{"meta":{"query_hash":"4f6799dd99f6","filters":{"topic":"Reliability and Agreement in Measurement"},"cohort_total":305,"direct_labels_cover":3,"predictions_cover":305,"exported":305,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/4f6799dd99f6","api":"https://metacan.xera.ac/api/v1/cohort?topic=Reliability+and+Agreement+in+Measurement"},"results":[{"id":"W1518867746","doi":"10.1108/09526861311297325","title":"A comparison of two diagnostic performance measures","year":2013,"lang":"en","type":"article","venue":"International Journal of Health Care Quality Assurance","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Measure (data warehouse); Point (geometry); Receiver operating characteristic; Signal-to-noise ratio (imaging); Taguchi methods; Noise (video); Diagnostic accuracy; Computer science; Index (typography); Statistics; Mathematics; Artificial intelligence; Data mining; Medicine; Radiology","score_opus":0.2317686654379214,"score_gpt":0.5212233067730002,"score_spread":0.2894546413350788,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1518867746","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96969664,0.004093476,0.010412452,0.012392351,0.00246676,0.0002708259,0.000018059261,0.0000073273686,0.0006421251],"genre_scores_gemma":[0.9967335,0.00019927342,0.0021417222,0.00066716503,0.00021201391,0.000007694159,0.0000018092896,0.000005445973,0.00003140538],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9895951,0.0011438756,0.0034018066,0.00022707708,0.0054008286,0.00023126749],"domain_scores_gemma":[0.9859018,0.003734626,0.003023934,0.00034016455,0.006857799,0.00014170777],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0104817655,0.00013791236,0.00068306923,0.00023834055,0.00010336359,0.00013758351,0.0014699387,0.000043337222,0.00027485372],"category_scores_gemma":[0.010728641,0.000098998105,0.00023591849,0.00025048447,0.00012367999,0.00062516314,0.000069496054,0.00029510597,0.000084850835],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007291079,0.00027131816,0.6615962,0.00007696701,0.000072275405,0.0000014622782,0.0046101674,0.007097602,0.00028767672,0.0004978767,0.008256896,0.31715867],"study_design_scores_gemma":[0.0018141455,0.0006891796,0.9466435,0.00075837434,0.000012377439,0.000020160438,0.00887457,0.0005214539,0.0032713886,0.005010105,0.032137744,0.00024703544],"about_ca_topic_score_codex":0.00047247618,"about_ca_topic_score_gemma":0.000107475564,"teacher_disagreement_score":0.31691164,"about_ca_system_score_codex":0.00020862467,"about_ca_system_score_gemma":0.00049596326,"threshold_uncertainty_score":0.9976044},"labels":[],"label_agreement":null},{"id":"W1754785110","doi":"10.1186/1756-0500-7-939","title":"A note on statistical and biological communication: a case study of the 2009 H1N1 pandemic","year":2014,"lang":"en","type":"letter","venue":"BMC Research Notes","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"University of Waterloo; York University","keywords":"Pandemic; Coronavirus disease 2019 (COVID-19); 2019-20 coronavirus outbreak; Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); H1n1 pandemic; Medicine; Data science; Computer science; Virology; Outbreak; Pathology; Disease; Infectious disease (medical specialty)","score_opus":0.7445059638980688,"score_gpt":0.555417270094464,"score_spread":0.18908869380360482,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1754785110","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.75583845,0.00039255028,0.0013313222,0.23804116,0.00020813211,0.0032849677,0.00014468862,0.00002806011,0.0007306934],"genre_scores_gemma":[0.97824854,0.00006611009,0.0006633949,0.020205462,0.00037782316,0.00012402979,0.000009203267,0.0000129092605,0.0002925384],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9810319,0.01214835,0.00092769595,0.0008093363,0.004614569,0.00046812763],"domain_scores_gemma":[0.9385769,0.05766724,0.00025936676,0.0025547948,0.00085176225,0.00008995796],"candidate_categories":["metaresearch","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.02596244,0.00022798963,0.00054944574,0.0002549829,0.00059916626,0.00029056493,0.0021153288,0.00040229657,0.00025222322],"category_scores_gemma":[0.041082315,0.00010868411,0.00010682799,0.00059248763,0.0012864247,0.00004858702,0.0010615216,0.0027959899,0.0000981625],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001432423,0.000692195,0.1729541,0.000077256,0.000038556354,0.0002827968,0.0012260887,0.000038922062,0.000041765128,0.00019441682,0.81273884,0.011571836],"study_design_scores_gemma":[0.004199524,0.009611281,0.14389198,0.0008082647,0.00011625933,0.00071427884,0.008712202,0.0025644053,0.000098038065,0.113743246,0.7143013,0.0012392696],"about_ca_topic_score_codex":0.00096563034,"about_ca_topic_score_gemma":0.0009691893,"teacher_disagreement_score":0.2224101,"about_ca_system_score_codex":0.00007471479,"about_ca_system_score_gemma":0.00018607524,"threshold_uncertainty_score":0.9995046},"labels":[],"label_agreement":null},{"id":"W1829168467","doi":"10.1191/0962280204sm365ra","title":"Sample size requirements for the design of reliability study: review and new results","year":2004,"lang":"en","type":"article","venue":"Statistical Methods in Medical Research","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":468,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Intraclass correlation; Sample size determination; Reliability (semiconductor); Variance (accounting); Statistics; Sample (material); Computer science; Mathematics; Reliability engineering; Psychometrics","score_opus":0.7388653050513273,"score_gpt":0.6767105014196586,"score_spread":0.0621548036316687,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1829168467","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00027721745,0.001785707,0.9649975,0.028784528,0.00014654855,0.00372231,0.000070454145,0.000007060383,0.00020869287],"genre_scores_gemma":[0.015296034,0.0023716572,0.98125136,0.00058862905,0.000065816195,0.00027002313,0.0000013574527,0.000010426617,0.00014468501],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9661128,0.017477801,0.003102248,0.0013736129,0.0110077765,0.0009257876],"domain_scores_gemma":[0.21180284,0.78199196,0.00036175098,0.002526177,0.002214304,0.0011029821],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.30590615,0.00015862181,0.0007157914,0.00011922534,0.00023461602,0.00008057058,0.001493691,0.000117325944,0.0012308691],"category_scores_gemma":[0.88523096,0.0000838829,0.000066337685,0.0013711544,0.0012729593,0.00008711299,0.00051170145,0.0007985488,0.000010978118],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00089007017,0.000996033,0.0010864048,0.0003816549,0.0000367946,0.000009792651,0.0006431302,0.000101561694,0.00006885325,0.012304317,0.016762849,0.96671855],"study_design_scores_gemma":[0.0025577268,0.0015834908,0.01666613,0.00047584344,0.000043431966,9.721886e-7,0.00070483703,0.0013848571,0.000087738874,0.9667127,0.009666867,0.00011539585],"about_ca_topic_score_codex":0.00096829666,"about_ca_topic_score_gemma":0.000085926986,"teacher_disagreement_score":0.96660316,"about_ca_system_score_codex":0.00015765542,"about_ca_system_score_gemma":0.0010248302,"threshold_uncertainty_score":0.9996821},"labels":[],"label_agreement":null},{"id":"W1881569770","doi":"10.1002/9781118445112.stat05243","title":"Kappa and Its Dependence on Marginal Rates","year":2014,"lang":"en","type":"other","venue":"Wiley StatsRef: Statistics Reference Online","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":40,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Kappa; Cohen's kappa; Statistic; Statistics; Measure (data warehouse); Mathematics; Econometrics; Computer science; Data mining; Geometry","score_opus":0.18230828496112722,"score_gpt":0.4110886459547939,"score_spread":0.22878036099366666,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1881569770","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0045707948,0.012869554,0.24733669,0.00371092,0.0064005535,0.0059377574,0.112984404,0.0009963459,0.60519296],"genre_scores_gemma":[0.01658486,0.010843176,0.07746396,0.0014890564,0.0010577177,0.000093292605,0.002206324,0.0005930181,0.8896686],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9924707,0.0005656208,0.0011643139,0.0015468487,0.0035961648,0.0006563935],"domain_scores_gemma":[0.9951077,0.001774304,0.00093616254,0.0011922553,0.0005924616,0.0003970887],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0022853548,0.0007128985,0.0010006117,0.0005485315,0.00018526342,0.00037993182,0.0013105145,0.00039352739,0.009830364],"category_scores_gemma":[0.003746122,0.0005163636,0.00007073601,0.00039365064,0.00029002168,0.00009445089,0.0002854445,0.00078356336,0.0022756555],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001078388,0.00039257374,0.00067010114,0.00016759361,0.00007513898,0.00004489989,0.000056817145,0.00004343169,0.000074117626,0.03829796,0.88766736,0.07240217],"study_design_scores_gemma":[0.0007756625,0.0007877319,0.0015883164,0.00093664404,0.000076089615,0.000007344966,0.00012979361,0.0027402034,0.000038554736,0.047449023,0.9445784,0.00089225365],"about_ca_topic_score_codex":0.0001180352,"about_ca_topic_score_gemma":0.001236471,"teacher_disagreement_score":0.28447562,"about_ca_system_score_codex":0.00008925405,"about_ca_system_score_gemma":0.0002403428,"threshold_uncertainty_score":0.9997288},"labels":[],"label_agreement":null},{"id":"W1887790309","doi":"10.1186/1742-5573-3-11","title":"New patient-oriented summary measure of net total gain in certainty for dichotomous diagnostic tests","year":2006,"lang":"en","type":"article","venue":"Epidemiologic Perspectives & Innovations","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":96,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Reciprocal; Measure (data warehouse); Certainty; Youden's J statistic; Index (typography); Medicine; Statistics; Population; Disease; Computer science; Test (biology); Predictive value; Mathematics; Data mining; Internal medicine","score_opus":0.08524577208288033,"score_gpt":0.3642964651287134,"score_spread":0.2790506930458331,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1887790309","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9134075,0.0014704703,0.06910617,0.0077625685,0.00047555182,0.0020073326,0.00023519162,0.00007834889,0.0054568322],"genre_scores_gemma":[0.9896062,0.000015622836,0.009395287,0.00024753468,0.00014252824,0.00018735637,0.000072793795,0.000008522932,0.00032415465],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9950262,0.000767392,0.0019648338,0.0008364648,0.0009080434,0.0004970699],"domain_scores_gemma":[0.9818092,0.015054553,0.00081189367,0.00070497656,0.0015338014,0.00008556204],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.007246373,0.00026735576,0.00067756657,0.0005269886,0.0001905718,0.000038293485,0.00047379592,0.00016543402,0.00020271986],"category_scores_gemma":[0.09875732,0.00019726188,0.00022168388,0.0026280063,0.00023862683,0.00025540215,0.0001042011,0.00028626173,0.000018488985],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016701977,0.0011875955,0.7242026,0.000015227075,0.00005017987,0.000005356021,0.0018805544,0.021679757,0.0031169003,0.120765485,0.11300037,0.013928998],"study_design_scores_gemma":[0.0010813471,0.00052282686,0.71586686,0.00008200976,0.000019291101,0.0000024970477,0.006719258,0.0011019218,0.0002890593,0.27092978,0.003057321,0.00032781647],"about_ca_topic_score_codex":0.0010387016,"about_ca_topic_score_gemma":0.0005350477,"teacher_disagreement_score":0.1501643,"about_ca_system_score_codex":0.00030443986,"about_ca_system_score_gemma":0.0002280609,"threshold_uncertainty_score":0.9088342},"labels":[],"label_agreement":null},{"id":"W1949714730","doi":"10.5435/00124635-200207000-00007","title":"Classification Systems in Orthopaedics","year":2002,"lang":"en","type":"review","venue":"Journal of the American Academy of Orthopaedic Surgeons","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":164,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Medicine; Reliability (semiconductor); Documentation; Kappa; Measure (data warehouse); Orthopedic surgery; Medical physics; Cohen's kappa; Reliability engineering; Data mining; Machine learning; Computer science; Surgery","score_opus":0.24293474394203074,"score_gpt":0.4163155049620289,"score_spread":0.17338076101999816,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1949714730","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0019571104,0.9936304,0.000060274575,0.0016337764,0.0015340588,0.0007050229,0.000038510658,0.000007682295,0.00043316555],"genre_scores_gemma":[0.006966587,0.9919982,0.00008118189,0.00013914426,0.00031015015,0.000018731585,8.2997997e-7,0.000032611686,0.00045254306],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9829429,0.0039114286,0.0066313744,0.0005237982,0.0055101826,0.00048036643],"domain_scores_gemma":[0.97664684,0.0049201446,0.016935311,0.0008044552,0.00043961825,0.0002536088],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.022884315,0.00048834266,0.0041110544,0.0015136091,0.00011701221,0.00010142124,0.0034442449,0.0002928653,0.00012321334],"category_scores_gemma":[0.008399463,0.00025620408,0.0029833189,0.0047851177,0.00090392574,0.0002530377,0.0002721943,0.0021958738,0.000050747007],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000019434205,0.00019952952,0.015059928,0.00071695354,0.00015936838,0.000012289635,0.000053528373,0.00011773867,0.0000014870468,0.00014478578,0.015129188,0.96838576],"study_design_scores_gemma":[0.00020105451,0.00008689235,0.010295672,0.0071768546,0.0004253225,0.00019934398,0.00029739717,0.000080767495,2.6759287e-7,0.00023335288,0.9807619,0.00024116845],"about_ca_topic_score_codex":0.000047900252,"about_ca_topic_score_gemma":0.0000059706085,"teacher_disagreement_score":0.9681446,"about_ca_system_score_codex":0.00035511304,"about_ca_system_score_gemma":0.0005715385,"threshold_uncertainty_score":0.99998903},"labels":[],"label_agreement":null},{"id":"W1955533988","doi":"10.6000/1929-6029.2015.04.02.2","title":"On the Relationship between the Reliability and Accuracy of Bio-Behavioral Diagnoses: Simple Math to the Rescue","year":2015,"lang":"en","type":"article","venue":"International Journal of Statistics in Medical Research","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"","keywords":"Medical diagnosis; Kappa; Mathematics; Statistics; Cohen's kappa; Equivalence (formal languages); Statistic; Youden's J statistic; Combinatorics; Psychology; Medicine; Discrete mathematics; Pathology; Receiver operating characteristic; Geometry","score_opus":0.5558141781385264,"score_gpt":0.5812588006269411,"score_spread":0.025444622488414725,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1955533988","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.90117604,0.00005546113,0.0075064334,0.09035337,0.0003416706,0.00029138476,0.00009355734,0.0000011611255,0.00018091581],"genre_scores_gemma":[0.9986672,0.000059010497,0.00062033755,0.0003689583,0.00022340356,0.000011700302,0.0000020315445,0.0000040492882,0.00004333381],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98364115,0.0024100356,0.0012146579,0.00019429052,0.012328271,0.00021162043],"domain_scores_gemma":[0.8517298,0.14242655,0.0003795986,0.00061819033,0.0045119934,0.00033385877],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07744791,0.000079610094,0.00019330364,0.00024900952,0.00015613565,0.0002183794,0.0026752378,0.00006541773,0.00021447934],"category_scores_gemma":[0.4724514,0.00003116069,0.000050942934,0.000563599,0.0007482238,0.00011376406,0.00045969387,0.0012256592,0.000038149406],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00037228933,0.00037337508,0.67588747,0.0000064302512,0.000028748025,0.00005750653,0.003074527,0.0007601113,0.0000051472052,0.13069893,0.11945745,0.06927798],"study_design_scores_gemma":[0.00039415478,0.0004129487,0.47132337,0.00012442566,0.0000056145077,0.000006858912,0.002505464,0.00084253494,0.00002181154,0.51218337,0.012138731,0.000040728395],"about_ca_topic_score_codex":0.00029966334,"about_ca_topic_score_gemma":0.00027850523,"teacher_disagreement_score":0.39500347,"about_ca_system_score_codex":0.0001918248,"about_ca_system_score_gemma":0.00060644123,"threshold_uncertainty_score":0.94996154},"labels":[],"label_agreement":null},{"id":"W1968667761","doi":"10.1016/j.ijnurstu.2011.01.016","title":"Guidelines for Reporting Reliability and Agreement Studies (GRRAS) were proposed","year":2011,"lang":"en","type":"article","venue":"International Journal of Nursing Studies","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":985,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Western University","funders":"","keywords":"Agreement; Reliability (semiconductor); Psychology; Physics","score_opus":0.7846223701604345,"score_gpt":0.5880428148793927,"score_spread":0.19657955528104176,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1968667761","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9141483,0.016700048,0.0064282864,0.05032753,0.010404203,0.0006028932,0.000010765745,0.000021464424,0.0013565371],"genre_scores_gemma":[0.9526653,0.00125354,0.044425573,0.0004265851,0.00071396056,0.000009631301,3.941298e-7,0.0000085397305,0.0004964456],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99353766,0.00015015519,0.00361348,0.00034572376,0.0021610078,0.00019199392],"domain_scores_gemma":[0.9776951,0.0010620418,0.0038603654,0.00025227154,0.01705962,0.00007057136],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.015187533,0.00017602726,0.0005931967,0.0002300148,0.00021571845,0.000097568154,0.0005820048,0.00003846576,0.000024469748],"category_scores_gemma":[0.071750194,0.000106273976,0.0002996476,0.00011686439,0.00039420053,0.00039012282,0.00012810747,0.00013019356,0.0000031080235],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015968047,0.0014250781,0.10198204,0.00010732792,0.004539798,0.00009426978,0.052536476,0.0004452967,0.0020858492,0.0031463313,0.2053537,0.62668705],"study_design_scores_gemma":[0.002678173,0.0017395566,0.058354307,0.0034731568,0.00044097594,0.00027409868,0.1377573,0.0003956322,0.006433009,0.7658302,0.022103686,0.00051985425],"about_ca_topic_score_codex":0.000004845205,"about_ca_topic_score_gemma":0.000008293466,"teacher_disagreement_score":0.7626839,"about_ca_system_score_codex":0.00030293816,"about_ca_system_score_gemma":0.00007838821,"threshold_uncertainty_score":0.93606883},"labels":[],"label_agreement":null},{"id":"W1970325005","doi":"10.1037/a0029315","title":"When can categorical variables be treated as continuous? A comparison of robust continuous and categorical SEM estimation methods under suboptimal conditions.","year":2012,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2534,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Categorical variable; Statistics; Mathematics; Normality; Continuous variable; Sample size determination; Ordinal data; Sample (material); Latent variable; Confirmatory factor analysis; Econometrics; Structural equation modeling","score_opus":0.35917610160414,"score_gpt":0.5393551820295689,"score_spread":0.1801790804254289,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1970325005","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2184741,0.00065069646,0.7715866,0.0029581173,0.0006295535,0.00063354237,0.000019847663,0.00009562784,0.004951901],"genre_scores_gemma":[0.56340086,0.000014651143,0.4358103,0.00038298056,0.00006637195,0.000059472026,0.00001744137,0.000012116294,0.00023578166],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9880808,0.0070522563,0.0018335215,0.0009840264,0.0012995218,0.00074986217],"domain_scores_gemma":[0.9887494,0.008602803,0.0007382056,0.0008331003,0.0005398568,0.0005366234],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.019853534,0.00043089868,0.0014159856,0.00026614405,0.0002947318,0.00021130822,0.0006962457,0.00049215835,0.0017594387],"category_scores_gemma":[0.0092919925,0.0002864,0.00024615735,0.00089918356,0.0006933506,0.0003313724,0.00020047076,0.0005656711,0.00003739318],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008796751,0.008410262,0.2674582,0.000064654996,0.00044051534,0.000015965497,0.0049694953,0.004061446,0.03602914,0.15476674,0.027018527,0.49588537],"study_design_scores_gemma":[0.0034362245,0.0029654999,0.22082315,0.00004391505,0.00059820956,0.00021336961,0.0058752955,0.028391765,0.014317865,0.70704544,0.014880322,0.0014089294],"about_ca_topic_score_codex":0.0002334369,"about_ca_topic_score_gemma":0.000006330643,"teacher_disagreement_score":0.5522787,"about_ca_system_score_codex":0.00012126773,"about_ca_system_score_gemma":0.0000365577,"threshold_uncertainty_score":0.9999588},"labels":[],"label_agreement":null},{"id":"W1970971885","doi":"10.1016/j.jclinepi.2011.10.019","title":"A confidence interval approach to sample size estimation for interobserver agreement studies with multiple raters and outcomes","year":2012,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":131,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; York University","funders":"","keywords":"Confidence interval; Triage; Sample size determination; Reliability (semiconductor); Medicine; Estimation; Medical physics; Sample (material); Interval estimation; Statistics; Scale (ratio); Mathematics; Psychiatry; Power (physics)","score_opus":0.7278829789470725,"score_gpt":0.5892656106725227,"score_spread":0.13861736827454985,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1970971885","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5510898,0.00021653275,0.4377364,0.009640635,0.0009378351,0.0003496362,0.0000052715254,0.0000034124268,0.000020464518],"genre_scores_gemma":[0.7412383,0.000052657768,0.25461778,0.003840247,0.00017331465,0.000019500561,3.9343047e-7,0.000004288615,0.0000535332],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9912939,0.002782632,0.004568639,0.0003588049,0.00061785383,0.0003781493],"domain_scores_gemma":[0.6780325,0.31531006,0.0039408077,0.0005730192,0.0015092379,0.00063433225],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.090695836,0.0001803115,0.002053035,0.00009117781,0.0000845443,0.000032034313,0.00047776164,0.00010881442,0.000027110376],"category_scores_gemma":[0.53522193,0.000089277244,0.00038261423,0.00012186735,0.00032151854,0.00036790594,0.00016256984,0.00029087608,0.000010835931],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011038019,0.0003167372,0.9745274,0.00003082355,0.00025387557,3.5719765e-7,0.00068008684,0.001985873,0.000007718636,0.001437612,0.0053834654,0.014272238],"study_design_scores_gemma":[0.0022373649,0.0028781104,0.9346225,0.00012928038,0.00009984955,0.000015531436,0.002355095,0.004781458,0.000017428785,0.04831425,0.004381427,0.00016765641],"about_ca_topic_score_codex":0.000017589427,"about_ca_topic_score_gemma":0.0000239838,"teacher_disagreement_score":0.44452614,"about_ca_system_score_codex":0.000051516865,"about_ca_system_score_gemma":0.000049445098,"threshold_uncertainty_score":0.93632},"labels":[],"label_agreement":null},{"id":"W1976786126","doi":"10.1002/sim.911","title":"A general goodness‐of‐fit approach for inference procedures concerning the kappa statistic","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Goodness of fit; Statistics; Polytomous Rasch model; Statistic; Cohen's kappa; Inference; Mathematics; Statistical inference; Test statistic; Multinomial distribution; Dirichlet distribution; Outcome (game theory); Kappa; Range (aeronautics); Confidence interval; Econometrics; Statistical hypothesis testing; Computer science; Item response theory; Psychometrics; Artificial intelligence","score_opus":0.2882924109126018,"score_gpt":0.4485972092231941,"score_spread":0.1603047983105923,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1976786126","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.020719312,0.00037117253,0.97362036,0.0008422169,0.00034496645,0.00086314685,0.00016797546,0.000011117264,0.0030597392],"genre_scores_gemma":[0.8879613,0.00015096123,0.11005801,0.00037840143,0.00022514242,0.00014898437,0.00004103414,0.000011419365,0.0010246992],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99614334,0.00024846086,0.0011142162,0.0004300747,0.0017397022,0.00032418757],"domain_scores_gemma":[0.9931344,0.005273756,0.0004028415,0.00048649468,0.00062172604,0.000080782775],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006128714,0.00017309579,0.00045718628,0.00014242234,0.0001483617,0.00004448042,0.00077723485,0.00005204178,0.00035103966],"category_scores_gemma":[0.027906502,0.00009391504,0.00003324876,0.0006110486,0.00062644616,0.00006799417,0.00007292936,0.0002024107,0.000006735288],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006151115,0.0006510035,0.21932468,0.0006463564,0.00011211949,0.000032289332,0.013955891,0.028376866,0.0015828973,0.4283963,0.12300453,0.18330196],"study_design_scores_gemma":[0.00373466,0.0011638519,0.070077,0.00035303217,0.000114899805,0.000014623267,0.0106649725,0.21134102,0.00011384635,0.67557806,0.026378945,0.000465109],"about_ca_topic_score_codex":0.00019023631,"about_ca_topic_score_gemma":0.0002347758,"teacher_disagreement_score":0.86724204,"about_ca_system_score_codex":0.000053832937,"about_ca_system_score_gemma":0.00018569734,"threshold_uncertainty_score":0.98028183},"labels":[],"label_agreement":null},{"id":"W1978765295","doi":"10.2202/1557-4679.1275","title":"Sample Size Requirements for Interval Estimation of the Kappa Statistic for Interobserver Agreement Studies with a Binary Outcome and Multiple Raters","year":2010,"lang":"en","type":"article","venue":"The International Journal of Biostatistics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":127,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Statistics; Cohen's kappa; Confidence interval; Kappa; Intraclass correlation; Mathematics; Interval estimation; Statistic; Sample size determination; Sample (material); Binary number; Limit (mathematics); Interval (graph theory); Psychometrics; Combinatorics","score_opus":0.2216258336390316,"score_gpt":0.43800712383045454,"score_spread":0.21638129019142294,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1978765295","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.68724144,0.000036467747,0.30452096,0.005011971,0.0020031142,0.00057676557,0.0006028786,0.0000021120886,0.0000042817906],"genre_scores_gemma":[0.8858047,0.000012907027,0.11370487,0.00029963904,0.0000769631,0.00001203056,0.000003935047,0.0000068586305,0.00007806447],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9970289,0.0001053058,0.0011673896,0.00016616363,0.0014008649,0.00013136466],"domain_scores_gemma":[0.9857738,0.0104351,0.0012699215,0.0002525876,0.0022239909,0.000044623786],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0041139657,0.00013358652,0.00028243774,0.00007955485,0.00013568788,0.00013081508,0.0009536272,0.000026629055,0.00003126899],"category_scores_gemma":[0.025751788,0.000061026185,0.0001129646,0.000091023794,0.0003308829,0.00017073109,0.00017405467,0.00014133,9.1031797e-7],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.014822012,0.0020801928,0.54771596,0.00074401,0.0062130727,0.000024004725,0.017244702,0.031118978,0.071794756,0.044940155,0.06721252,0.19608964],"study_design_scores_gemma":[0.015584549,0.0061426708,0.27421665,0.0012391439,0.0010017998,0.0001133647,0.014398312,0.17465077,0.019064365,0.48324764,0.009562142,0.0007785992],"about_ca_topic_score_codex":0.000020039623,"about_ca_topic_score_gemma":0.00015474774,"teacher_disagreement_score":0.4383075,"about_ca_system_score_codex":0.00007405372,"about_ca_system_score_gemma":0.0000755367,"threshold_uncertainty_score":0.9824547},"labels":[],"label_agreement":null},{"id":"W1979334793","doi":"10.1111/1467-9884.00266","title":"Statistical Inferences For Interobserver Agreement Studies With Nominal Outcome Data","year":2001,"lang":"en","type":"article","venue":"Journal of the Royal Statistical Society Series D (The Statistician)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Categorical variable; Outcome (game theory); Statistical inference; Statistics; Econometrics; Statistical hypothesis testing; Nominal level; Inference; Multiple comparisons problem; Focus (optics); Computer science; Mathematics; Artificial intelligence; Confidence interval","score_opus":0.35846836881799066,"score_gpt":0.4458766577488586,"score_spread":0.08740828893086794,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1979334793","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02137651,0.0003438235,0.9576968,0.015676875,0.0016602852,0.0008460031,0.0021535219,0.000012735836,0.00023342125],"genre_scores_gemma":[0.72710526,0.00017155265,0.26522142,0.0026756884,0.0009216703,0.000054286258,0.000049617258,0.00004569533,0.0037548314],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9923142,0.00072817295,0.0023143329,0.00059824146,0.003406787,0.0006382595],"domain_scores_gemma":[0.9846421,0.010735158,0.0013739918,0.0012420173,0.001762576,0.00024416298],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.008997114,0.00039980895,0.0009420199,0.000037445767,0.00094242184,0.0005482832,0.0032915599,0.00008562523,0.0010136414],"category_scores_gemma":[0.016230488,0.00016879005,0.00028355178,0.00040016416,0.0017356159,0.00045408081,0.000842089,0.0006180723,0.00004309736],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0033087533,0.00091165304,0.110433936,0.00028260212,0.002517648,0.00009939782,0.004506975,0.00471128,0.000025008527,0.10086609,0.68939084,0.082945846],"study_design_scores_gemma":[0.0039150105,0.005149884,0.32403594,0.00040430855,0.0018251083,0.00015931953,0.036181692,0.025212038,0.000030618554,0.33550808,0.26651016,0.0010678535],"about_ca_topic_score_codex":0.00007731915,"about_ca_topic_score_gemma":0.0006640728,"teacher_disagreement_score":0.7057287,"about_ca_system_score_codex":0.00026186972,"about_ca_system_score_gemma":0.00029193782,"threshold_uncertainty_score":0.99989957},"labels":[],"label_agreement":null},{"id":"W1983075605","doi":"10.1038/nmeth.2659","title":"Error bars","year":2013,"lang":"en","type":"article","venue":"Nature Methods","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":289,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canada's Michael Smith Genome Sciences Centre","funders":"","keywords":"Computer science; Chemistry","score_opus":0.22189996935963935,"score_gpt":0.5448362889887357,"score_spread":0.32293631962909636,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1983075605","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2005043,0.004842004,0.4237961,0.03945188,0.01175299,0.0022004317,0.000013848086,0.00031140004,0.31712705],"genre_scores_gemma":[0.2024047,0.000009355923,0.7835223,0.00391298,0.00024099565,0.000050517137,0.0000016103485,0.000011130141,0.009846417],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9956036,0.0014766251,0.00049811025,0.00049223664,0.0016568182,0.0002726161],"domain_scores_gemma":[0.99598646,0.002325654,0.0001327875,0.00091410184,0.0004989491,0.00014205235],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0149639025,0.00013712376,0.00025473745,0.00013933827,0.00012624924,0.00025537066,0.00094163127,0.0003343286,0.008351367],"category_scores_gemma":[0.01221792,0.00008143799,0.00014100554,0.000651349,0.00007608941,0.00029130877,0.00012453852,0.0007337779,0.001975726],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000075817225,0.000082432605,0.00383696,0.000004519363,0.000016995533,0.0000012833987,0.00020074313,0.000058600475,0.011614073,0.0038601062,0.22061478,0.7597019],"study_design_scores_gemma":[0.00018044897,0.000053232077,0.0472484,0.0000073724773,0.000008434353,0.000002326228,0.00032667792,0.0009819947,0.0073907482,0.22615813,0.7174686,0.00017364147],"about_ca_topic_score_codex":0.000021127627,"about_ca_topic_score_gemma":0.0000032006071,"teacher_disagreement_score":0.7595283,"about_ca_system_score_codex":0.000037562895,"about_ca_system_score_gemma":0.000034019922,"threshold_uncertainty_score":0.99880135},"labels":[],"label_agreement":null},{"id":"W1987916785","doi":"10.1016/j.jclinepi.2008.05.003","title":"Commentary on “Alternative graphs for diagnostic tests: the agreement chart and the receiver operating characteristic curve”","year":2008,"lang":"en","type":"letter","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Joseph’s Healthcare Hamilton; McMaster University","funders":"","keywords":"Receiver operating characteristic; Computer science; Chart; Perception; Feature (linguistics); Artificial intelligence; Information retrieval; Statistics; Psychology; Mathematics; Machine learning; Linguistics","score_opus":0.49507110752261396,"score_gpt":0.520213486178791,"score_spread":0.025142378656177078,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1987916785","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01384038,0.0011473614,0.00176267,0.9758306,0.0058453297,0.0013130372,0.000080899954,0.000004370229,0.00017534035],"genre_scores_gemma":[0.011438186,0.009475777,0.00067455485,0.9665728,0.011382324,0.00009098307,0.000019341505,0.000020535608,0.0003254642],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9590475,0.02757781,0.010045779,0.0008494858,0.0018431729,0.00063625246],"domain_scores_gemma":[0.44451317,0.54509443,0.008301892,0.0009150272,0.0010155838,0.00015988962],"candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.14348626,0.00046157461,0.003771098,0.00017621757,0.00058471353,0.000104827704,0.002232592,0.0006052018,0.00018367839],"category_scores_gemma":[0.47795868,0.00018118533,0.0016088955,0.00017704625,0.0021937152,0.00014480045,0.00027082665,0.0048009786,0.00004684435],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00040164095,0.0001288394,0.030110864,0.000019738321,0.00035422665,0.000055497145,0.00019738528,0.00010889401,3.0661562e-7,0.0003494783,0.95736283,0.010910309],"study_design_scores_gemma":[0.002905281,0.0019108343,0.05329249,0.0003751136,0.0002093101,0.00007202749,0.00010625293,0.0007369848,7.509389e-7,0.08815455,0.8520125,0.00022389651],"about_ca_topic_score_codex":0.000054517994,"about_ca_topic_score_gemma":0.000012773465,"teacher_disagreement_score":0.5175166,"about_ca_system_score_codex":0.00008541581,"about_ca_system_score_gemma":0.00014307884,"threshold_uncertainty_score":0.997495},"labels":[],"label_agreement":null},{"id":"W1989853314","doi":"10.1186/1471-2288-9-5","title":"Measuring agreement of administrative data with chart data using prevalence unadjusted and adjusted kappa","year":2009,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":177,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Calgary General Hospital; Alberta Bone and Joint Health Institute; University of Calgary","funders":"Canadian Institutes of Health Research; Fondation pour la Recherche Médicale","keywords":"Kappa; Cohen's kappa; Statistics; Medicine; Sampling (signal processing); Sample size determination; Sample (material); Mathematics; Computer science","score_opus":0.9684346225908497,"score_gpt":0.6570201568897845,"score_spread":0.3114144657010651,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1989853314","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5973017,0.002872126,0.3907014,0.0053936136,0.00027112686,0.001746694,0.0002083691,0.000032118918,0.0014728119],"genre_scores_gemma":[0.73024946,0.00073992024,0.2682155,0.00023268219,0.00021544792,0.0000145444865,0.000053886717,0.000012697442,0.0002658691],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9596533,0.023688285,0.0015916175,0.002099676,0.012024033,0.00094307074],"domain_scores_gemma":[0.9358465,0.05494029,0.00050868525,0.0059418636,0.0018305515,0.00093208445],"candidate_categories":["metaresearch","open_science","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.18285584,0.00022656185,0.00080535054,0.000405469,0.00026327002,0.000096260424,0.0057533225,0.00022798835,0.0011884006],"category_scores_gemma":[0.24845958,0.00014285053,0.000038199338,0.001287191,0.0016291391,0.00050303305,0.0025977402,0.00082962844,0.000019816522],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0055043017,0.0022120813,0.13044031,0.0014943552,0.000405046,0.00030232256,0.0029590833,0.0002245794,0.02797928,0.0070339725,0.01093176,0.8105129],"study_design_scores_gemma":[0.0092830425,0.009673474,0.6395493,0.0043789656,0.00041933043,0.000432117,0.022421196,0.24415717,0.011662184,0.040653888,0.01580456,0.0015647456],"about_ca_topic_score_codex":0.0002220004,"about_ca_topic_score_gemma":0.0011910191,"teacher_disagreement_score":0.80894816,"about_ca_system_score_codex":0.00007229082,"about_ca_system_score_gemma":0.0023064837,"threshold_uncertainty_score":0.9997246},"labels":[],"label_agreement":null},{"id":"W1990424977","doi":"10.1007/s00180-013-0424-7","title":"Evaluation of confidence intervals for the kappa statistic when the assumption of marginal homogeneity is violated","year":2013,"lang":"en","type":"article","venue":"Computational Statistics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; McMaster University; Ontario Clinical Oncology Group","funders":"","keywords":"Homogeneity (statistics); Kappa; Statistic; Confidence interval; Statistics; Mathematics; Cohen's kappa; Econometrics; Test statistic; Statistical hypothesis testing","score_opus":0.3043425736713799,"score_gpt":0.4321275538967644,"score_spread":0.12778498022538448,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1990424977","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.049903527,0.00019649445,0.9459013,0.0016014912,0.00024152022,0.0012340422,0.0008566742,0.0000043622344,0.000060632847],"genre_scores_gemma":[0.9513144,0.000009056163,0.048261207,0.00016771902,0.00003020129,0.00010157904,0.000049315662,0.000005000416,0.00006152449],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9945778,0.0006517194,0.0009604007,0.00024338129,0.0034278594,0.00013880024],"domain_scores_gemma":[0.98196936,0.009931479,0.0006902279,0.00033558116,0.0070378575,0.00003548508],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01108366,0.000108889,0.00020799408,0.000058366073,0.00018339542,0.000110070796,0.00057534914,0.000034877707,0.0019114383],"category_scores_gemma":[0.006603666,0.000059915586,0.00007301322,0.0001985109,0.00029662502,0.00012374198,0.00006383814,0.000079746525,0.000058847876],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009285208,0.0002808275,0.007044494,0.00008839821,0.0002946211,1.5203639e-7,0.0032193172,0.3627656,0.0006701977,0.07843429,0.113122225,0.43398702],"study_design_scores_gemma":[0.00021976176,0.000063970176,0.13042477,0.000014925108,0.0000813108,4.5111142e-7,0.0001949497,0.4158411,0.00011036382,0.4528456,0.000163859,0.000038933635],"about_ca_topic_score_codex":0.00017750189,"about_ca_topic_score_gemma":0.00003715284,"teacher_disagreement_score":0.9014109,"about_ca_system_score_codex":0.00006735007,"about_ca_system_score_gemma":0.0002297441,"threshold_uncertainty_score":0.99900097},"labels":[],"label_agreement":null},{"id":"W1992959707","doi":"10.1093/ps/79.8.1194","title":"Bacteriological and Histological Profile of Turkeys Condemned for Cyanosis","year":2000,"lang":"en","type":"article","venue":"Poultry Science","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Health Canada; University of Guelph","funders":"Health Canada","keywords":"Salmonella; Veterinary medicine; Rappaport; Medicine; Biology; Bacteria","score_opus":0.17528353550879233,"score_gpt":0.38849228119078877,"score_spread":0.21320874568199644,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1992959707","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9907906,0.00008144967,0.0005333266,0.0012683566,0.000140959,0.0003557724,0.000020516569,0.000016937154,0.006792056],"genre_scores_gemma":[0.99300396,0.000020173884,0.0056614955,0.00028588698,0.00002198334,0.000028388657,0.0000010580343,0.0000016038688,0.0009754622],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99733436,0.000112491936,0.00050989946,0.0006300245,0.0011160787,0.0002971407],"domain_scores_gemma":[0.99809676,0.0009227619,0.00013029926,0.00039313626,0.00032149345,0.00013552664],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0055597695,0.0001023624,0.00026375233,0.000081498685,0.00026380835,0.00010440231,0.00086649525,0.000067347566,0.0026431258],"category_scores_gemma":[0.0045574335,0.000058890968,0.000072098606,0.00056324527,0.0014104189,0.00028067254,0.00008280621,0.000076133445,0.00007028178],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028175281,0.00039376772,0.5174047,0.000024071374,0.000008386544,0.0000018799703,0.0004710036,0.0000756601,0.20162062,0.0042154314,0.004262615,0.27124014],"study_design_scores_gemma":[0.00064233865,0.0007575021,0.9142465,0.000026230979,0.000012385501,0.000009552752,0.00060059526,0.0017488342,0.011247154,0.030266419,0.04018289,0.00025958652],"about_ca_topic_score_codex":0.000008355699,"about_ca_topic_score_gemma":0.0000010153068,"teacher_disagreement_score":0.39684185,"about_ca_system_score_codex":0.000030737676,"about_ca_system_score_gemma":0.000089987814,"threshold_uncertainty_score":0.9982686},"labels":[],"label_agreement":null},{"id":"W1993809841","doi":"10.1016/j.jclinepi.2004.11.026","title":"The problem of imperfect reference standards","year":2005,"lang":"en","type":"letter","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Gold standard (test); Imperfect; Class (philosophy); Latent class model; Standard error; Statistics; Scopus; Computer science; Econometrics; Mathematics; MEDLINE; Artificial intelligence; Philosophy; Linguistics; Political science","score_opus":0.6177846126145073,"score_gpt":0.5901486659943236,"score_spread":0.02763594662018376,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1993809841","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0032895997,0.002716842,0.0022385542,0.98351675,0.0026018785,0.00028451212,0.00006105661,0.0000054145216,0.0052853613],"genre_scores_gemma":[0.011942791,0.017057812,0.019414566,0.9202157,0.026149422,0.000013282913,0.000007650794,0.00004651521,0.0051522166],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9091976,0.047289524,0.033962145,0.0010777568,0.0073482227,0.0011247075],"domain_scores_gemma":[0.63948995,0.32944342,0.023696532,0.0015482698,0.0055719437,0.0002498728],"candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.40660533,0.00033752384,0.0050662933,0.00019653616,0.00016258624,0.00005470199,0.0036038393,0.0016679718,0.00065899827],"category_scores_gemma":[0.36856198,0.00014474851,0.002248474,0.00028533518,0.0013410837,0.00014644895,0.00026912117,0.007253008,0.000067184556],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00022274084,0.000056168064,0.012585777,0.000022147844,0.00015377361,0.000019166235,0.000012876832,0.00012189999,0.0000019180218,0.0003970875,0.86223435,0.1241721],"study_design_scores_gemma":[0.00043075587,0.0013508808,0.0072246976,0.00016355861,0.0000755478,0.000033244673,0.000021892807,0.00003554542,0.0000029184127,0.11139904,0.87913924,0.00012265598],"about_ca_topic_score_codex":0.000014625609,"about_ca_topic_score_gemma":0.00002544061,"teacher_disagreement_score":0.2821539,"about_ca_system_score_codex":0.00013162526,"about_ca_system_score_gemma":0.0015612944,"threshold_uncertainty_score":0.99962807},"labels":[],"label_agreement":null},{"id":"W1994298865","doi":"10.1371/journal.pone.0052221","title":"Assessing Diagnostic Tests: How to Correct for the Combined Effects of Interpretation and Reference Standard","year":2012,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"SUNY Downstate Medical Center; National Institutes of Health; National Institute of Neurological Disorders and Stroke; York University; State University of New York","keywords":"Sensitivity (control systems); Reliability (semiconductor); Diagnostic accuracy; Gold standard (test); Reference values; Standard deviation; Computer science; Interpreter; Diagnostic test; Statistics; Standard error; Calibration; Interpretation (philosophy); Medical physics; Medicine; Mathematics; Radiology; Internal medicine; Pediatrics; Physics","score_opus":0.2178926583735101,"score_gpt":0.3706645710522915,"score_spread":0.1527719126787814,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1994298865","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9627332,0.0003549946,0.03380107,0.0018303428,0.00018410698,0.00094313355,0.0000073343695,0.000010557369,0.00013527139],"genre_scores_gemma":[0.99645406,0.000022918897,0.003188555,0.00014110279,0.000036637193,0.0000981019,9.0946384e-7,0.0000039610704,0.000053762862],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99835515,0.00016073168,0.00023151588,0.00016542771,0.00093612476,0.00015104814],"domain_scores_gemma":[0.97914314,0.020000199,0.00013651057,0.0002794476,0.0003719126,0.00006879656],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0029414096,0.0000738135,0.0002054094,0.00005100498,0.00010505408,0.00020500144,0.00022193186,0.000026570662,0.0000168374],"category_scores_gemma":[0.04526049,0.000042304164,0.000026733269,0.00018200834,0.00005390645,0.00033614907,0.00006330241,0.000060795403,0.000007218643],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006421738,0.002696162,0.54249877,0.0008151719,0.00029991678,6.5139926e-7,0.007107463,0.00006844896,0.15160766,0.0013763617,0.006582953,0.2863043],"study_design_scores_gemma":[0.0015425071,0.0025698435,0.7398353,0.0020413788,0.00042436185,4.975342e-7,0.002641626,0.0034567288,0.23679867,0.009453291,0.0008968589,0.0003389217],"about_ca_topic_score_codex":0.000004195913,"about_ca_topic_score_gemma":0.000009464347,"teacher_disagreement_score":0.28596535,"about_ca_system_score_codex":0.000031876556,"about_ca_system_score_gemma":0.000019421792,"threshold_uncertainty_score":0.96278167},"labels":[],"label_agreement":null},{"id":"W1994361491","doi":"10.1016/j.rhum.2003.07.008","title":"Évaluation des destructions dans les articulations analysées individuellement au cours de la polyarthrite rhumatoïde : une nouvelle méthode fondée sur celle de Larsen","year":2003,"lang":"fr","type":"article","venue":"Revue du Rhumatisme","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital; Sunnybrook Hospital; Mount Sinai Hospital","funders":"","keywords":"Medicine; Humanities; Gynecology; Philosophy","score_opus":0.15403118193672483,"score_gpt":0.33993183251218906,"score_spread":0.18590065057546423,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1994361491","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.74262166,0.0025294528,0.24334122,0.0033333232,0.0005645528,0.0006278556,0.00008568197,0.00005784412,0.0068384195],"genre_scores_gemma":[0.9057411,0.0012722473,0.08808042,0.000090102476,0.00016256623,0.000113369286,0.00002975282,0.000039244824,0.0044711917],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99112475,0.004172295,0.0015578135,0.0007772787,0.0013837632,0.0009840957],"domain_scores_gemma":[0.9951031,0.002453404,0.00056160765,0.0009023833,0.0005587507,0.00042074817],"candidate_categories":["metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.012595964,0.00045833187,0.00059520965,0.00038628696,0.0021060682,0.00061435066,0.00058825,0.00034017643,0.0032242597],"category_scores_gemma":[0.0037674636,0.00046123948,0.00040350977,0.0014478231,0.0008976732,0.000585044,0.00012056969,0.00043975143,0.00021663783],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000016092295,0.0010870881,0.59971476,0.00020177646,0.00024071649,0.000028577559,0.06342823,0.15401219,0.0042551914,0.06606625,0.0013931681,0.109556],"study_design_scores_gemma":[0.0021336328,0.00030467505,0.5868321,0.0012041692,0.00087023235,0.0003106327,0.034180447,0.09719093,0.005908548,0.22613531,0.043786198,0.001143139],"about_ca_topic_score_codex":0.0029734008,"about_ca_topic_score_gemma":0.0084022805,"teacher_disagreement_score":0.16311945,"about_ca_system_score_codex":0.0013423575,"about_ca_system_score_gemma":0.0012055614,"threshold_uncertainty_score":0.99978393},"labels":[],"label_agreement":null},{"id":"W1995530352","doi":"10.2466/pr0.2000.87.3.791","title":"Expertise Sensitive Item Selection","year":2000,"lang":"en","type":"article","venue":"Psychological Reports","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Christian Studies; University of Toronto; Nipissing University","funders":"","keywords":"Certification; Index (typography); Test (biology); Selection (genetic algorithm); Psychology; Computer science; Applied psychology; Statistics; Artificial intelligence; Mathematics; World Wide Web; Management","score_opus":0.1941270554244931,"score_gpt":0.44065364488865594,"score_spread":0.24652658946416284,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1995530352","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8961761,0.000031637297,0.0010577341,0.0011318126,0.0004197185,0.00021711946,6.6059454e-7,0.00008610785,0.10087913],"genre_scores_gemma":[0.9904013,0.000024923303,0.000635016,0.001246973,0.00013438429,0.000022457474,0.0000013235191,0.000004588633,0.0075290073],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.996077,0.00031357925,0.00086452364,0.0008635939,0.0015959286,0.00028534626],"domain_scores_gemma":[0.9984762,0.00037214745,0.00018497644,0.00057767465,0.00023283467,0.00015614493],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0037041437,0.00014012853,0.00023560823,0.00006299002,0.00016517243,0.00013995651,0.00018036309,0.00010216017,0.011549231],"category_scores_gemma":[0.0028796683,0.000085477324,0.00015650272,0.00052617915,0.00009895057,0.00015963054,0.000023806902,0.0001727012,0.0011615193],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001289094,0.0006316629,0.035466623,0.0000010598347,0.000013148037,0.0005252873,0.00030228196,0.00020994757,0.0053719706,0.00006867169,0.04574989,0.91153055],"study_design_scores_gemma":[0.00026296006,0.00047792017,0.7654434,0.000018916991,0.0000106905845,0.0009451143,0.0001490447,0.00045604949,0.0029209428,0.04332236,0.18563867,0.00035398043],"about_ca_topic_score_codex":0.000011876377,"about_ca_topic_score_gemma":0.000004454621,"teacher_disagreement_score":0.91117656,"about_ca_system_score_codex":0.000054946155,"about_ca_system_score_gemma":0.000009902681,"threshold_uncertainty_score":0.9996162},"labels":[],"label_agreement":null},{"id":"W2004544084","doi":"10.1016/j.jclinepi.2006.02.006","title":"Interrater reliability of measurements of comorbid illness should be reported","year":2006,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":59,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Baycrest Hospital; Queen's University","funders":"Cancer Care Ontario","keywords":"Inter-rater reliability; Intraclass correlation; Medicine; Reliability (semiconductor); Rating scale; Comorbidity; Physical therapy; Psychometrics; Statistics; Psychiatry; Clinical psychology","score_opus":0.7802261328157349,"score_gpt":0.5869670242980766,"score_spread":0.19325910851765837,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2004544084","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9735734,0.00019610419,0.008541066,0.012205254,0.0028968835,0.00020632082,0.000010195643,0.000005623547,0.0023651654],"genre_scores_gemma":[0.9925303,0.000052625735,0.005791944,0.0010882826,0.000336233,0.000001971002,0.0000020150997,0.000007775091,0.00018886331],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.95847356,0.013246404,0.02447588,0.00063569995,0.002738953,0.00042952268],"domain_scores_gemma":[0.9320835,0.043728016,0.017746508,0.0014126492,0.0047648293,0.00026448784],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.23719656,0.00021501684,0.0037759547,0.00022741384,0.000052696167,0.000010654765,0.0012600612,0.00041454064,0.000707714],"category_scores_gemma":[0.32174549,0.00012484074,0.001502039,0.00040454417,0.0010142839,0.0002259124,0.0001841061,0.0008228759,0.0000099208355],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006809109,0.00097392546,0.9587648,0.00002925989,0.000102891834,0.000007685612,0.000043225613,0.0018344288,0.00088863826,0.0002822942,0.030881494,0.0055104685],"study_design_scores_gemma":[0.0015279579,0.0013367429,0.8491168,0.00014163667,0.00011793443,0.000029337918,0.00014125096,0.00031808615,0.0027449494,0.12697586,0.017384084,0.00016537376],"about_ca_topic_score_codex":0.0001835824,"about_ca_topic_score_gemma":0.000060767903,"teacher_disagreement_score":0.12669358,"about_ca_system_score_codex":0.000054714274,"about_ca_system_score_gemma":0.00027838617,"threshold_uncertainty_score":0.7854667},"labels":[],"label_agreement":null},{"id":"W2007345088","doi":"10.1371/journal.pone.0073990","title":"The Case for Using the Repeatability Coefficient When Calculating Test–Retest Reliability","year":2013,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":474,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"Curtin University of Technology","keywords":"Intraclass correlation; Repeatability; Reliability (semiconductor); Strengths and weaknesses; Computer science; Pearson product-moment correlation coefficient; Statistics; Test (biology); Measure (data warehouse); Reliability engineering; Correlation coefficient; Data mining; Medical physics; Machine learning; Mathematics; Reproducibility; Medicine; Psychology; Engineering","score_opus":0.29659504026263267,"score_gpt":0.3528709709396792,"score_spread":0.056275930677046504,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2007345088","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9828515,0.00011992285,0.0053063217,0.008109681,0.00015496589,0.0029437675,0.000031382795,0.000048538262,0.00043393846],"genre_scores_gemma":[0.9882015,0.000002757638,0.010703216,0.00026686775,0.00014050788,0.0002682327,0.0000017060972,0.0000128964675,0.00040233642],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9948629,0.0006426198,0.0012202653,0.0007740533,0.0019762323,0.00052392326],"domain_scores_gemma":[0.9732891,0.020826532,0.00048520838,0.0029533245,0.002271377,0.00017447214],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.017753977,0.00020721453,0.00033213265,0.0000356502,0.001985018,0.00079362205,0.0010268501,0.00007934347,0.000290215],"category_scores_gemma":[0.06078167,0.0000963998,0.00020369144,0.00040532817,0.00049382326,0.00027709137,0.00031545595,0.00026495825,0.00009350516],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00035399207,0.0180725,0.781798,0.00062403973,0.0005908543,0.00005420441,0.009725716,0.015530758,0.06705296,0.004561966,0.017620329,0.08401469],"study_design_scores_gemma":[0.0008324501,0.0005734155,0.0155393435,0.00017765639,0.00026096648,0.00003672189,0.0067096213,0.85000336,0.011095144,0.11136771,0.002813156,0.0005904653],"about_ca_topic_score_codex":0.00055387546,"about_ca_topic_score_gemma":0.00017016722,"teacher_disagreement_score":0.8344726,"about_ca_system_score_codex":0.00020708419,"about_ca_system_score_gemma":0.00009606268,"threshold_uncertainty_score":0.99931425},"labels":[],"label_agreement":null},{"id":"W2008192038","doi":"10.1007/s00477-012-0561-5","title":"Parameter and quantile estimation of the 2-parameter kappa distribution by maximum likelihood","year":2012,"lang":"en","type":"article","venue":"Stochastic Environmental Research and Risk Assessment","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Moncton","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Quantile; Kappa; Mathematics; Statistics; Maximum likelihood; Shape parameter; Estimation theory; Estimation; Distribution (mathematics); Applied mathematics; Econometrics; Mathematical analysis; Economics","score_opus":0.07148164151909714,"score_gpt":0.3891359155073917,"score_spread":0.31765427398829454,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2008192038","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92406857,0.000933395,0.07351641,0.0003697983,0.00013297315,0.0006427586,0.00026438577,0.0000045172615,0.00006720391],"genre_scores_gemma":[0.9981385,0.0003781424,0.0012503925,0.000015585669,0.000025240975,0.00006058919,0.000026200452,0.0000070461015,0.00009832112],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99534225,0.00064235134,0.0004917309,0.00039641763,0.002617682,0.0005095538],"domain_scores_gemma":[0.9969898,0.0020477902,0.000195488,0.0004887805,0.00003940273,0.00023876945],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006540348,0.00015338592,0.00021370771,0.00006405938,0.00047384785,0.00013760863,0.00030389885,0.00007619471,0.00014221604],"category_scores_gemma":[0.0012873355,0.00008812145,0.00006573795,0.00020038434,0.00080328627,0.0003074245,0.00043863626,0.0003810059,0.000026310461],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015955519,0.0021709928,0.6711848,0.00004177249,0.00011481654,5.349438e-7,0.0007738524,0.00052988785,0.007147154,0.0011761895,0.0052385675,0.31146193],"study_design_scores_gemma":[0.0008513251,0.00083196774,0.8727956,0.000060555914,0.00005634855,0.0000057861034,0.0019714911,0.024252553,0.0025243845,0.09496585,0.0014392202,0.0002449585],"about_ca_topic_score_codex":0.000075016804,"about_ca_topic_score_gemma":0.0000059849062,"teacher_disagreement_score":0.31121695,"about_ca_system_score_codex":0.0001579489,"about_ca_system_score_gemma":0.000028339848,"threshold_uncertainty_score":0.36445025},"labels":[],"label_agreement":null},{"id":"W2012185914","doi":"10.1016/j.prevetmed.2009.07.011","title":"Traditional descriptive analysis and novel visual representation of diagnostic repeatability and reproducibility: Application to an infectious salmon anaemia virus RT-PCR assay","year":2009,"lang":"en","type":"article","venue":"Preventive Veterinary Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Fisheries and Oceans Canada; University of Prince Edward Island","funders":"","keywords":"Repeatability; Reproducibility; Homogenization (climate); Transferability; Replicate; Biology; Statistics; Veterinary medicine; Mathematics; Pathology; Medicine; Ecology","score_opus":0.17558859261345333,"score_gpt":0.41348204531252536,"score_spread":0.23789345269907203,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2012185914","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8947136,0.000114138245,0.102875784,0.0009859212,0.0001232867,0.00097348937,0.000040520357,0.000025939868,0.00014736045],"genre_scores_gemma":[0.9984301,0.00004829653,0.0011161598,0.00012092346,0.000109382694,0.000072710165,0.000062955456,0.000005393776,0.00003411892],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9946526,0.0010603179,0.0010721369,0.0017808306,0.0012437798,0.00019030225],"domain_scores_gemma":[0.9949423,0.002543474,0.00041871233,0.0013039316,0.0005699664,0.00022161985],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.012636157,0.00020770283,0.0006121355,0.0004775247,0.00014964765,0.000041013063,0.00022778533,0.00007191892,0.00008965742],"category_scores_gemma":[0.023151278,0.00016265045,0.000096732634,0.0017354303,0.00032986587,0.000495171,0.00010116581,0.000139026,0.0000024935814],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00080186035,0.00213266,0.42353192,0.000050135488,0.0002664159,0.0000091488355,0.0050416593,0.0004056152,0.35154748,0.000460268,0.00015320635,0.21559964],"study_design_scores_gemma":[0.0005781424,0.0042733713,0.9694546,0.000047486155,0.0002840588,0.000014553914,0.00086444267,0.0018457149,0.0031461087,0.019229474,0.000107820204,0.00015424786],"about_ca_topic_score_codex":0.00047513144,"about_ca_topic_score_gemma":0.00010519947,"teacher_disagreement_score":0.5459227,"about_ca_system_score_codex":0.000111393514,"about_ca_system_score_gemma":0.000031261763,"threshold_uncertainty_score":0.98507714},"labels":[],"label_agreement":null},{"id":"W2014089987","doi":"10.1002/j.2334-5837.2007.tb02909.x","title":"4.6.3 A Self‐assessment Scheme and an Evaluation of its Reliability based on ISO 9004:2000","year":2007,"lang":"en","type":"article","venue":"INCOSE International Symposium","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"","keywords":"Reliability (semiconductor); Audit; Computer science; Quality (philosophy); Self-assessment; Process (computing); Scheme (mathematics); Quality audit; Reliability engineering; Risk analysis (engineering); Process management; Business; Accounting; Mathematics; Engineering; Psychology","score_opus":0.09485833129002842,"score_gpt":0.4272861330272782,"score_spread":0.3324278017372498,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2014089987","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9705305,0.000033240878,0.0026543923,0.0028269764,0.0008415484,0.00065316504,0.000029569648,0.000041418945,0.02238915],"genre_scores_gemma":[0.993547,0.000010634002,0.005777855,0.00036499283,0.00016281508,0.0000314454,0.000014788311,0.000009207724,0.000081256854],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9917185,0.00038411125,0.0010382697,0.00066855655,0.0059733703,0.00021723144],"domain_scores_gemma":[0.9953817,0.0009120997,0.0003908273,0.0006367532,0.0025283021,0.00015031527],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.025743078,0.00017982147,0.00023045913,0.0002507764,0.00011047237,0.00014524818,0.000693433,0.00010509326,0.0006479923],"category_scores_gemma":[0.0011375058,0.00014142165,0.000103838676,0.0003285163,0.00008852053,0.00046061678,0.00010608362,0.00017772752,0.00004949821],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020757848,0.013719675,0.5915393,0.00015173848,0.00028331744,0.000022359261,0.0034811771,0.062832944,0.16874143,0.022555346,0.001876824,0.13272005],"study_design_scores_gemma":[0.0024942616,0.0008141355,0.22378163,0.00010438145,0.00005450393,0.00000361621,0.00028527947,0.72374606,0.02619569,0.012342821,0.009835157,0.0003424318],"about_ca_topic_score_codex":0.000037767095,"about_ca_topic_score_gemma":0.000053769858,"teacher_disagreement_score":0.66091317,"about_ca_system_score_codex":0.000788759,"about_ca_system_score_gemma":0.00031878112,"threshold_uncertainty_score":0.8922088},"labels":[],"label_agreement":null},{"id":"W2016242508","doi":"10.1016/j.jclinepi.2010.12.001","title":"The difference between reliability and agreement","year":2011,"lang":"en","type":"letter","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":93,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Scopus; Inter-rater reliability; Intraclass correlation; Reliability (semiconductor); Kappa; Cohen's kappa; Statistics; Psychology; Medicine; Mathematics; MEDLINE; Physics; Psychometrics; Political science","score_opus":0.6917992205636752,"score_gpt":0.5490965920628558,"score_spread":0.14270262850081938,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2016242508","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03808391,0.001321954,0.0051259855,0.9488102,0.0052436832,0.00033475054,0.000017311591,0.000006484633,0.0010557229],"genre_scores_gemma":[0.03287585,0.008884266,0.00660981,0.91987073,0.025358975,0.000016647014,0.000006767917,0.000037137786,0.006339813],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9308183,0.04178641,0.022028076,0.0012870965,0.0030885243,0.0009915679],"domain_scores_gemma":[0.5006245,0.4758477,0.018300792,0.0024824105,0.002227405,0.00051725784],"candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.29836318,0.0004179197,0.0049178526,0.00015859165,0.0002974881,0.000093269206,0.0032785386,0.0016731332,0.00029065693],"category_scores_gemma":[0.48132014,0.0001817077,0.0016474816,0.00020336782,0.0020815318,0.00011783543,0.0005648548,0.0072759176,0.00013399884],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006318836,0.000034193465,0.35788372,0.000013403757,0.00010436449,0.000016394744,0.000018037903,0.00000147778,1.8907208e-7,0.000097937176,0.55782825,0.08393882],"study_design_scores_gemma":[0.00020756433,0.0005425882,0.32198378,0.00004897955,0.00007399208,0.00000903469,0.0000120552795,0.0000092346345,5.966838e-7,0.21733142,0.45968404,0.00009669408],"about_ca_topic_score_codex":0.000045142664,"about_ca_topic_score_gemma":0.000015826383,"teacher_disagreement_score":0.43406126,"about_ca_system_score_codex":0.00006771901,"about_ca_system_score_gemma":0.00036662386,"threshold_uncertainty_score":0.9996229},"labels":[],"label_agreement":null},{"id":"W2017183230","doi":"10.1002/bsl.730","title":"Do tests of malingering concur? Concordance among malingering measures","year":2006,"lang":"en","type":"article","venue":"Behavioral Sciences & the Law","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Columbia College","funders":"","keywords":"Malingering; Concordance; Psychology; Clinical psychology; Psychiatry; Medicine; Internal medicine","score_opus":0.22399606406696035,"score_gpt":0.40694419476798366,"score_spread":0.1829481307010233,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2017183230","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9902506,0.00048211307,0.0003560374,0.000681859,0.0006199456,0.00035321806,0.000008699373,0.00004475834,0.007202785],"genre_scores_gemma":[0.9986449,0.00000773272,0.00082146405,0.00009307331,0.00013290523,0.000024580813,5.4253684e-7,0.000007791742,0.00026703504],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99417907,0.0002603115,0.0010180528,0.0006713318,0.00335827,0.0005129859],"domain_scores_gemma":[0.9977664,0.0005293272,0.00051660056,0.0007239845,0.0003752808,0.00008841187],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009982182,0.00021847019,0.00036999077,0.00009138393,0.0007465605,0.0005942475,0.0021196918,0.00007020236,0.00019816848],"category_scores_gemma":[0.00026820033,0.0001256623,0.00019184448,0.0010666618,0.0024743383,0.00056477584,0.00027035654,0.00018372672,0.000042346328],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000026516085,0.00029363262,0.86701906,0.000011560787,0.00000724807,0.000010821672,0.0006724084,0.010203887,0.07573075,0.032976538,0.00079749635,0.012250082],"study_design_scores_gemma":[0.0010867036,0.00085126923,0.74497783,0.0004421723,0.0001369253,0.000021047024,0.004289848,0.0014083063,0.1314704,0.08528712,0.028796704,0.0012316707],"about_ca_topic_score_codex":0.0035670614,"about_ca_topic_score_gemma":0.0021041462,"teacher_disagreement_score":0.122041225,"about_ca_system_score_codex":0.000058419497,"about_ca_system_score_gemma":0.00007132975,"threshold_uncertainty_score":0.9116802},"labels":[],"label_agreement":null},{"id":"W2018205196","doi":"10.1111/j.0006-341x.2001.00584.x","title":"Inference Procedures for Assessing Interobserver Agreement among Multiple Raters","year":2001,"lang":"en","type":"article","venue":"Biometrics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Goodness of fit; Mathematics; Nominal level; Confidence interval; Binomial distribution; Range (aeronautics); Inference; Sample size determination; Binomial (polynomial); Multiple comparisons problem; Statistical inference; Computer science; Artificial intelligence","score_opus":0.3335008582243348,"score_gpt":0.43055005559190307,"score_spread":0.09704919736756829,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2018205196","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8721611,0.00014999896,0.12491678,0.000505751,0.00082519476,0.000651903,0.000011173411,0.000039598854,0.00073851313],"genre_scores_gemma":[0.99216366,0.000046035082,0.0061534094,0.00025553256,0.00009918466,0.00007141311,0.000008675055,0.0000097453185,0.0011923565],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99618274,0.00012371778,0.00082200987,0.00062715204,0.0018246293,0.00041972977],"domain_scores_gemma":[0.99592364,0.0021353406,0.00034410984,0.0005887104,0.00084828393,0.0001598927],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.005012,0.00019633934,0.000280413,0.0012251597,0.00024613933,0.0008785136,0.00090784056,0.00009975907,0.0002331778],"category_scores_gemma":[0.023816785,0.00013870139,0.00017146148,0.0047872053,0.00012229061,0.00066699524,0.00018218445,0.00009668478,0.000117019656],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000326098,0.0002356668,0.93248504,0.00003115914,0.000021093714,0.0000025583538,0.00016626838,0.00023009913,0.0025908188,0.000077721576,0.007986862,0.05614012],"study_design_scores_gemma":[0.0012716079,0.0002642357,0.9088506,0.000101563266,0.000027130834,0.0000015962643,0.0012630161,0.008810538,0.00356037,0.0042660655,0.07113828,0.00044501576],"about_ca_topic_score_codex":0.00004830451,"about_ca_topic_score_gemma":0.000111690824,"teacher_disagreement_score":0.12000256,"about_ca_system_score_codex":0.00015346683,"about_ca_system_score_gemma":0.00009355439,"threshold_uncertainty_score":0.984406},"labels":[],"label_agreement":null},{"id":"W2022913092","doi":"10.1007/s11135-013-9919-0","title":"On the choice of measures of reliability and validity in the content-analysis of texts","year":2013,"lang":"en","type":"article","venue":"Quality & Quantity","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":61,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland","funders":"","keywords":"Reliability (semiconductor); Rhetorical question; Interpretation (philosophy); Reading comprehension; Complement (music); Reading (process); Kappa; Computer science; Measure (data warehouse); Content (measure theory); Psychology; Natural language processing; Content validity; Comprehension; Linguistics; Statistics; Social psychology; Cognitive psychology; Mathematics; Psychometrics; Data mining; Philosophy","score_opus":0.6107004055274742,"score_gpt":0.45660162382610464,"score_spread":0.15409878170136954,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2022913092","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9949781,0.0000858313,0.0003830842,0.0031357917,0.00007052761,0.0006311501,0.00005094063,0.0000043760488,0.00066023064],"genre_scores_gemma":[0.999601,0.000021787919,0.000070839065,0.00024715674,0.0000079173515,0.000018095914,0.0000014376602,0.0000023753084,0.000029369792],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9890718,0.0051936097,0.0019511173,0.0004576677,0.0031326795,0.00019313654],"domain_scores_gemma":[0.97636014,0.01916376,0.0011266093,0.002017598,0.001278828,0.00005305616],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.048469774,0.00015113108,0.000825978,0.0001943807,0.00010079612,0.00006156922,0.0011487524,0.000083528495,0.00041740484],"category_scores_gemma":[0.039300732,0.00006999595,0.00040318002,0.0015916982,0.0007542975,0.00019746488,0.0001229438,0.00022511093,0.000011677292],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000056798122,0.0006928099,0.9720294,0.000047767793,0.00013130176,6.744084e-8,0.001274483,0.00027174212,0.0033635385,0.019972105,0.00061809976,0.0015419121],"study_design_scores_gemma":[0.00019186796,0.00012779278,0.9583821,0.000020557307,0.00013010908,4.0477122e-8,0.0018757089,0.00041870287,0.0025909173,0.036053963,0.00013336653,0.00007483747],"about_ca_topic_score_codex":0.016729683,"about_ca_topic_score_gemma":0.004504059,"teacher_disagreement_score":0.01608186,"about_ca_system_score_codex":0.000028484483,"about_ca_system_score_gemma":0.00004681244,"threshold_uncertainty_score":0.989818},"labels":[],"label_agreement":null},{"id":"W2023293941","doi":"10.2190/iq.29.3.c","title":"High-Risk Patients' Readiness to Undergo BMD Testing for Osteoporosis Diagnosis in Pennsylvania","year":2009,"lang":"en","type":"article","venue":"International Quarterly of Community Health Education","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"National Institute of Arthritis and Musculoskeletal and Skin Diseases; Canadian Institutes of Health Research; U.S. Public Health Service","keywords":"Osteoporosis; Medicine; Psychological intervention; Bone mineral; Physical therapy; Randomized controlled trial; Summative assessment; Internal medicine; Psychology; Formative assessment","score_opus":0.12825880433268882,"score_gpt":0.4289646787318542,"score_spread":0.30070587439916535,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2023293941","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97376937,0.00002192931,0.0020072944,0.021548068,0.0012888693,0.00097384286,0.00006870775,0.000016485474,0.0003054568],"genre_scores_gemma":[0.9878756,0.000007326023,0.010015136,0.0016827206,0.0000956807,0.0001823398,0.00007973762,0.000006584838,0.00005492076],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99625796,0.0008483177,0.0013263277,0.00024256449,0.00109337,0.00023146386],"domain_scores_gemma":[0.9926607,0.003928126,0.00079141406,0.0006549988,0.0018180638,0.00014669073],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0060002035,0.00014359062,0.0003096577,0.00051294075,0.00029627903,0.00013725495,0.0010914397,0.00005543916,0.00005471483],"category_scores_gemma":[0.006490167,0.0001336256,0.00007579874,0.0006341336,0.000033799333,0.00038017056,0.00003105013,0.0002605196,0.000025949084],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000718522,0.0022351858,0.25316888,0.000021192605,0.000006445653,1.4425252e-8,0.007453401,0.00025895302,0.00003405258,0.0012704906,0.0074923416,0.7279872],"study_design_scores_gemma":[0.0005793802,0.0032679555,0.8916311,0.00016714528,0.000004732551,2.2044014e-7,0.0066037728,0.000089218855,0.000050820687,0.09487839,0.0026093423,0.00011787513],"about_ca_topic_score_codex":0.015383683,"about_ca_topic_score_gemma":0.0060186004,"teacher_disagreement_score":0.72786933,"about_ca_system_score_codex":0.0004910198,"about_ca_system_score_gemma":0.00041338857,"threshold_uncertainty_score":0.99117297},"labels":[],"label_agreement":null},{"id":"W2030560722","doi":"10.1136/bmj.b4012","title":"Risk of bias versus quality assessment of randomised controlled trials: cross sectional study","year":2009,"lang":"en","type":"article","venue":"BMJ","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":345,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Jadad scale; Medicine; Kappa; Publication bias; Funnel plot; Sample size determination; Cohen's kappa; Statistics; Meta-analysis; Correlation; Relative risk; Internal medicine; Mathematics; Confidence interval","score_opus":0.553386919390064,"score_gpt":0.5654401656422524,"score_spread":0.012053246252188399,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2030560722","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9869348,0.000047319536,0.0018747213,0.00033707172,0.0010125388,0.0035102514,0.000040293242,0.000011966768,0.006231035],"genre_scores_gemma":[0.9988416,0.0000081714215,0.00066825765,0.000024796267,0.00014189738,0.000071141745,0.0000016654989,0.000002430903,0.000240031],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9796926,0.010142384,0.005047834,0.00043886589,0.004502318,0.00017601822],"domain_scores_gemma":[0.96703285,0.02743007,0.0034042653,0.00082950987,0.0012284373,0.00007488337],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.19993502,0.00015046846,0.0018019879,0.0001492118,0.00011815191,0.00011044177,0.0005073936,0.00007140502,0.0007189066],"category_scores_gemma":[0.11503973,0.00008583317,0.00090391655,0.00036276094,0.00009896282,0.00012527165,0.000043902306,0.00013958539,0.000011934172],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.061338093,0.0027565146,0.9219787,0.0000065834174,0.00060319214,0.0000010476958,0.00034185534,0.004842432,0.0010439937,0.00050069316,0.0020120912,0.0045747925],"study_design_scores_gemma":[0.15082878,0.0009760218,0.8422504,0.000006319858,0.00010749868,1.0220105e-7,0.00071505527,0.0005874126,0.00026337293,0.004070699,0.00010937701,0.000084953215],"about_ca_topic_score_codex":0.000117359574,"about_ca_topic_score_gemma":0.000042653108,"teacher_disagreement_score":0.08949069,"about_ca_system_score_codex":0.000061039405,"about_ca_system_score_gemma":0.00020372683,"threshold_uncertainty_score":0.8924147},"labels":[],"label_agreement":null},{"id":"W2032172756","doi":"10.1016/j.ygyno.2007.07.064","title":"p16 and MIB1 improve the sensitivity and specificity of the diagnosis of high grade squamous intraepithelial lesions: Methodological issues in a report of 447 biopsies with consensus diagnosis and HPV HCII testing","year":2007,"lang":"en","type":"article","venue":"Gynecologic Oncology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Vancouver General Hospital; BC Cancer Agency","funders":"National Cancer Institute","keywords":"Medicine; Grading (engineering); Biopsy; Pathology; Atypia; Intraepithelial neoplasia; Internal medicine; Cancer; Biology","score_opus":0.2530757024059997,"score_gpt":0.41026272798023233,"score_spread":0.15718702557423264,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2032172756","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9897448,0.0006949196,0.00008757241,0.008145832,0.00021566507,0.0008867084,0.000029768265,0.000009700377,0.00018505674],"genre_scores_gemma":[0.9882707,0.0004308253,0.011064599,0.00013725561,0.000033089047,0.000046627938,4.106354e-7,0.0000045080437,0.000011999354],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.994445,0.00266244,0.001319163,0.0006085104,0.0006453602,0.00031948663],"domain_scores_gemma":[0.92030483,0.077573836,0.0011415745,0.00048758698,0.00042170187,0.00007048513],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.020931672,0.00020691156,0.00095014315,0.00013602416,0.00017490849,0.000021619579,0.00026321848,0.000252551,0.000035788038],"category_scores_gemma":[0.06493196,0.00009431463,0.000067517525,0.00058784697,0.003017702,0.00005047506,0.00042255464,0.00030807932,2.8033298e-7],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00020922697,0.00056508835,0.90502226,0.00003631684,0.000025909192,0.00013934993,0.0007301818,0.00006189223,0.0062473887,0.00045320767,0.00010329541,0.08640588],"study_design_scores_gemma":[0.00068886945,0.0019424815,0.9548075,0.000034121098,0.00006085212,0.00017946567,0.0038089838,0.000069061134,0.027455427,0.010349041,0.00048627597,0.00011790951],"about_ca_topic_score_codex":0.0014289409,"about_ca_topic_score_gemma":0.004308373,"teacher_disagreement_score":0.08628797,"about_ca_system_score_codex":0.00006426038,"about_ca_system_score_gemma":0.00010851438,"threshold_uncertainty_score":0.99969554},"labels":[],"label_agreement":null},{"id":"W2036060563","doi":"10.1186/1471-2288-10-82","title":"Inter-rater agreement and reliability of the COSMIN (COnsensus-based Standards for the selection of health status Measurement Instruments) Checklist","year":2010,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":289,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"Vrije Universiteit Amsterdam; University of Oxford; McMaster University; Goddard Space Flight Center; University of Washington","keywords":"Checklist; Kappa; Inter-rater reliability; Intraclass correlation; Cohen's kappa; Reliability (semiconductor); Medicine; Quality (philosophy); Agreement; Terminology; Psychology; Applied psychology; Physical therapy; Medical physics; Clinical psychology; Statistics; Psychometrics; Rating scale; Mathematics","score_opus":0.5948043968788056,"score_gpt":0.5666070490965269,"score_spread":0.02819734778227867,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2036060563","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7360402,0.00026790256,0.23604037,0.023667121,0.0011681955,0.0025354896,0.00008794098,0.000008788287,0.00018402388],"genre_scores_gemma":[0.9704114,0.00007649354,0.028815242,0.0003273527,0.00007062288,0.00017400643,0.0000013686418,0.000008853127,0.00011463904],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9401533,0.03720393,0.0030697882,0.0011458405,0.01720643,0.0012206894],"domain_scores_gemma":[0.93022543,0.06034505,0.00081088947,0.0016480009,0.006530828,0.00043981196],"candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.39401495,0.00016993107,0.0006664849,0.00020724724,0.00048070043,0.000051671,0.001216097,0.00020023438,0.0010586313],"category_scores_gemma":[0.38449427,0.000079416146,0.00021222742,0.0007389415,0.0028427758,0.000042088726,0.00044543628,0.00094143,0.0000024414717],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002453449,0.00089207065,0.36769968,0.00085631653,0.00009789977,3.107363e-7,0.0010337174,0.00013585147,0.01665502,0.0025695155,0.018010149,0.58959603],"study_design_scores_gemma":[0.007085396,0.0044682357,0.56682533,0.0006157815,0.000078459896,0.000008793487,0.0054242373,0.026630612,0.07637274,0.07185221,0.2402325,0.00040569247],"about_ca_topic_score_codex":0.001463431,"about_ca_topic_score_gemma":0.009785491,"teacher_disagreement_score":0.5891903,"about_ca_system_score_codex":0.00033257526,"about_ca_system_score_gemma":0.00629961,"threshold_uncertainty_score":0.9998709},"labels":[],"label_agreement":null},{"id":"W2040932637","doi":"10.1002/bimj.200310154","title":"A Non-iterative Confidence Interval Estimating Procedure for the Intraclass Kappa Statistic with Multinomial Outcomes","year":2005,"lang":"en","type":"article","venue":"Biometrical Journal","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cancer Care Ontario; Western University; Robarts Clinical Trials","funders":"","keywords":"Multinomial distribution; Confidence interval; Statistics; Mathematics; Statistic; Intraclass correlation; Interval estimation; Sample size determination; Kappa; Interval (graph theory); Combinatorics; Reproducibility","score_opus":0.140185236435446,"score_gpt":0.4138179222094404,"score_spread":0.2736326857739944,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2040932637","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09867907,0.00012292247,0.8916852,0.008117249,0.0005978428,0.00065948284,0.000026494718,0.000013504448,0.00009821721],"genre_scores_gemma":[0.84448206,0.0000038980484,0.15414025,0.00052883156,0.0005445995,0.000038398546,8.1466754e-7,0.000008505739,0.00025264564],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9958843,0.00014786771,0.0010253862,0.00038582511,0.002161113,0.00039548095],"domain_scores_gemma":[0.99115044,0.006740208,0.0005872285,0.00028721982,0.0010151966,0.00021971989],"candidate_categories":["metaresearch","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.0055830763,0.00020835847,0.0003798238,0.0005351294,0.0006037636,0.0011621212,0.00097523746,0.00006409598,0.00023293687],"category_scores_gemma":[0.01494174,0.00009076347,0.00017630206,0.0016603251,0.00022365576,0.0004142241,0.000091206035,0.00041379716,0.0000671584],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007558217,0.0007019387,0.07861964,0.000034714853,0.00030331686,0.000026233864,0.0027731487,0.0070920954,0.0009077514,0.000713978,0.030488208,0.87758315],"study_design_scores_gemma":[0.005600556,0.0024638714,0.21930234,0.0002614753,0.00018245346,0.00039596017,0.003316668,0.73352903,0.0012192527,0.008305657,0.024690786,0.0007319689],"about_ca_topic_score_codex":0.000007802533,"about_ca_topic_score_gemma":0.000017333792,"teacher_disagreement_score":0.8768512,"about_ca_system_score_codex":0.00017925356,"about_ca_system_score_gemma":0.00019187325,"threshold_uncertainty_score":0.9998748},"labels":[],"label_agreement":null},{"id":"W2042490722","doi":"10.1179/2042618611y.0000000001","title":"Clinician’s ability to identify neck and low back interventions: an inter-rater chance-corrected agreement pilot study","year":2011,"lang":"en","type":"article","venue":"Journal of Manual & Manipulative Therapy","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"U.S. Air Force; McGill University","keywords":"Medicine; Psychological intervention; Physical therapy; Physical medicine and rehabilitation; Agreement; Nursing","score_opus":0.5696956122938136,"score_gpt":0.500385077475087,"score_spread":0.06931053481872662,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2042490722","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9935605,0.00017151593,0.0029016018,0.00039928211,0.001387826,0.0012510779,0.00000632403,0.000012011352,0.00030987596],"genre_scores_gemma":[0.9975627,0.00006502833,0.0012027343,0.0005174544,0.00021945643,0.000020955833,0.0000011439652,0.000016026173,0.00039451738],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9928776,0.001852751,0.0025613236,0.0006433706,0.001723884,0.00034104177],"domain_scores_gemma":[0.9958706,0.0003513002,0.001237784,0.00077850133,0.0014511648,0.0003106313],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.012155573,0.00032909503,0.0007552526,0.0004314155,0.00017177689,0.00030478887,0.0011263822,0.00005378298,0.0036779626],"category_scores_gemma":[0.0006760941,0.00021601078,0.0002908815,0.00045816795,0.00013604265,0.0010587745,0.00023665435,0.00038821305,0.0002482516],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0078173755,0.019729044,0.8101723,0.000054333734,0.000522761,0.000045061795,0.039792962,0.00010008362,0.005054348,0.000071044626,0.0028267663,0.11381391],"study_design_scores_gemma":[0.002105385,0.026175626,0.95168716,0.00022912082,0.000033510907,0.000008471128,0.008377336,0.00005960496,0.00082600006,0.009816066,0.0004268963,0.00025481562],"about_ca_topic_score_codex":0.00013468236,"about_ca_topic_score_gemma":0.0005583385,"teacher_disagreement_score":0.14151485,"about_ca_system_score_codex":0.0001429059,"about_ca_system_score_gemma":0.000042005944,"threshold_uncertainty_score":0.9972328},"labels":[],"label_agreement":null},{"id":"W2043640852","doi":"10.1177/0962280213476771","title":"Adjusted inference procedures for the interobserver agreement in twin studies","year":2013,"lang":"en","type":"article","venue":"Statistical Methods in Medical Research","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"London Health Sciences Centre; Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Inference; Statistics; Confidence interval; Statistic; Statistical inference; Computer science; Interval estimation; Statistical hypothesis testing; Multiple comparisons problem; Sample size determination; Cohen's kappa; Kappa; Test statistic; Mathematics; Econometrics; Artificial intelligence","score_opus":0.7237431735697322,"score_gpt":0.680536609079811,"score_spread":0.04320656448992122,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2043640852","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.026959958,0.0020380772,0.91454524,0.049598075,0.0007740607,0.0044193463,0.000021690917,0.000018991695,0.0016245357],"genre_scores_gemma":[0.5705475,0.000859089,0.4213044,0.0015061229,0.00020029815,0.004205858,0.000003671205,0.000020061125,0.0013529769],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9807243,0.008307147,0.0017207309,0.00090811605,0.007288438,0.0010512819],"domain_scores_gemma":[0.50127786,0.49087238,0.0002603459,0.0020758954,0.0045612785,0.000952243],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.11421105,0.00019007131,0.0005605791,0.00039698128,0.00020797232,0.0002391555,0.0020414994,0.0001649981,0.0074444185],"category_scores_gemma":[0.6344833,0.00009572729,0.0000748693,0.0016241755,0.0015630176,0.00016783048,0.00085662625,0.0012807924,0.00024061234],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000114913724,0.00032995417,0.01147687,0.00014969136,0.000032768196,0.0000137489815,0.0012881386,0.000044688524,0.0001409662,0.028953765,0.032792144,0.92466235],"study_design_scores_gemma":[0.0010022672,0.00043636098,0.16511637,0.0004097923,0.0000064201377,9.625664e-7,0.009585211,0.037000734,0.00017070943,0.7768029,0.009294668,0.00017358357],"about_ca_topic_score_codex":0.000516375,"about_ca_topic_score_gemma":0.0021128883,"teacher_disagreement_score":0.9244888,"about_ca_system_score_codex":0.00026456517,"about_ca_system_score_gemma":0.0005156538,"threshold_uncertainty_score":0.9934629},"labels":[],"label_agreement":null},{"id":"W2044059462","doi":"10.1097/acm.0b013e31822a6cf8","title":"Rater-Based Assessments as Social Judgments: Rethinking the Etiology of Rater Errors","year":2011,"lang":"en","type":"review","venue":"Academic Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":194,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Northern British Columbia","funders":"","keywords":"Categorical variable; Psychology; Categorization; Impression formation; Inter-rater reliability; Social psychology; Construct (python library); Cognitive psychology; Social perception; Rating scale; Computer science; Perception; Developmental psychology; Artificial intelligence; Machine learning","score_opus":0.6437870321086094,"score_gpt":0.5584685302281239,"score_spread":0.08531850188048551,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2044059462","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00043198187,0.98379415,0.0007982573,0.004180676,0.0028544874,0.0016877365,0.000018816272,0.000034523564,0.006199378],"genre_scores_gemma":[0.001708759,0.9919846,0.00007063522,0.002715185,0.00086883456,0.00015050886,0.00005791099,0.000035614168,0.002407931],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9884249,0.0035780726,0.0034710108,0.0009705822,0.0030177727,0.00053766137],"domain_scores_gemma":[0.9924318,0.0031390765,0.0027678458,0.0010938613,0.00043548032,0.00013193542],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.026088126,0.00056508824,0.002658526,0.00041397102,0.0002885668,0.000021323067,0.0032904618,0.001061323,0.0040003196],"category_scores_gemma":[0.0044962573,0.00025341322,0.00045014083,0.0009071265,0.001110094,0.00012949905,0.00032616648,0.0022592654,0.00041889638],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009914445,0.00006793999,0.00044980418,0.0030626715,0.00046558687,0.000012263537,0.0048523033,0.000002718505,0.00003674251,0.004058517,0.12230534,0.86458695],"study_design_scores_gemma":[0.00049686444,0.00033201222,0.00003835015,0.0060032182,0.00065978075,0.0000066664825,0.0005592464,0.000007683276,0.000011967487,0.025695015,0.96594757,0.00024162667],"about_ca_topic_score_codex":0.00006995632,"about_ca_topic_score_gemma":0.0000042471206,"teacher_disagreement_score":0.8643453,"about_ca_system_score_codex":0.00017265289,"about_ca_system_score_gemma":0.0005659985,"threshold_uncertainty_score":0.99999183},"labels":[],"label_agreement":null},{"id":"W2044481915","doi":"10.1038/modpathol.3800496","title":"Interobserver agreement and reproducibility in classification of invasive breast carcinoma: an NCI breast cancer family registry study","year":2005,"lang":"en","type":"article","venue":"Modern Pathology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":116,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Mount Sinai Hospital; University of Toronto; Jewish General Hospital; Statistics Canada","funders":"National Cancer Institute; National Institutes of Health; National Institute of Informatics","keywords":"Medicine; Reproducibility; Breast cancer; Pathology; Breast carcinoma; Cancer registry; Cancer; Oncology; Radiology; Internal medicine; Epidemiology","score_opus":0.18480263545995282,"score_gpt":0.37383187807648444,"score_spread":0.18902924261653162,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2044481915","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9958175,0.0002205886,0.00033488456,0.0023675477,0.00020950788,0.0007483629,0.00008484458,0.0000129467535,0.00020382753],"genre_scores_gemma":[0.9992701,0.000019125271,0.00020312649,0.00018418336,0.00007187137,0.00008605952,0.000005619534,0.000008187208,0.00015174529],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9944434,0.0010881464,0.0011904789,0.002076925,0.00091346976,0.0002875441],"domain_scores_gemma":[0.995874,0.00017174984,0.00042460868,0.00292276,0.0005076555,0.000099217636],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0103280265,0.00019950511,0.00046938207,0.00017885609,0.00007587572,0.000049951137,0.0006467848,0.00011647243,0.00008361291],"category_scores_gemma":[0.00070370897,0.00015589719,0.00006273251,0.00030278612,0.0002690189,0.00036547484,0.00026933453,0.00021594622,0.000013170683],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017178222,0.0014034166,0.91665435,0.000015694282,0.000009188963,0.000011970165,0.004692938,0.00048543638,0.024698548,0.000028564018,0.00017425422,0.051653873],"study_design_scores_gemma":[0.0008592479,0.00022291164,0.98991364,0.000018878289,0.00001736796,0.00003423497,0.0037146101,0.0018936808,0.00028863476,0.0028696104,0.000025254953,0.00014192918],"about_ca_topic_score_codex":0.000774084,"about_ca_topic_score_gemma":0.009485027,"teacher_disagreement_score":0.0732593,"about_ca_system_score_codex":0.0002082196,"about_ca_system_score_gemma":0.00012578997,"threshold_uncertainty_score":0.63573027},"labels":[],"label_agreement":null},{"id":"W2050257625","doi":"10.1016/j.jclinepi.2007.10.023","title":"Reliability studies of diagnostic tests are not using enough observers for robust estimation of interobserver agreement: a simulation study","year":2008,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":14,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; University of British Columbia","funders":"","keywords":"Kappa; Reliability (semiconductor); Statistics; Sample (material); Variance (accounting); Cohen's kappa; Sample size determination; Variation (astronomy); Inter-rater reliability; Computer science; Population; Variable (mathematics); Econometrics; Mathematics; Medicine","score_opus":0.8782298280510391,"score_gpt":0.620131203012851,"score_spread":0.25809862503818815,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2050257625","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94049704,0.0002841422,0.055926852,0.001284826,0.001270666,0.0007174982,0.0000116297415,0.0000037520517,0.0000035779092],"genre_scores_gemma":[0.9608097,0.00018044128,0.038544606,0.0002508945,0.00018696935,0.0000063167154,8.5048504e-7,0.000006782358,0.000013448677],"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","domain_scores_codex":[0.9774507,0.0074603464,0.012924083,0.0005308105,0.0013357827,0.00029826805],"domain_scores_gemma":[0.50787836,0.46410236,0.019915951,0.0011196557,0.006739813,0.0002438418],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.10365155,0.00020713432,0.0032790226,0.00018394706,0.00012730059,0.0000067168394,0.0006826267,0.0002001976,0.000038086535],"category_scores_gemma":[0.77846193,0.00013272808,0.0009856916,0.0003570678,0.00064795476,0.0003874422,0.00016355424,0.0003870304,0.0000031269333],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00046134985,0.0010251763,0.4723378,0.000048839516,0.00012662531,0.0000025915162,0.00030481518,0.52329993,0.000012432991,0.000021214799,0.00027410037,0.0020850939],"study_design_scores_gemma":[0.0023098788,0.0032548944,0.85579324,0.00025974284,0.00017818222,0.0000041121148,0.0011434099,0.12633163,0.000037989386,0.010508118,0.000076642114,0.000102141355],"about_ca_topic_score_codex":0.000033948214,"about_ca_topic_score_gemma":0.000035368124,"teacher_disagreement_score":0.6748104,"about_ca_system_score_codex":0.00010981267,"about_ca_system_score_gemma":0.0001797572,"threshold_uncertainty_score":0.92297935},"labels":[],"label_agreement":null},{"id":"W2053630609","doi":"10.1097/ede.0b013e318207fc5c","title":"Adjusting for Differential-verification Bias in Diagnostic-accuracy Studies","year":2011,"lang":"en","type":"article","venue":"Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Test (biology); Computer science; Statistics; Bayesian probability; Reference values; Population; Data mining; Medicine; Mathematics; Internal medicine","score_opus":0.8295205122545626,"score_gpt":0.5205216194963798,"score_spread":0.3089988927581828,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2053630609","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.93891716,0.0016695531,0.0518718,0.0040846323,0.0016185105,0.0008651493,0.000009959593,0.00003886551,0.0009243716],"genre_scores_gemma":[0.98937017,0.0003911816,0.0089907525,0.00067856803,0.00015003786,0.00024794,0.0000049862856,0.0000062333734,0.00016011237],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9955817,0.0015023822,0.0016553386,0.0006008938,0.0002665063,0.00039318018],"domain_scores_gemma":[0.8709549,0.12690732,0.0008348433,0.0007697102,0.00044303873,0.00009024332],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01959785,0.00014316505,0.0006042516,0.00015583314,0.00012096995,0.000010769901,0.0005035346,0.00010116181,0.00027941697],"category_scores_gemma":[0.55152345,0.00009557728,0.00013110686,0.00025238158,0.00017491814,0.00014453164,0.000108778564,0.00012719873,0.00012928959],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009680018,0.0002284795,0.84613174,0.000037116304,0.000043153665,0.0000020450898,0.0029730198,0.0002010463,0.00041053863,0.023284985,0.017585274,0.10900581],"study_design_scores_gemma":[0.00040920178,0.00016835438,0.6486715,0.00004745088,0.000017985958,0.0000010240556,0.0017153916,0.0013461866,0.00046712515,0.33677542,0.010221987,0.00015834431],"about_ca_topic_score_codex":0.00009449418,"about_ca_topic_score_gemma":0.00014863232,"teacher_disagreement_score":0.5319256,"about_ca_system_score_codex":0.000059625254,"about_ca_system_score_gemma":0.000026404185,"threshold_uncertainty_score":0.67922634},"labels":[],"label_agreement":null},{"id":"W2055948452","doi":"10.1016/j.jmpt.2006.06.009","title":"Applying Generalizability Theory to High-Stakes Objective Structured Clinical Examinations in a Naturalistic Environment","year":2006,"lang":"en","type":"article","venue":"Journal of Manipulative and Physiological Therapeutics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Generalizability theory; Cronbach's alpha; Medicine; Variance (accounting); Chiropractic; Checklist; Objective structured clinical examination; Reliability (semiconductor); Conceptualization; Licensure; Statistics; Applied psychology; Medical education; Clinical psychology; Psychology; Psychometrics; Alternative medicine; Mathematics; Computer science; Cognitive psychology","score_opus":0.5803482065378721,"score_gpt":0.4441747944179522,"score_spread":0.1361734121199199,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2055948452","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98699915,0.00030771107,0.0116428165,0.00035393966,0.00019589344,0.000407982,0.00000873873,0.0000039291504,0.000079864614],"genre_scores_gemma":[0.99589306,0.00007273819,0.0031102384,0.00067100825,0.00019220209,0.000012247203,0.0000018495207,0.0000045253078,0.000042106327],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99556565,0.0018255484,0.0012531488,0.00034271512,0.0008202307,0.0001927233],"domain_scores_gemma":[0.9966997,0.002264727,0.0004874341,0.0002310653,0.00023632235,0.00008075158],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0054225544,0.00017032879,0.0005089549,0.000120037155,0.00011494996,0.0000716956,0.00033496114,0.00011843865,0.00012359697],"category_scores_gemma":[0.0007161185,0.000095140604,0.00015983458,0.00025265056,0.0002268846,0.00013227914,0.00008957206,0.00037727537,0.000005790299],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024571083,0.0034948236,0.4500573,0.00003202346,0.00042945196,0.000053508327,0.0032721658,0.08652616,0.100855224,0.10219695,0.0006308977,0.2499944],"study_design_scores_gemma":[0.00032235225,0.00036994347,0.61171967,0.000008884043,0.000017671191,0.0000016386459,0.0003204623,0.00033864,0.00018299025,0.38551402,0.0011194783,0.000084226944],"about_ca_topic_score_codex":0.000016153284,"about_ca_topic_score_gemma":0.000013589965,"teacher_disagreement_score":0.2833171,"about_ca_system_score_codex":0.00009652689,"about_ca_system_score_gemma":0.000021930198,"threshold_uncertainty_score":0.38797212},"labels":[],"label_agreement":null},{"id":"W2056109249","doi":"10.1080/00016470412331294445","title":"How reliable are reliability studies of fracture classifications?A systematic review of their methodologies","year":2004,"lang":"en","type":"review","venue":"Acta Orthopaedica Scandinavica","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":184,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; McMaster University Medical Centre","funders":"","keywords":"Checklist; Medicine; Reliability (semiconductor); Kappa; Cohen's kappa; MEDLINE; Statistics; Sample size determination; Medical physics; Data mining; Computer science; Mathematics; Psychology; Cognitive psychology","score_opus":0.37890643944898705,"score_gpt":0.46675751809186394,"score_spread":0.08785107864287689,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2056109249","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000005939002,0.9886809,0.00031956926,0.004711944,0.00054663594,0.0051267645,0.00025132365,0.00006688498,0.00029002927],"genre_scores_gemma":[0.00040525675,0.99496484,0.0028641445,0.00022135492,0.000078182275,0.0007394561,0.000050897306,0.00004831462,0.0006275662],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9813682,0.0051763733,0.0065325517,0.0017253443,0.0045784335,0.0006190836],"domain_scores_gemma":[0.9618553,0.01625462,0.012301885,0.0058525833,0.0034706108,0.0002650033],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.032975588,0.0010613273,0.011643535,0.00059838267,0.0002825702,0.00014837792,0.003706552,0.0007016168,0.00015212229],"category_scores_gemma":[0.10411196,0.0005488904,0.0030173482,0.002986034,0.0012455386,0.00041819422,0.00064613455,0.0010542148,0.000027387909],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000009375584,0.0002024773,0.00006945527,0.97543913,0.0005787341,0.0000033510794,0.0001365575,0.0000025841387,0.0000010953753,0.00014832534,0.008458263,0.014950669],"study_design_scores_gemma":[0.00012559183,0.00013230187,0.000030888266,0.64296246,0.0025441323,0.000012072048,0.0010141078,4.8843526e-7,0.000007906009,0.004710937,0.34809116,0.00036794465],"about_ca_topic_score_codex":0.00001648936,"about_ca_topic_score_gemma":0.0000067125397,"teacher_disagreement_score":0.3396329,"about_ca_system_score_codex":0.0005567155,"about_ca_system_score_gemma":0.0010832062,"threshold_uncertainty_score":0.99969625},"labels":[],"label_agreement":null},{"id":"W2060814138","doi":"10.1159/000054764","title":"Statistical versus Fuzzy Measures of Variable Interaction in Patients with Stroke","year":2001,"lang":"en","type":"article","venue":"Neuroepidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University","funders":"","keywords":"Spearman's rank correlation coefficient; Statistics; Variable (mathematics); Medicine; Correlation; Rank correlation; Variables; Fuzzy logic; Mathematics; Artificial intelligence; Computer science","score_opus":0.21595298442250044,"score_gpt":0.38740726112740365,"score_spread":0.1714542767049032,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2060814138","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9827176,0.000013701161,0.009592407,0.00054762006,0.0008523488,0.00022818403,0.000020259817,0.000011985945,0.006015921],"genre_scores_gemma":[0.9974288,0.000013822938,0.002131868,0.00028564112,0.000030564377,0.000010293424,0.0000066784896,0.0000057067,0.00008663764],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9960328,0.0014038894,0.0009281962,0.00047367165,0.00086138526,0.00030005738],"domain_scores_gemma":[0.99265784,0.0061860518,0.00030325295,0.00044925875,0.00032723823,0.000076368415],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.004101341,0.00012735522,0.00045265234,0.00018893469,0.000040383453,0.000014403754,0.0003494424,0.000075411815,0.0003996746],"category_scores_gemma":[0.017716335,0.00008386597,0.000041625102,0.0003735975,0.00014871066,0.00016040784,0.00006346643,0.00025656965,0.000055665154],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002125851,0.0002532489,0.97889996,0.0000030240185,0.000012512718,0.0000034647858,0.000028008792,0.003530827,0.00012381969,0.0032490727,0.001918488,0.0098517295],"study_design_scores_gemma":[0.0018930134,0.0011980571,0.97529036,0.000011872273,0.000011609542,0.0000011403177,0.0000689463,0.0003943867,0.000028517481,0.008489926,0.012520789,0.00009135884],"about_ca_topic_score_codex":0.00016894194,"about_ca_topic_score_gemma":0.00018344398,"teacher_disagreement_score":0.014711216,"about_ca_system_score_codex":0.00005391774,"about_ca_system_score_gemma":0.00004308859,"threshold_uncertainty_score":0.99055785},"labels":[],"label_agreement":null},{"id":"W2060998495","doi":"10.1002/cjs.10103","title":"Planning and analysis of measurement reliability studies","year":2011,"lang":"en","type":"article","venue":"Canadian Journal of Statistics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":true,"ca_institutions":"University of Guelph; University of Waterloo","funders":"","keywords":"Reliability (semiconductor); Plan (archaeology); Measure (data warehouse); Mathematics; Statistics; Humanities; Computer science; Geography; Philosophy; Data mining; Physics","score_opus":0.4922490454234743,"score_gpt":0.3848306343614787,"score_spread":0.10741841106199562,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2060998495","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8692992,0.00648817,0.12027474,0.0004170615,0.0010231715,0.0002085028,0.000405343,0.0000033053088,0.0018805265],"genre_scores_gemma":[0.98115814,0.000043788372,0.018698703,0.000055553093,0.00001710346,5.510883e-7,5.2215483e-7,0.0000029810833,0.000022681135],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.996953,0.00020843839,0.0011470963,0.0001633289,0.0013456447,0.0001824849],"domain_scores_gemma":[0.99493027,0.0005321247,0.0006714135,0.00027712045,0.0032135684,0.0003755346],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009263743,0.000096823365,0.00051709346,0.00068472064,0.00010903287,0.000040880128,0.00033657244,0.000033818967,0.00025323912],"category_scores_gemma":[0.010981474,0.000067953486,0.00009855678,0.0006800096,0.0003209941,0.00011613673,0.000019501998,0.00012534077,0.0000020696737],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003912107,0.000041723823,0.9619161,0.000032557266,0.0012120304,0.00006699218,0.011618984,0.0010042328,0.00005344729,0.0027677736,0.009338873,0.011908162],"study_design_scores_gemma":[0.00023304025,0.00031258242,0.93508,0.00008079864,0.00094894937,0.000005335329,0.0057819644,0.0004673072,0.00017612334,0.05436087,0.0024284131,0.0001246488],"about_ca_topic_score_codex":0.0009710394,"about_ca_topic_score_gemma":0.008186801,"teacher_disagreement_score":0.111858934,"about_ca_system_score_codex":0.00016104229,"about_ca_system_score_gemma":0.00053318834,"threshold_uncertainty_score":0.99734944},"labels":[],"label_agreement":null},{"id":"W2061141931","doi":"10.1007/s10459-010-9222-y","title":"Likert scales, levels of measurement and the “laws” of statistics","year":2010,"lang":"en","type":"article","venue":"Advances in Health Sciences Education","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4546,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Likert scale; Statistics; Parametric statistics; Econometrics; Nonparametric statistics; Variance (accounting); Regression analysis; Sample size determination; Sample (material); Ordinal data; Statistical analysis; Computer science; Psychology; Mathematics","score_opus":0.1216247962799418,"score_gpt":0.4436029242752067,"score_spread":0.32197812799526493,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2061141931","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.901521,0.02831988,0.023624543,0.026868608,0.008683003,0.0029817815,0.00007742801,0.000018026692,0.007905695],"genre_scores_gemma":[0.97944134,0.000846199,0.019324632,0.000290829,0.00003873431,0.000024798766,3.839893e-7,0.0000018864405,0.00003118037],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9955663,0.0003928354,0.0011780198,0.00034218197,0.0023136002,0.00020708235],"domain_scores_gemma":[0.996653,0.0014468168,0.0007749992,0.0003876763,0.00066311314,0.00007439042],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0272799,0.000086979715,0.00027949014,0.00017720746,0.0002022139,0.000043562464,0.0006373961,0.000028590259,0.00006132293],"category_scores_gemma":[0.0057235346,0.00004940073,0.000028269043,0.0008494486,0.0016791997,0.00040362365,0.000055289875,0.00014455755,0.0000026583866],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000046136127,0.00029272263,0.12453406,0.00019158692,0.0000024005126,2.8900358e-8,0.0020901468,0.0004946366,0.0008065314,0.17886245,0.0006521027,0.6920272],"study_design_scores_gemma":[0.00078612234,0.000326582,0.2945156,0.00025822097,0.0000052156865,0.0000026232747,0.004815977,0.0014413854,0.001583191,0.670577,0.025526114,0.00016196101],"about_ca_topic_score_codex":0.00034322336,"about_ca_topic_score_gemma":0.0029226062,"teacher_disagreement_score":0.6918652,"about_ca_system_score_codex":0.00004405821,"about_ca_system_score_gemma":0.0009322807,"threshold_uncertainty_score":0.94547236},"labels":[],"label_agreement":null},{"id":"W2065426364","doi":"10.1159/000054768","title":"Reliability of Information Collected by Proxy in Family Studies of Alzheimer’s Disease","year":2001,"lang":"en","type":"article","venue":"Neuroepidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Institute on Aging; National Institutes of Health","keywords":"Proxy (statistics); Intraclass correlation; Medicine; Cohen's kappa; Kappa; Epidemiology; Reliability (semiconductor); Statistic; Gerontology; Demography; Clinical psychology; Statistics; Psychometrics; Internal medicine","score_opus":0.22839435769597524,"score_gpt":0.41856244089106565,"score_spread":0.1901680831950904,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2065426364","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9941697,0.0010162977,0.0003895836,0.002761235,0.000291194,0.00060285995,0.000032578148,0.000014555546,0.0007219778],"genre_scores_gemma":[0.99813265,0.000597401,0.00031126567,0.00084840535,0.0000105483805,0.000042561507,0.000007334572,0.0000030512917,0.000046815538],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9943497,0.0019658762,0.0023360134,0.00035670737,0.0007293097,0.00026237397],"domain_scores_gemma":[0.99329007,0.0043090875,0.00081324636,0.00068127806,0.0008169737,0.00008936306],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008551486,0.00014058716,0.0007505494,0.00024780078,0.000045252313,0.0000066156254,0.00045418873,0.000073513576,0.000050907034],"category_scores_gemma":[0.06380586,0.00009723251,0.00012381095,0.0009768457,0.00040886534,0.00033502106,0.00013285199,0.0001575977,0.000022110153],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00044945575,0.0003068296,0.94118965,0.000042379983,0.000023021796,0.0000022699078,0.00041668979,0.0030380448,0.0032662323,0.00041249124,0.045974724,0.0048782257],"study_design_scores_gemma":[0.0005248025,0.00028038633,0.95033956,0.000026825111,0.000023602677,7.100906e-7,0.0004278678,0.0022095344,0.00038483608,0.029106125,0.016577635,0.00009812696],"about_ca_topic_score_codex":0.00011147944,"about_ca_topic_score_gemma":0.000020162695,"teacher_disagreement_score":0.055254377,"about_ca_system_score_codex":0.00003731127,"about_ca_system_score_gemma":0.00010062923,"threshold_uncertainty_score":0.9440801},"labels":[],"label_agreement":null},{"id":"W2066581144","doi":"10.1002/sim.2546","title":"Bivariate models for co‐aggregation of dichotomous traits in twins","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Bivariate analysis; Twin study; Correlation; Dizygotic twin; Trait; Statistics; Similarity (geometry); Econometrics; Dizygotic twins; Population; Genetic correlation; Monozygotic twin; Psychology; Heritability; Mathematics; Demography; Biology; Computer science; Medicine; Genetics; Genetic variation; Artificial intelligence","score_opus":0.13561655662648225,"score_gpt":0.4075748772475037,"score_spread":0.27195832062102143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2066581144","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0881031,0.00017530676,0.9042212,0.00091677427,0.00030357583,0.00085080456,0.00030774964,0.000008954804,0.005112575],"genre_scores_gemma":[0.973384,0.000028297507,0.026034232,0.000107079926,0.00008822525,0.000038080725,0.0000407643,0.000006873981,0.00027243872],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99671274,0.00015391165,0.0012349961,0.00030926662,0.0013761822,0.00021287706],"domain_scores_gemma":[0.9971119,0.0020295207,0.00026116497,0.00023512103,0.00032304556,0.000039259867],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0059015867,0.000110227855,0.00039024235,0.0003613638,0.00003236062,0.000014377702,0.0003020841,0.0000576423,0.00013502706],"category_scores_gemma":[0.003534555,0.00007959294,0.000025640458,0.00053958583,0.00018013161,0.00009454748,0.000016353048,0.00010749289,0.000004674424],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005130135,0.0011612097,0.051407054,0.00029538808,0.000024102086,0.000037313293,0.00537659,0.10094725,0.0072144847,0.6232868,0.06809132,0.14164545],"study_design_scores_gemma":[0.0020020313,0.00024621657,0.042255007,0.00013743789,0.000009625846,5.408459e-7,0.0003220523,0.072430216,0.00031861739,0.88112575,0.0010645853,0.00008791293],"about_ca_topic_score_codex":0.0005883636,"about_ca_topic_score_gemma":0.0013530137,"teacher_disagreement_score":0.8852809,"about_ca_system_score_codex":0.00007305351,"about_ca_system_score_gemma":0.000059116064,"threshold_uncertainty_score":0.4231448},"labels":[],"label_agreement":null},{"id":"W2069378996","doi":"10.1016/j.jclinepi.2010.03.002","title":"Guidelines for Reporting Reliability and Agreement Studies (GRRAS) were proposed","year":2010,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2097,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Western University","funders":"","keywords":"Reliability (semiconductor); Psychology; Medicine; Statistics; Mathematics; Physics","score_opus":0.8672390330587612,"score_gpt":0.6777059197779776,"score_spread":0.1895331132807836,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2069378996","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.86515754,0.00081930705,0.009263499,0.11820952,0.005886674,0.00046457764,0.0000036511228,0.000008726315,0.00018651423],"genre_scores_gemma":[0.74720305,0.0011730612,0.23945084,0.007989421,0.003441096,0.000014099624,8.075982e-7,0.000012807161,0.0007148376],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95896083,0.004705852,0.03348568,0.000917647,0.0013962543,0.0005337567],"domain_scores_gemma":[0.42399958,0.44324395,0.093427196,0.003587839,0.03416727,0.001574159],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.45064372,0.00021836588,0.003410418,0.00012306408,0.00019645992,0.000036675057,0.0007325085,0.0003323937,0.000116874406],"category_scores_gemma":[0.94168174,0.00011435207,0.0010388456,0.00015202149,0.00080750836,0.00022719738,0.00022781966,0.0010628202,0.000012046918],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00041550235,0.00025348336,0.7635147,0.000054199194,0.00016140574,0.000008256494,0.00009904367,0.0002754978,0.0004237409,0.0019835006,0.0986567,0.13415396],"study_design_scores_gemma":[0.001543044,0.0017198691,0.25295988,0.00010894832,0.00010334914,0.00006973462,0.0009580551,0.0018206442,0.00009909769,0.54122025,0.19920006,0.00019706109],"about_ca_topic_score_codex":0.000008643571,"about_ca_topic_score_gemma":0.000060091228,"teacher_disagreement_score":0.5392367,"about_ca_system_score_codex":0.000028629456,"about_ca_system_score_gemma":0.00029350707,"threshold_uncertainty_score":0.5656779},"labels":[],"label_agreement":null},{"id":"W2074163347","doi":"10.1016/j.jclinepi.2007.05.010","title":"A review of five cardiology journals found that observer variability of measured variables was infrequently reported","year":2007,"lang":"en","type":"review","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":11,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Children's Hospital of Eastern Ontario; University of Ottawa","funders":"","keywords":"Medicine; Categorical variable; Continuous variable; Cohen's kappa; Kappa; Statistic; Statistics; Internal medicine; Mathematics","score_opus":0.8734374309110979,"score_gpt":0.6395238285885412,"score_spread":0.23391360232255676,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2074163347","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0001583807,0.98385376,0.007476886,0.0009777028,0.005018267,0.0011834101,0.00007293357,0.00000775968,0.0012508915],"genre_scores_gemma":[0.00012741343,0.98945516,0.0074903965,0.0015479543,0.0012205332,0.000013271163,0.000012010136,0.000033026572,0.00010020912],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.7620946,0.13426879,0.09404967,0.0022393684,0.0058751763,0.0014723996],"domain_scores_gemma":[0.4292139,0.42355308,0.12744696,0.0048698573,0.013718341,0.0011978442],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","research_integrity"],"category_scores_codex":[0.6338015,0.00087895093,0.03615671,0.0007922468,0.00009600933,0.000025922332,0.0039740126,0.0024654302,0.0010737982],"category_scores_gemma":[0.795835,0.0004940671,0.012314747,0.0015768822,0.0016660796,0.00035102598,0.0005259726,0.0040575885,0.000028896244],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00040938315,0.00088168046,0.06380491,0.05163168,0.0056545585,0.00016554711,0.000026762338,0.00017021624,0.0000014474984,0.00092402636,0.057157546,0.81917226],"study_design_scores_gemma":[0.0006606966,0.00082563824,0.008996502,0.08758862,0.0033142057,0.0006554328,0.000020721742,0.000008341111,3.8443224e-7,0.057756606,0.8398267,0.00034616774],"about_ca_topic_score_codex":0.000054239827,"about_ca_topic_score_gemma":0.0000094327415,"teacher_disagreement_score":0.8188261,"about_ca_system_score_codex":0.00029270706,"about_ca_system_score_gemma":0.0036756834,"threshold_uncertainty_score":0.99983937},"labels":[],"label_agreement":null},{"id":"W2074840499","doi":"10.1207/s15324796abm2603_02","title":"Classification and regression tree analysis in public health: Methodological review and comparison with logistic regression","year":2003,"lang":"en","type":"review","venue":"Annals of Behavioral Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":898,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Stephen's University","funders":"","keywords":"Logistic regression; Public health; Identification (biology); Population; Nonparametric statistics; Logistic model tree; Regression analysis; Health psychology; Decision tree; Statistics; Stepwise regression; Regression; Psychology; Medicine; Applied psychology; Computer science; Data mining; Mathematics; Environmental health; Pathology; Biology","score_opus":0.9360556982721563,"score_gpt":0.6560495613872027,"score_spread":0.28000613688495357,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2074840499","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0003618745,0.9890538,0.00047181814,0.008739704,0.000070348055,0.0011814748,0.000014494275,0.000011007432,0.00009547257],"genre_scores_gemma":[0.0019372442,0.9959616,0.0013391864,0.00047846435,0.000022895163,0.000072899566,0.00008441168,0.000011619453,0.00009165607],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9876019,0.005634188,0.003242534,0.0011071627,0.0020212461,0.0003930097],"domain_scores_gemma":[0.9930894,0.0020388004,0.002948306,0.0010017388,0.00055953057,0.00036222461],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027979089,0.0004877235,0.0058968267,0.0010944738,0.000109920824,0.00004510117,0.00051613373,0.0002885011,0.00019845425],"category_scores_gemma":[0.0036306265,0.00021614469,0.00025466885,0.0033386848,0.00070662075,0.00015287098,0.00011434227,0.0005625115,0.0000029488413],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000014200414,0.0002527144,0.008471751,0.004681008,0.00005774329,0.000004920176,0.00009569665,2.2336808e-7,6.982272e-7,0.0000760193,0.0043629236,0.9819821],"study_design_scores_gemma":[0.0006145511,0.0027014683,0.028501943,0.08580411,0.0047183004,0.000025289844,0.0007334967,0.000029740353,6.732441e-7,0.0006841428,0.8756337,0.00055257865],"about_ca_topic_score_codex":0.000107102984,"about_ca_topic_score_gemma":0.0001492383,"teacher_disagreement_score":0.9814295,"about_ca_system_score_codex":0.000058118796,"about_ca_system_score_gemma":0.00017173398,"threshold_uncertainty_score":0.969705},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2075431957","doi":"10.1007/s10877-008-9127-y","title":"A Measure of Confidence in Bland–Altman Analysis for the Interchangeability of Two Methods of Measurement","year":2008,"lang":"en","type":"article","venue":"Journal of Clinical Monitoring and Computing","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":40,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University Health Network","funders":"","keywords":"Interchangeability; Measure (data warehouse); Confidence interval; Limits of agreement; Medicine; Bland–Altman plot; Statistics; Medical physics; Computer science; Mathematics; Data mining; Internal medicine; Nuclear medicine","score_opus":0.6837745116735408,"score_gpt":0.587837256182952,"score_spread":0.09593725549058874,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2075431957","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.82037973,0.0010022818,0.177654,0.0002213203,0.0005770855,0.00014128025,0.0000011101143,0.0000012703247,0.000021921689],"genre_scores_gemma":[0.9768368,0.00012880174,0.022797842,0.0000071440004,0.0002229632,0.0000010421033,2.4750305e-8,0.00000276551,0.0000026244497],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9918317,0.0016035393,0.004150102,0.00024717842,0.0020079797,0.00015949157],"domain_scores_gemma":[0.9794673,0.014535529,0.0028040928,0.00039462786,0.0027116602,0.00008684712],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0847835,0.000107774875,0.001170716,0.00023814102,0.000078485114,0.000017077053,0.0006272217,0.00006513083,0.00000448544],"category_scores_gemma":[0.024859482,0.00006105094,0.00065858517,0.00071189564,0.0003074612,0.00008113493,0.00011770182,0.00032288712,7.093408e-8],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023883156,0.00023173743,0.89879,0.000040554456,0.00031071051,7.921353e-7,0.0010911993,0.005470381,0.00113298,0.00003343655,0.000008122311,0.092651226],"study_design_scores_gemma":[0.0013447846,0.00068520766,0.97322416,0.00051187456,0.00031129312,0.0000042745323,0.0015899283,0.00998935,0.008223477,0.0039762184,0.00005914558,0.0000803079],"about_ca_topic_score_codex":0.0000882181,"about_ca_topic_score_gemma":0.000019101317,"teacher_disagreement_score":0.15645705,"about_ca_system_score_codex":0.000031485037,"about_ca_system_score_gemma":0.00013980252,"threshold_uncertainty_score":0.98335457},"labels":[],"label_agreement":null},{"id":"W2077240755","doi":"10.1016/j.annemergmed.2011.06.423","title":"391 Between-Rater Reliability of Historical Variables Used in a Routine Emergency Department Patient Evaluation","year":2011,"lang":"en","type":"article","venue":"Annals of Emergency Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Medicine; Emergency department; Inter-rater reliability; Reliability (semiconductor); Medical emergency; Emergency medicine; Statistics; Nursing","score_opus":0.5818926185803324,"score_gpt":0.4632303106880315,"score_spread":0.11866230789230087,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2077240755","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98743075,0.0012113131,0.0006952647,0.0021943378,0.0025138645,0.0008814094,0.000018306977,0.00001187763,0.005042864],"genre_scores_gemma":[0.9987464,0.00047477218,0.00028064338,0.00003156525,0.00013412064,0.00006626291,0.000017281256,0.000010115639,0.00023884469],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9894651,0.001100993,0.004259087,0.000703656,0.0040664403,0.00040471955],"domain_scores_gemma":[0.9946095,0.00033286525,0.0012051157,0.0011281562,0.0025388312,0.00018553056],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.019922242,0.00026494963,0.0008803452,0.0003949937,0.000058547634,0.0000016820335,0.0006562144,0.00012749156,0.02057719],"category_scores_gemma":[0.010466832,0.00017611783,0.0002481335,0.0013691036,0.00010666888,0.0002424662,0.00012239788,0.00016590724,0.000025534693],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015401855,0.0010524994,0.9504666,0.00006135655,0.000044630517,8.6891026e-7,0.0038097545,0.00012508185,0.0013344574,0.00022883265,0.037385006,0.0053369366],"study_design_scores_gemma":[0.00097007817,0.0023271248,0.8937091,0.00015509325,0.00012851295,1.846854e-7,0.0006770993,0.00037953575,0.008904382,0.08381177,0.008643134,0.0002939391],"about_ca_topic_score_codex":0.0015823966,"about_ca_topic_score_gemma":0.0001937873,"teacher_disagreement_score":0.08358294,"about_ca_system_score_codex":0.00013610421,"about_ca_system_score_gemma":0.00010217188,"threshold_uncertainty_score":0.9978684},"labels":[],"label_agreement":null},{"id":"W2079234813","doi":"10.1111/j.0006-341x.2002.00209.x","title":"Interval Estimation for a Difference Between Intraclass Kappa Statistics","year":2002,"lang":"en","type":"article","venue":"Biometrics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":48,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Confidence interval; Cohen's kappa; Mathematics; Interval estimation; Kappa; Statistic; Sample size determination; Intraclass correlation; Reproducibility","score_opus":0.42560737613053634,"score_gpt":0.418085093842297,"score_spread":0.007522282288239368,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2079234813","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.090064935,0.000120899495,0.9074925,0.0007320164,0.00050189346,0.0003744795,0.0003080836,0.000030740506,0.00037443716],"genre_scores_gemma":[0.9084426,0.00002447549,0.08995767,0.00010142417,0.00012835297,0.00002741038,0.000022305478,0.000007736553,0.0012879974],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9968911,0.0001013549,0.00071540027,0.00039002602,0.0016428512,0.00025926644],"domain_scores_gemma":[0.99510074,0.0035713431,0.0002495965,0.00042977376,0.00052784104,0.00012070438],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0028637892,0.00013006983,0.00025797318,0.000938941,0.00013937576,0.00032182544,0.00064846186,0.00008446605,0.00035346596],"category_scores_gemma":[0.016152184,0.000095891,0.00008377169,0.0031776484,0.00008518841,0.0001359684,0.000088924186,0.00009148603,0.00035945405],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000064301457,0.00013701993,0.024434526,0.000020711748,0.000014740395,6.265117e-7,0.00014866266,0.00003575697,0.00017701059,0.0013304463,0.03159675,0.9420973],"study_design_scores_gemma":[0.0016727042,0.0013568419,0.30197936,0.00005898427,0.00011068051,0.0000022318927,0.00028805446,0.35628706,0.0017312801,0.17399158,0.16175331,0.00076790707],"about_ca_topic_score_codex":0.000008432401,"about_ca_topic_score_gemma":0.0000034857608,"teacher_disagreement_score":0.9413294,"about_ca_system_score_codex":0.000100335616,"about_ca_system_score_gemma":0.000015338604,"threshold_uncertainty_score":0.99213517},"labels":[],"label_agreement":null},{"id":"W2086743973","doi":"10.1016/j.cmpb.2011.09.011","title":"Improving the medical scale predictability by the pairwise comparisons method: Evidence from a clinical data study","year":2011,"lang":"en","type":"article","venue":"Computer Methods and Programs in Biomedicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":22,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Laurentian University; University of Sudbury","funders":"","keywords":"Predictability; Pairwise comparison; Reliability (semiconductor); Scale (ratio); Computer science; Clinical Practice; Medical diagnosis; Psychology; Machine learning; Clinical psychology; Artificial intelligence; Data mining; Medicine; Statistics; Family medicine; Mathematics","score_opus":0.7015660725220738,"score_gpt":0.562736366494007,"score_spread":0.13882970602806677,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2086743973","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19901317,0.0028435797,0.79004526,0.004965369,0.001388695,0.0016618037,0.000016533226,0.00004435506,0.000021202943],"genre_scores_gemma":[0.24368995,0.00017055236,0.7543101,0.0010483689,0.0006278991,0.00010636236,0.000017247505,0.000012232213,0.000017334132],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9760513,0.016300377,0.0024894716,0.0017783354,0.0028957727,0.0004847735],"domain_scores_gemma":[0.97386605,0.021354858,0.00048254966,0.003689768,0.00024393773,0.000362854],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.17938633,0.00028941594,0.0008363718,0.00008299777,0.0003191493,0.00024916546,0.004832724,0.00017563887,0.00021739032],"category_scores_gemma":[0.009513216,0.000120021185,0.00011168916,0.001061813,0.0016490087,0.00028117155,0.0030498235,0.0009183004,0.0000067362953],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006075152,0.0007554344,0.3858015,0.000005924553,0.000033272125,0.0000040152836,0.0022255043,1.9990628e-7,0.00001315093,0.0000025029435,0.0036815389,0.6074162],"study_design_scores_gemma":[0.0020443397,0.002515148,0.7219742,0.0003085799,0.00021382495,0.000015435562,0.007361924,0.2323458,0.000030326672,0.0048907935,0.027959967,0.00033967174],"about_ca_topic_score_codex":0.0033801263,"about_ca_topic_score_gemma":0.00063638063,"teacher_disagreement_score":0.6070765,"about_ca_system_score_codex":0.000026569172,"about_ca_system_score_gemma":0.00012225696,"threshold_uncertainty_score":0.9988301},"labels":[],"label_agreement":null},{"id":"W2088910496","doi":"10.3109/0142159x.2012.703791","title":"Generalizability theory for the perplexed: A practical introduction and guide: AMEE Guide No. 68","year":2012,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":250,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Generalizability theory; Generalization; Reliability (semiconductor); Computer science; Variance (accounting); Psychology; Management science; Cognitive psychology; Statistics; Mathematics","score_opus":0.16901637720693752,"score_gpt":0.45187119296090483,"score_spread":0.2828548157539673,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2088910496","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.37178043,0.0034998837,0.2375172,0.35631582,0.008717683,0.002736955,0.0000112078615,0.0001683677,0.019252468],"genre_scores_gemma":[0.83076644,0.00032397974,0.020609183,0.010580835,0.022460626,0.0006459841,0.000015868703,0.00005816309,0.11453891],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9942915,0.0013448166,0.0007941172,0.000506057,0.0026284123,0.00043512008],"domain_scores_gemma":[0.9950133,0.0031053757,0.00015864878,0.000925715,0.0004222743,0.00037469473],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05644593,0.00015994557,0.00025872837,0.0000386504,0.00030116882,0.00012969189,0.00048909016,0.00016783441,0.014213303],"category_scores_gemma":[0.123287946,0.00007815242,0.00014054119,0.00017699166,0.00059868884,0.00031026202,0.00020651493,0.0003672504,0.00038082985],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008474184,0.00037951526,0.03989009,0.000010388578,0.000040242896,4.4978097e-7,0.0006533791,0.0000031597992,0.00011495528,0.0057786116,0.8784801,0.0745644],"study_design_scores_gemma":[0.00034459188,0.00007146253,0.0071948967,0.000004889312,0.00004198372,0.000012282379,0.0008709682,0.0033450061,0.00005488939,0.012155992,0.9757878,0.00011524379],"about_ca_topic_score_codex":0.000037827675,"about_ca_topic_score_gemma":0.000010057186,"teacher_disagreement_score":0.45898604,"about_ca_system_score_codex":0.00008642022,"about_ca_system_score_gemma":0.000118414115,"threshold_uncertainty_score":0.98668784},"labels":[],"label_agreement":null},{"id":"W2091019309","doi":"10.1002/sim.5466","title":"Sample size formulas for estimating intraclass correlation coefficients with precision and assurance","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":467,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"","keywords":"Intraclass correlation; Confidence interval; Reliability (semiconductor); Sample size determination; Statistics; Interval (graph theory); Mathematics; Coverage probability; Limit (mathematics); Interval estimation; Sample (material); Correlation coefficient; Correlation; Computer science; Reproducibility; Power (physics); Mathematical analysis; Combinatorics; Physics","score_opus":0.0854428886232476,"score_gpt":0.3958068923931848,"score_spread":0.31036400376993717,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2091019309","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03243521,0.00010128357,0.9651323,0.00061456853,0.00073737855,0.00061404746,0.00008665168,0.000010270444,0.00026826697],"genre_scores_gemma":[0.5394728,0.000008728366,0.46020034,0.00008952562,0.00010990037,0.00002323863,0.00001157251,0.000005826205,0.000078045574],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99727046,0.000117423035,0.00066958315,0.00029017255,0.0013546677,0.00029768917],"domain_scores_gemma":[0.97717905,0.021839257,0.00025683033,0.00025805025,0.000351443,0.00011536452],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.00701403,0.00012944343,0.00029271565,0.000097374934,0.0001445583,0.00004037315,0.0001777118,0.000048677823,0.00011377545],"category_scores_gemma":[0.058951147,0.00007927259,0.000011404964,0.00031914256,0.00018170373,0.00019465249,0.00003749401,0.000128265,0.000007101997],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00049228786,0.00032291983,0.44257334,0.00016265754,0.000023073188,0.0000023071145,0.005014019,0.020674469,0.000284349,0.026756644,0.020592978,0.48310095],"study_design_scores_gemma":[0.003485309,0.00095118873,0.21504311,0.0005213569,0.000050617444,0.000006225203,0.0012084448,0.5956204,0.00005170642,0.17636253,0.0064261793,0.00027295997],"about_ca_topic_score_codex":0.00003909961,"about_ca_topic_score_gemma":0.000076827804,"teacher_disagreement_score":0.5749459,"about_ca_system_score_codex":0.000064033564,"about_ca_system_score_gemma":0.000031253658,"threshold_uncertainty_score":0.9489757},"labels":[],"label_agreement":null},{"id":"W2091307601","doi":"10.1111/j.1743-498x.2012.00607.x","title":"A method for identifying extreme OSCE examiners","year":2013,"lang":"en","type":"article","venue":"The Clinical Teacher","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Medical Council of Canada","funders":"","keywords":"Psychology; Cohort; Applied psychology; Standard deviation; Quality assurance; Inter-rater reliability; Objective structured clinical examination; Statistics; Computer science; Medicine; Rating scale; Mathematics; Developmental psychology","score_opus":0.753723718068258,"score_gpt":0.5667969616218294,"score_spread":0.18692675644642864,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2091307601","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.43325454,0.000288938,0.49854675,0.04424181,0.0020919023,0.00280295,0.0000050058766,0.00014547352,0.018622618],"genre_scores_gemma":[0.7491208,0.000029785133,0.12914832,0.0057728314,0.0014842956,0.00049110345,0.0000028264533,0.000037859514,0.113912225],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9948904,0.0016525631,0.0013523389,0.00059077895,0.0011878005,0.0003261382],"domain_scores_gemma":[0.9864279,0.0115113575,0.00033701232,0.0012006748,0.00037341708,0.00014961879],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.042268768,0.00014334884,0.00035949922,0.000044520715,0.00020727332,0.00032456417,0.001379488,0.00011877711,0.003085907],"category_scores_gemma":[0.01741883,0.000070311566,0.00042204568,0.0002642553,0.00019373001,0.00022309914,0.00020389918,0.00033401058,0.001920563],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000039661034,0.00037858274,0.02179337,0.000007106759,0.00006579011,4.3324567e-7,0.0005409369,0.000041883457,0.001271131,0.00095995585,0.3262363,0.64866483],"study_design_scores_gemma":[0.0012477572,0.00025059705,0.1578628,0.000032619213,0.00008037842,0.0000017708688,0.0023227006,0.014064924,0.00018767841,0.37591714,0.44768953,0.0003421009],"about_ca_topic_score_codex":0.000084668216,"about_ca_topic_score_gemma":0.00001738311,"teacher_disagreement_score":0.64832276,"about_ca_system_score_codex":0.00003009092,"about_ca_system_score_gemma":0.00003772812,"threshold_uncertainty_score":0.99885654},"labels":[],"label_agreement":null},{"id":"W2095752327","doi":"10.1002/(sici)1097-0258(20000315)19:5<723::aid-sim379>3.0.co;2-a","title":"Interval estimation for Cohen's kappa as a measure of agreement","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":217,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Confidence interval; Statistics; Mathematics; Measure (data warehouse); Kappa; Statistic; Variance (accounting); Delta method; Coverage probability; Asymptotic analysis; Computation; Asymptotic distribution; Cohen's kappa; Applied mathematics; Computer science; Algorithm","score_opus":0.11948518704376823,"score_gpt":0.4260948892710158,"score_spread":0.3066097022272476,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2095752327","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09164632,0.00033246726,0.87748086,0.0059719835,0.0009288197,0.0017594662,0.00026011167,0.000020903772,0.021599073],"genre_scores_gemma":[0.934467,0.000054860157,0.062390182,0.0004828379,0.0001126822,0.00006699333,0.00004351584,0.00000935096,0.0023725657],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9958637,0.00018771716,0.0012035646,0.0003503025,0.0021770194,0.00021769441],"domain_scores_gemma":[0.9970811,0.001633237,0.00023951916,0.00044417116,0.0005190375,0.00008291486],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.007252351,0.00013812908,0.00043697792,0.00018062303,0.00005625889,0.0000204042,0.00043934936,0.000052849133,0.006227912],"category_scores_gemma":[0.008945916,0.000094976735,0.00004560084,0.00037748177,0.00024921555,0.000073867865,0.000025884367,0.000116392934,0.00011006977],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004506093,0.0003054597,0.0033102415,0.000110227964,0.000039016802,0.00000870051,0.0032759362,0.004834862,0.00050358963,0.027285425,0.09017988,0.869696],"study_design_scores_gemma":[0.0042837616,0.002271252,0.015318377,0.0006632088,0.00008327864,0.000004389221,0.0020945803,0.08427042,0.00056159747,0.8536577,0.03650804,0.00028335804],"about_ca_topic_score_codex":0.00014507158,"about_ca_topic_score_gemma":0.0001421121,"teacher_disagreement_score":0.8694127,"about_ca_system_score_codex":0.00007963402,"about_ca_system_score_gemma":0.00007118362,"threshold_uncertainty_score":0.99940217},"labels":[],"label_agreement":null},{"id":"W2096649070","doi":"10.1002/sim.3387","title":"Bivariate modeling of interobserver agreement coefficients","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Bivariate analysis; Anxiety; Outcome (game theory); Trait; Psychology; Statistics; Depression (economics); Clinical psychology; Medicine; Psychiatry; Mathematics; Computer science","score_opus":0.2636804934739743,"score_gpt":0.41067999707973896,"score_spread":0.14699950360576464,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2096649070","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2562768,0.00014396648,0.7381275,0.00053375424,0.00079823186,0.00026093132,0.00004384974,0.000007953743,0.003807012],"genre_scores_gemma":[0.98481953,0.00010700493,0.014295311,0.00020606845,0.000055735276,0.000006116722,0.000010557031,0.0000060429,0.0004936606],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9951486,0.00022310561,0.0014185057,0.0003845095,0.0025781605,0.0002471381],"domain_scores_gemma":[0.99780047,0.00072645344,0.00025080977,0.00056390726,0.0005689121,0.00008946227],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0053847698,0.0001389238,0.00044916346,0.00029184166,0.00007451432,0.000008749466,0.00059224485,0.000044746106,0.0012153013],"category_scores_gemma":[0.005526881,0.00009320396,0.00003546298,0.0006102101,0.00030098358,0.00006446388,0.00012241516,0.00016654002,0.00007196934],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005473346,0.0021188394,0.28653452,0.00023853712,0.0001483354,0.00029726865,0.026290126,0.459126,0.0035042919,0.059830327,0.112770505,0.04859392],"study_design_scores_gemma":[0.0036221629,0.0008715349,0.04007524,0.00049408793,0.000036468948,0.000008516381,0.0024116375,0.80993825,0.0001914076,0.13907543,0.0029443402,0.00033094067],"about_ca_topic_score_codex":0.00030912846,"about_ca_topic_score_gemma":0.00010183453,"teacher_disagreement_score":0.7285427,"about_ca_system_score_codex":0.00007322352,"about_ca_system_score_gemma":0.00005767747,"threshold_uncertainty_score":0.99969774},"labels":[],"label_agreement":null},{"id":"W2097681968","doi":"10.1093/biostatistics/2.3.323","title":"Efficiency considerations in the analysis of inter-observer agreement","year":2001,"lang":"en","type":"article","venue":"Biostatistics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph","funders":"","keywords":"Cohen's kappa; Replicate; Statistics; Inter-rater reliability; Reliability (semiconductor); Statistic; Kappa; Mathematics; Binary number; Computer science; Binomial distribution; Econometrics; Arithmetic; Physics","score_opus":0.23113506979270315,"score_gpt":0.40521817031104557,"score_spread":0.17408310051834242,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2097681968","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.82730573,0.0001551022,0.15947591,0.0047084456,0.00035604378,0.00052876445,0.00019989398,0.000009547115,0.007260562],"genre_scores_gemma":[0.99684566,0.000020638226,0.0024713185,0.00047356015,0.0000123585205,0.000007414286,0.000009034284,0.0000016518524,0.00015838661],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99717414,0.00036686196,0.0008408043,0.0002481063,0.0012138607,0.00015622073],"domain_scores_gemma":[0.9964512,0.00238932,0.00020572735,0.0005977027,0.00032414965,0.000031920565],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.004319277,0.00008682029,0.00023774954,0.00032961622,0.00008877992,0.000105676765,0.00042311038,0.000029459794,0.0015348521],"category_scores_gemma":[0.0040573995,0.000050311533,0.00010438814,0.0019602573,0.00013317085,0.00005766561,0.00005186372,0.00007584735,0.00005976952],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000029891105,0.0013374675,0.8517873,0.000011358577,0.00028521413,0.000053213676,0.0062540094,0.01117279,0.00062281,0.07529371,0.032562364,0.020589838],"study_design_scores_gemma":[0.000325485,0.00013651798,0.93634456,0.00001395574,0.00026359851,0.0000019763793,0.0034783967,0.0154708335,0.00011236907,0.035753552,0.007955147,0.00014360766],"about_ca_topic_score_codex":0.00012718576,"about_ca_topic_score_gemma":0.0015706324,"teacher_disagreement_score":0.1695399,"about_ca_system_score_codex":0.000033956716,"about_ca_system_score_gemma":0.000043356216,"threshold_uncertainty_score":0.9993779},"labels":[],"label_agreement":null},{"id":"W2099077995","doi":"10.1177/0962280215601133","title":"Assessing agreement between two measurement systems: An alternative to the limits of agreement approach","year":2015,"lang":"en","type":"article","venue":"Statistical Methods in Medical Research","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":55,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Interchangeability; Agreement; Metric (unit); Limits of agreement; Computer science; Units of measurement; Physics; Engineering","score_opus":0.8455312682296875,"score_gpt":0.6788734831139276,"score_spread":0.1666577851157599,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2099077995","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011491402,0.00032868056,0.96496737,0.0049651666,0.00069213734,0.0018865867,0.00003972745,0.000012590646,0.015616356],"genre_scores_gemma":[0.6608213,0.000015546693,0.33805937,0.00018720844,0.00046641406,0.00031499847,0.000007411359,0.000017214292,0.00011051873],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.90176743,0.04603641,0.0034198512,0.001767464,0.045421597,0.0015872299],"domain_scores_gemma":[0.96619326,0.024394171,0.00033659494,0.0018917132,0.005061493,0.0021227675],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3749643,0.00030738555,0.0009721339,0.0006872685,0.00027325907,0.000609609,0.0035681117,0.00017360438,0.0006258779],"category_scores_gemma":[0.2304376,0.00017501293,0.00009620861,0.0024496578,0.0009993035,0.00031551314,0.0009649156,0.001505947,0.00017208036],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016506291,0.001129789,0.0073038535,0.00009439813,0.00009324845,0.00003767438,0.003018375,0.002225057,0.0002898197,0.035049196,0.0075629684,0.94303054],"study_design_scores_gemma":[0.00698043,0.006074854,0.089565635,0.0018063405,0.00013658123,0.000015175086,0.081105605,0.23762509,0.0033789997,0.49072278,0.081134416,0.0014540886],"about_ca_topic_score_codex":0.001354066,"about_ca_topic_score_gemma":0.0002028967,"teacher_disagreement_score":0.9415765,"about_ca_system_score_codex":0.0010024825,"about_ca_system_score_gemma":0.0013651608,"threshold_uncertainty_score":0.7760447},"labels":[],"label_agreement":null},{"id":"W2099954138","doi":"10.1016/j.ygyno.2005.07.040","title":"Kappa statistics to measure interrater and intrarater agreement for 1790 cervical biopsy specimens among twelve pathologists: Qualitative histopathologic analysis and methodologic issues","year":2005,"lang":"en","type":"article","venue":"Gynecologic Oncology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":90,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Cancer Institute","keywords":"Medicine; Biopsy; Inter-rater reliability; Cohen's kappa; Cervical cancer; Radiology; Kappa; Family medicine; Cancer; Statistics; Internal medicine","score_opus":0.26558037577816274,"score_gpt":0.4873603395796752,"score_spread":0.22177996380151244,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2099954138","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.80787635,0.00060708687,0.1714945,0.015575338,0.00060803816,0.0018162534,0.00025111937,0.00006817509,0.0017031332],"genre_scores_gemma":[0.7277134,0.00009657118,0.26701748,0.0028413984,0.0001637549,0.00042142824,0.000033775174,0.000012922239,0.0016992679],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9901624,0.0045703463,0.0017416966,0.0017557071,0.00091488473,0.00085497013],"domain_scores_gemma":[0.98894143,0.008439044,0.0006521557,0.00073586253,0.00085301115,0.00037848958],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.021674277,0.0005250087,0.0016039616,0.0006906674,0.0003879808,0.00023710234,0.0007766491,0.00052692165,0.0024286439],"category_scores_gemma":[0.012183192,0.00034455772,0.00023583593,0.00089731085,0.0011065946,0.00024002348,0.0005930126,0.000406082,0.00019806065],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0025477973,0.0035297659,0.3018108,0.0001602781,0.0020384856,0.00048300717,0.08651789,0.0019509419,0.017716704,0.016796323,0.08312747,0.48332053],"study_design_scores_gemma":[0.0026249469,0.009718316,0.72090083,0.000022904038,0.0013036733,0.000058716076,0.031557087,0.001814052,0.0010780765,0.13137174,0.09815412,0.0013955601],"about_ca_topic_score_codex":0.000055957084,"about_ca_topic_score_gemma":0.0018151705,"teacher_disagreement_score":0.48192498,"about_ca_system_score_codex":0.00058206933,"about_ca_system_score_gemma":0.000085264975,"threshold_uncertainty_score":0.99990064},"labels":[],"label_agreement":null},{"id":"W2102888726","doi":"10.1002/cjs.5550360306","title":"On population‐based measures of agreement for binary classifications","year":2008,"lang":"en","type":"article","venue":"Canadian Journal of Statistics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"","keywords":"Cohen's kappa; Kappa; Statistic; Statistics; Population; Binary number; Inference; Binary data; Contrast (vision); Mathematics; Agreement; Econometrics; Computer science; Medicine; Artificial intelligence; Linguistics","score_opus":0.3749121106352241,"score_gpt":0.3604218302652674,"score_spread":0.014490280369956698,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2102888726","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.33911937,0.00026389345,0.65055,0.0047665085,0.0015182359,0.0006408337,0.0018974929,0.0000040609925,0.0012396179],"genre_scores_gemma":[0.9739378,0.00000792109,0.025539733,0.00021377734,0.000052824904,0.00000361305,0.00001284587,0.0000055125342,0.00022596154],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99742854,0.00012080262,0.00097636826,0.00013156602,0.0011662669,0.00017644896],"domain_scores_gemma":[0.9958609,0.0011980281,0.0006276628,0.0002900095,0.0016868047,0.0003366084],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0020427625,0.0000887048,0.00024114826,0.0004667626,0.00026435909,0.000035721252,0.0004119518,0.00004063152,0.00039810938],"category_scores_gemma":[0.005721161,0.0000681789,0.000106208616,0.00028151882,0.00012037516,0.00007447111,0.0000038313333,0.00009967521,0.000013695218],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002018499,0.00034476278,0.22422504,0.000052585103,0.00011670151,0.0000734838,0.0009949384,0.05493079,0.00060408877,0.07178181,0.61105376,0.035620168],"study_design_scores_gemma":[0.0014275791,0.0015401306,0.81664777,0.00013655318,0.00007682503,0.000016785185,0.000472108,0.0049474686,0.00047256312,0.12486916,0.049127825,0.00026522018],"about_ca_topic_score_codex":0.0005723071,"about_ca_topic_score_gemma":0.004787682,"teacher_disagreement_score":0.63481843,"about_ca_system_score_codex":0.00019190615,"about_ca_system_score_gemma":0.0013516172,"threshold_uncertainty_score":0.68491775},"labels":[],"label_agreement":null},{"id":"W2105852486","doi":"10.1186/1472-6963-8-277","title":"Reliability of the interRAI suite of assessment instruments: a 12-country study of an integrated health information system","year":2008,"lang":"en","type":"article","venue":"BMC Health Services Research","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":677,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Homewood Research Institute; University of Waterloo","funders":"University of Waterloo; Curtin University of Technology","keywords":"Health informatics; Reliability (semiconductor); Medicine; Nursing research; Cohen's kappa; Kappa; Suite; Health care; Health administration; Nursing; Family medicine; Public health; Statistics","score_opus":0.19404930026780062,"score_gpt":0.47833794597111684,"score_spread":0.28428864570331625,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105852486","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9953597,0.00003887503,0.00024264486,0.00021465778,0.0002855801,0.0033408385,0.00009046346,0.000017728607,0.00040949826],"genre_scores_gemma":[0.9988275,0.000050204548,0.00082853425,0.00015789113,0.000018304288,0.00007439555,0.000015451455,0.0000069009325,0.00002083],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9793022,0.0074656014,0.0038844834,0.00051111606,0.008233674,0.000602919],"domain_scores_gemma":[0.9918459,0.00068544724,0.0018248986,0.0020844606,0.0032957105,0.0002635986],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0445347,0.00018308207,0.0007992351,0.00079417654,0.00054222,0.00006614435,0.0022084643,0.000091205824,0.000054700522],"category_scores_gemma":[0.00060477486,0.000106522006,0.00010800666,0.0035135413,0.00030953594,0.0009829114,0.0005730197,0.00054375693,0.000008744917],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000383511,0.002080101,0.9487247,0.00464759,0.00002404433,1.2540458e-7,0.035344154,0.0009892086,0.00004543488,0.00012489395,0.00016087486,0.007475389],"study_design_scores_gemma":[0.0010365729,0.0026026596,0.82907146,0.0006726564,0.0000026861399,0.0000018768336,0.15585972,0.009736136,0.00010323682,0.000104893035,0.0007397592,0.000068330155],"about_ca_topic_score_codex":0.02217947,"about_ca_topic_score_gemma":0.0064192098,"teacher_disagreement_score":0.12051557,"about_ca_system_score_codex":0.0007132391,"about_ca_system_score_gemma":0.0027863367,"threshold_uncertainty_score":0.9843319},"labels":[],"label_agreement":null},{"id":"W2106963718","doi":"10.1093/annhyg/mel021","title":"Attenuation in Risk Estimates in Logistic and Cox Proportional-Hazards Models due to Group-Based Exposure Assessment Strategy","year":2006,"lang":"en","type":"article","venue":"The Annals of Occupational Hygiene","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Provincial Laboratory of Public Health; University of Alberta","funders":"","keywords":"Statistics; Proportional hazards model; Logistic regression; Mathematics; Attenuation; Variance (accounting); Regression analysis; Econometrics; Medicine; Physics","score_opus":0.2942759238130756,"score_gpt":0.44869054056029417,"score_spread":0.15441461674721857,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2106963718","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9458406,0.00013233698,0.050414823,0.0025311932,0.000044026754,0.0005251262,0.00008156743,0.000008164708,0.0004221947],"genre_scores_gemma":[0.9961887,0.000009915782,0.003403616,0.00020315162,0.00003057302,0.000078186924,0.00004951321,0.000004994915,0.000031362255],"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","domain_scores_codex":[0.99630433,0.00034529675,0.0009927483,0.0003797748,0.0017642999,0.00021355295],"domain_scores_gemma":[0.99749947,0.0011045259,0.00036694467,0.00034237446,0.0006387038,0.000047998165],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007291771,0.0001441499,0.00027279745,0.000305357,0.00009996953,0.000077474135,0.00031673044,0.000054048305,0.000064947075],"category_scores_gemma":[0.00100091,0.000094586976,0.000057459114,0.0006587013,0.00013512318,0.00026959972,0.000057307912,0.00012460095,0.000009603967],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019838882,0.00034498968,0.3222839,0.000009937845,0.000005109896,0.0000028604063,0.00004849227,0.65522146,0.00022072783,0.01500594,0.00044491648,0.0062132594],"study_design_scores_gemma":[0.00029679047,0.00015361662,0.676596,0.000032101307,0.000004322069,5.821151e-7,0.000064417116,0.1372115,0.00020948699,0.1853329,0.000022087499,0.000076219294],"about_ca_topic_score_codex":0.0008357479,"about_ca_topic_score_gemma":0.0009691758,"teacher_disagreement_score":0.51800996,"about_ca_system_score_codex":0.000050579743,"about_ca_system_score_gemma":0.0002106938,"threshold_uncertainty_score":0.3857145},"labels":[],"label_agreement":null},{"id":"W2107006322","doi":"10.1370/afm.1195","title":"Methods to Achieve High Interrater Reliability in Data Collection From Primary Care Medical Records","year":2011,"lang":"en","type":"article","venue":"The Annals of Family Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":83,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Élisabeth Bruyère Hospital; Bruyère; University of Ottawa","funders":"","keywords":"Medicine; Inter-rater reliability; Reliability (semiconductor); Quality assurance; Chart; Confidence interval; Audit; Statistics; Sample size determination; Data collection; Control chart; Quality (philosophy); Sample (material); Quality management; Statistic; Medical physics; Operations management; Computer science; Mathematics","score_opus":0.7358387871238369,"score_gpt":0.5387267113294011,"score_spread":0.1971120757944358,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2107006322","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9377157,0.0010277014,0.013613968,0.033479825,0.0020002946,0.0006954911,0.000052538115,0.000024505567,0.011389957],"genre_scores_gemma":[0.9692243,0.00045193738,0.018077407,0.011384026,0.00034016446,0.000035378118,0.00007191067,0.000014531071,0.0004003511],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9923752,0.0023903798,0.0015006495,0.0007730561,0.0026701982,0.00029052325],"domain_scores_gemma":[0.993553,0.0025282537,0.00029488432,0.002709132,0.00070491043,0.00020981854],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03377779,0.00019199129,0.00066731806,0.00023339802,0.0000861706,0.000016229722,0.0030484684,0.00013317169,0.0014601838],"category_scores_gemma":[0.013763852,0.0000971556,0.0000697654,0.0010583475,0.00042131852,0.00022455242,0.000875101,0.0003745536,0.00004197165],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0040252176,0.00063675607,0.044233516,0.00007976044,0.00010850087,0.000015155974,0.040144246,0.0000318631,0.005974001,0.000115238516,0.3169057,0.58773005],"study_design_scores_gemma":[0.0012085895,0.0015849101,0.8841581,0.0006690299,0.0000490107,0.0000014195007,0.012288588,0.000493831,0.0028081497,0.063365884,0.033113524,0.0002590028],"about_ca_topic_score_codex":0.009034255,"about_ca_topic_score_gemma":0.0010824329,"teacher_disagreement_score":0.8399245,"about_ca_system_score_codex":0.00006376385,"about_ca_system_score_gemma":0.00018303939,"threshold_uncertainty_score":0.9994526},"labels":[],"label_agreement":null},{"id":"W2111817119","doi":"10.1016/j.jpainsymman.2009.06.013","title":"Evaluating Correlation and Interrater Reliability for Four Performance Scales in the Palliative Care Setting","year":2010,"lang":"en","type":"article","venue":"Journal of Pain and Symptom Management","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":56,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; Sunnybrook Health Science Centre","funders":"","keywords":"Inter-rater reliability; Correlation; Medicine; Scale (ratio); Reliability (semiconductor); Kappa; Palliative care; Population; Statistics; Spearman's rank correlation coefficient; Intraclass correlation; Psychometrics; Clinical psychology; Rating scale; Nursing; Mathematics","score_opus":0.11947524826273714,"score_gpt":0.3982255808639538,"score_spread":0.2787503326012167,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2111817119","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98822534,0.00009389439,0.008151303,0.002227382,0.00030751235,0.0005040256,0.0000010795676,0.0000022739105,0.00048721407],"genre_scores_gemma":[0.9928301,0.0000492477,0.0066740005,0.00030352676,0.00007651515,0.00001661538,5.649089e-7,0.0000030489077,0.000046402638],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9974122,0.00056335237,0.00074552797,0.00020942124,0.0009262624,0.00014322154],"domain_scores_gemma":[0.99744076,0.00155321,0.00040365773,0.0002014167,0.00036242965,0.000038522212],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03358426,0.000098920005,0.00017253868,0.00013163233,0.00018872325,0.00021908725,0.00028781442,0.00003965884,0.000014015853],"category_scores_gemma":[0.0019651172,0.000053370895,0.00006521247,0.00013360534,0.00007116329,0.0002376993,0.000072254756,0.00023751252,0.0000010779131],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000100280326,0.00005532001,0.6799285,0.0001609719,0.00002314671,0.0000022318761,0.010096424,0.0018240281,0.00015011698,0.00045273788,0.0008865212,0.3063197],"study_design_scores_gemma":[0.000842412,0.00052169967,0.8873509,0.00016978575,0.000030219377,0.0000027355832,0.018856276,0.079139866,0.000019960813,0.010729684,0.0022376291,0.00009883405],"about_ca_topic_score_codex":0.0000020741238,"about_ca_topic_score_gemma":0.000047021444,"teacher_disagreement_score":0.3062209,"about_ca_system_score_codex":0.00003674344,"about_ca_system_score_gemma":0.000013285954,"threshold_uncertainty_score":0.9951284},"labels":[],"label_agreement":null},{"id":"W2113321769","doi":"10.1136/eb-2013-101243","title":"The two-step Fagan's nomogram: ad hoc interpretation of a diagnostic test result without calculation","year":2013,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":136,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Prince Edward Island","funders":"","keywords":"Nomogram; Fagan inspection; Computer science; Test (biology); Medicine; Software engineering; Geology","score_opus":0.11527103256885256,"score_gpt":0.3861066713332384,"score_spread":0.2708356387643858,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2113321769","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7494009,0.011811107,0.10407061,0.1273637,0.0019633612,0.0038095196,0.000010703839,0.00010798088,0.001462136],"genre_scores_gemma":[0.99691755,0.0003676884,0.0009826577,0.0008622043,0.00019211535,0.00014134099,0.000005971085,0.000009324641,0.0005211396],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99414116,0.000623463,0.001401503,0.00049337535,0.0030421857,0.0002982831],"domain_scores_gemma":[0.9470395,0.04936058,0.0007290411,0.0010523929,0.0016443896,0.00017408871],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009809416,0.00020454408,0.00039324578,0.0001851621,0.00023216716,0.00013033957,0.0007973243,0.00006535591,0.0004274227],"category_scores_gemma":[0.19284351,0.00010247157,0.000118707634,0.00087188487,0.0006280133,0.00052319956,0.000062140614,0.00020672417,0.00025288714],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00033833028,0.00022261465,0.16421555,0.00007757479,0.000033191092,0.000004540905,0.0017859722,0.0031792317,0.009445188,0.00017502696,0.035462804,0.78506],"study_design_scores_gemma":[0.005642888,0.00479252,0.61651516,0.01038941,0.00029304583,0.0000066611233,0.003598252,0.30220664,0.004936134,0.014059868,0.036869287,0.0006901431],"about_ca_topic_score_codex":0.0004606884,"about_ca_topic_score_gemma":0.0003156608,"teacher_disagreement_score":0.7843698,"about_ca_system_score_codex":0.000108082575,"about_ca_system_score_gemma":0.00015925545,"threshold_uncertainty_score":0.8139555},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2116873208","doi":"10.1071/nb12116","title":"Are pregnancy outcomes associated with risk factor reporting in routinely collected perinatal data?","year":2013,"lang":"en","type":"article","venue":"New South Wales Public Health Bulletin","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Aging","funders":"","keywords":"Medicine; Pregnancy; Medical record; Risk factor; Obstetrics; Kappa; Internal medicine","score_opus":0.37495525527933143,"score_gpt":0.39076785888721083,"score_spread":0.015812603607879394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2116873208","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9353535,0.00030425465,0.0007340864,0.060226727,0.00028708225,0.0017881104,0.00043248452,0.00018057055,0.00069319265],"genre_scores_gemma":[0.99162024,0.0000139735985,0.0015703836,0.0014948708,0.000054974826,0.000057845755,0.000071721915,0.000032630738,0.005083355],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9887494,0.0015473507,0.003922525,0.0014666298,0.0030497462,0.0012643778],"domain_scores_gemma":[0.9871626,0.0015699257,0.007521432,0.0020614604,0.0008792091,0.00080535765],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.011576726,0.00041029893,0.0011152608,0.0003995446,0.00051654776,0.0009080827,0.001752055,0.00019467475,0.004354607],"category_scores_gemma":[0.10646274,0.00027126455,0.0001289657,0.0016071725,0.00012773271,0.0004401563,0.00050513947,0.00073013594,0.0006547214],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000009178744,0.00018803331,0.9527,0.000024400873,0.00003532548,0.000013502091,0.0028044132,0.000028402597,9.081937e-7,0.000012149042,0.022014491,0.022169163],"study_design_scores_gemma":[0.0013709463,0.00013658678,0.9617214,0.00030420336,0.0000046179302,0.0000022574789,0.0041796262,0.00036372573,0.0000012693984,0.00018046296,0.03143156,0.00030331477],"about_ca_topic_score_codex":0.0040235403,"about_ca_topic_score_gemma":0.0035495872,"teacher_disagreement_score":0.09488602,"about_ca_system_score_codex":0.00045451676,"about_ca_system_score_gemma":0.0013782618,"threshold_uncertainty_score":0.99997395},"labels":[],"label_agreement":null},{"id":"W2122573908","doi":"10.1177/0013164415574086","title":"A Ratio Test of Interrater Agreement With High Specificity","year":2015,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université du Québec à Trois-Rivières; University of Ottawa","funders":"","keywords":"Inter-rater reliability; Agreement; Psychology; Validation test; Test validity; Test (biology); Psychometrics; Statistics; Clinical psychology; Developmental psychology; Rating scale; Mathematics; Geology","score_opus":0.4746679401283067,"score_gpt":0.40888115156470517,"score_spread":0.06578678856360154,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2122573908","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9590281,0.00026492556,0.0029383341,0.020894632,0.0007964507,0.0004871495,0.000013889053,0.0000118836,0.015564583],"genre_scores_gemma":[0.996219,0.00000853102,0.0022677628,0.00049437454,0.00019155044,0.000052131207,0.000004024907,0.0000037447808,0.00075892784],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9951897,0.00023349216,0.0006947219,0.0005495645,0.003134748,0.00019773492],"domain_scores_gemma":[0.9972599,0.00034613925,0.00024184467,0.0004372252,0.0014736382,0.00024122339],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0052833296,0.00017015386,0.00026969425,0.00009512017,0.000083143656,0.00009737622,0.00038083596,0.00005455595,0.0017031877],"category_scores_gemma":[0.0019463446,0.000092129885,0.000052544536,0.00031576006,0.00023127042,0.0001226023,0.000055589688,0.00012587327,0.00015645582],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013313801,0.01477403,0.6793236,0.000044786415,0.0001753233,0.0000056489034,0.003288074,0.00034459258,0.017989663,0.03277455,0.21578151,0.03416684],"study_design_scores_gemma":[0.0012008036,0.0020276483,0.8835647,0.00006686495,0.000021134061,0.000011518307,0.0011293471,0.000023446295,0.0017933976,0.09393515,0.015975077,0.00025091038],"about_ca_topic_score_codex":0.000028590961,"about_ca_topic_score_gemma":0.000033550918,"teacher_disagreement_score":0.2042411,"about_ca_system_score_codex":0.00010234507,"about_ca_system_score_gemma":0.000108766566,"threshold_uncertainty_score":0.9992094},"labels":[],"label_agreement":null},{"id":"W2123033778","doi":"10.1177/0962280210383082","title":"Closed-form confidence intervals for functions of the normal mean and standard deviation","year":2010,"lang":"en","type":"article","venue":"Statistical Methods in Medical Research","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":172,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"Ontario Ministry of Research and Innovation; Natural Sciences and Engineering Research Council of Canada","keywords":"Standard deviation; Confidence interval; Percentile; Mathematics; Statistics; Coefficient of variation; Robust confidence intervals; Normal distribution; Geometric standard deviation; Standard error; CDF-based nonparametric confidence interval; Variance (accounting); Mean difference; Range (aeronautics); Limits of agreement; Relative standard deviation; Medicine; Detection limit; Nuclear medicine","score_opus":0.35473691480919106,"score_gpt":0.6128973161725231,"score_spread":0.258160401363332,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2123033778","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.052668586,0.000040515486,0.93829817,0.005628624,0.00064934476,0.0006895743,0.00009816025,0.000005139726,0.0019218824],"genre_scores_gemma":[0.6721897,0.0000249475,0.3271143,0.00014894128,0.000095799274,0.0001044804,0.0000021368103,0.000006607864,0.00031307407],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9890402,0.003324019,0.0010477146,0.0004582904,0.0056920745,0.00043772138],"domain_scores_gemma":[0.9306525,0.06664287,0.00013728043,0.00066605146,0.0015385784,0.00036273804],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.11007071,0.00008988532,0.00031068604,0.00018262716,0.00023344407,0.00009938571,0.0008777013,0.00015703536,0.0028721017],"category_scores_gemma":[0.34850413,0.00004952302,0.000063534746,0.0007182384,0.0017101875,0.00010201585,0.00037061874,0.00111577,0.000010059624],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017618667,0.00007759803,0.0054638404,0.00004866521,0.000008367719,0.0000012322457,0.00040470657,0.0000018255514,0.0020888152,0.12643728,0.002488403,0.8628031],"study_design_scores_gemma":[0.0006617922,0.00041057877,0.066613436,0.00009610029,0.00000979521,0.0000023256164,0.0009008228,0.010950559,0.003013095,0.89752686,0.019725276,0.00008934994],"about_ca_topic_score_codex":0.00011278957,"about_ca_topic_score_gemma":0.0010551981,"teacher_disagreement_score":0.86271375,"about_ca_system_score_codex":0.00004761361,"about_ca_system_score_gemma":0.00048759108,"threshold_uncertainty_score":0.9980394},"labels":[],"label_agreement":null},{"id":"W2124853083","doi":"10.1080/19338240903348220","title":"Bayesian Method for Improving Logistic Regression Estimates under Group-Based Exposure Assessment with Additive Measurement Errors","year":2009,"lang":"en","type":"article","venue":"Archives of Environmental & Occupational Health","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta; University of Calgary","funders":"Canadian Institutes of Health Research","keywords":"Statistics; Bayesian probability; Logistic regression; Variance (accounting); Mathematics; Econometrics; Observational error; Odds ratio","score_opus":0.11740485703262345,"score_gpt":0.4164750196733869,"score_spread":0.29907016264076347,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2124853083","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017750045,0.00017845073,0.97790325,0.002239865,0.00008370601,0.001304318,0.00032314862,0.000032238586,0.00018499896],"genre_scores_gemma":[0.6865998,0.000008680196,0.31244007,0.00071124756,0.000030376228,0.000062599494,0.00012119854,0.00001316497,0.000012855569],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9943653,0.0005096902,0.0010298769,0.0007763172,0.0028493814,0.00046945197],"domain_scores_gemma":[0.99640024,0.0019438847,0.00085155555,0.00050897215,0.000058473637,0.00023686567],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0037887986,0.00030664454,0.00047450277,0.0002145573,0.0003970766,0.000048449736,0.00043768497,0.00005103564,0.00011301133],"category_scores_gemma":[0.00032047817,0.00020694494,0.00018831175,0.00014981345,0.00025213644,0.00019604033,0.00005879985,0.00017338018,0.0000030104256],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0038537353,0.004896588,0.10288632,0.00023030203,0.00013763054,0.000004381809,0.0009725141,0.054375462,0.036801226,0.009618643,0.0006374828,0.7855857],"study_design_scores_gemma":[0.0017308695,0.0039203726,0.90530634,0.00020302695,0.000033297376,0.0000020544076,0.00077291374,0.044165228,0.0015805652,0.04181594,0.00020584324,0.00026352663],"about_ca_topic_score_codex":0.000046727862,"about_ca_topic_score_gemma":0.000059126687,"teacher_disagreement_score":0.80242,"about_ca_system_score_codex":0.0003762194,"about_ca_system_score_gemma":0.00036688303,"threshold_uncertainty_score":0.8438969},"labels":[],"label_agreement":null},{"id":"W2129937394","doi":"10.1093/annhyg/46.suppl_1.247","title":"A Latent Variable Model for the Analysis of Variability in the Classification of Radiographs of Pneumoconioses","year":2002,"lang":"en","type":"article","venue":"The Annals of Occupational Hygiene","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Radiography; Pneumoconiosis; Context (archaeology); Statistic; Chest radiograph; Scan statistic; Medicine; Orthodontics; Radiology; Statistics; Mathematics; Pathology; Geography","score_opus":0.4967850990249517,"score_gpt":0.44147257714594407,"score_spread":0.05531252187900765,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2129937394","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9361182,0.0002743966,0.058274806,0.0037656338,0.00004593412,0.00077521295,0.0002831103,0.0000026381886,0.0004600584],"genre_scores_gemma":[0.99882585,0.000097777105,0.0008470071,0.000106631334,0.000009324431,0.000058088466,0.00000851757,0.0000023700443,0.000044434346],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99598867,0.00070286624,0.0013118914,0.00023347391,0.0016253055,0.00013779012],"domain_scores_gemma":[0.9881844,0.008287699,0.0009252462,0.0010273772,0.001557426,0.000017884015],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.022731567,0.000096159725,0.0003905468,0.0003102137,0.00008759519,0.000016382739,0.0011793758,0.0000452553,0.00014008959],"category_scores_gemma":[0.0040014163,0.000043022847,0.00039757442,0.0026230789,0.00035859016,0.00012294116,0.00005028401,0.000068444,7.6304366e-7],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00044077673,0.001815857,0.1731002,0.0000596432,0.0006830151,3.244579e-8,0.0034660443,0.73653257,0.007692082,0.06385466,0.0017578771,0.010597261],"study_design_scores_gemma":[0.00011846004,0.00004747345,0.41193435,0.000009883707,0.00014293929,5.996727e-8,0.00026121974,0.55610037,0.00096173136,0.03032269,0.00006858626,0.00003226575],"about_ca_topic_score_codex":0.00012603342,"about_ca_topic_score_gemma":0.0000385742,"teacher_disagreement_score":0.23883413,"about_ca_system_score_codex":0.000009359463,"about_ca_system_score_gemma":0.00005918286,"threshold_uncertainty_score":0.7878353},"labels":[],"label_agreement":null},{"id":"W2134076777","doi":"10.1016/j.jclinepi.2009.08.005","title":"A valid and reliable belief elicitation method for Bayesian priors","year":2009,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":80,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Women's College Hospital; University of Toronto; Toronto General Hospital; University Health Network; Toronto Western Hospital; Hospital for Sick Children; Public Health Ontario","funders":"Canadian Institutes of Health Research","keywords":"Prior probability; Bayesian probability; Expert elicitation; Statistics; Computer science; Econometrics; Mathematics","score_opus":0.5853364478923457,"score_gpt":0.6222949697557054,"score_spread":0.036958521863359683,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2134076777","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.096276365,0.00048836274,0.822445,0.07867032,0.0012178881,0.00028592494,0.0000025088052,0.0000071477884,0.0006065065],"genre_scores_gemma":[0.22310011,0.0006096329,0.7577446,0.017189855,0.0009963679,0.000003860361,8.1645794e-7,0.0000067698916,0.00034798618],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9856853,0.00556186,0.0072616693,0.00044229714,0.0007113589,0.00033753566],"domain_scores_gemma":[0.87421566,0.119271375,0.00450992,0.00042574268,0.001192671,0.00038462222],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2154536,0.00013575652,0.001906673,0.00016666942,0.00011212691,0.000034480454,0.0005043676,0.00027671867,0.00007204201],"category_scores_gemma":[0.3805293,0.00008129854,0.0006667847,0.0002097568,0.00013361144,0.00022240411,0.000033715616,0.0004813214,0.000010873747],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009934983,0.0004438891,0.12481695,0.00001524373,0.00008388076,0.0000052221308,0.0002200946,0.003069977,0.00024414275,0.018345347,0.061411683,0.7903501],"study_design_scores_gemma":[0.0011590153,0.0029435712,0.1722225,0.000044689863,0.000050430008,0.000024413643,0.00011814898,0.011771343,0.000027315478,0.72384447,0.087698765,0.000095334064],"about_ca_topic_score_codex":0.000006124475,"about_ca_topic_score_gemma":0.000004321033,"teacher_disagreement_score":0.7902548,"about_ca_system_score_codex":0.000024922547,"about_ca_system_score_gemma":0.00013426707,"threshold_uncertainty_score":0.8078556},"labels":[],"label_agreement":null},{"id":"W2134578783","doi":"10.1007/bf02294802","title":"A Two-Stage Logistic Regression Model for Analyzing Inter-Rater Agreement","year":2003,"lang":"en","type":"article","venue":"Psychometrika","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cancer Care Ontario","funders":"","keywords":"Logistic regression; Covariate; Spurious relationship; Psychology; Statistics; Psychopathology; Inter-rater reliability; Population; Odds ratio; Odds; Regression analysis; Econometrics; Clinical psychology; Mathematics; Rating scale; Medicine","score_opus":0.39673202417244113,"score_gpt":0.4638417272771476,"score_spread":0.06710970310470649,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2134578783","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09618266,0.00073628046,0.88823855,0.0010141319,0.001400213,0.0009769334,0.00004472705,0.000047606332,0.011358892],"genre_scores_gemma":[0.9640072,0.000046220084,0.020509833,0.0005060258,0.00007280681,0.00010141408,0.0000056368444,0.00001598478,0.014734851],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.99547434,0.00030425529,0.0010745133,0.0009376857,0.0016771221,0.00053210347],"domain_scores_gemma":[0.99638766,0.00130911,0.00037035064,0.0012111928,0.00051605335,0.00020561018],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009530053,0.00027396478,0.0004321776,0.0008812228,0.0002616594,0.0003693441,0.00082149816,0.00008658316,0.0012894616],"category_scores_gemma":[0.009352503,0.00017681601,0.000311598,0.0017959828,0.0000913677,0.00025824108,0.0000748777,0.0001771797,0.0003329713],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012789593,0.0037951565,0.3596429,0.00026212787,0.00044845341,0.000017907856,0.003701245,0.059976276,0.014795334,0.054179654,0.18329324,0.31860876],"study_design_scores_gemma":[0.009259835,0.0011123102,0.005763337,0.000376581,0.00016172437,0.0000060835837,0.00239772,0.27246216,0.0070224223,0.2741029,0.42552212,0.0018128149],"about_ca_topic_score_codex":0.000009615589,"about_ca_topic_score_gemma":0.000032681815,"teacher_disagreement_score":0.86782455,"about_ca_system_score_codex":0.00014839493,"about_ca_system_score_gemma":0.000059403155,"threshold_uncertainty_score":0.9996235},"labels":[],"label_agreement":null},{"id":"W2134799834","doi":"10.1093/ije/29.6.1070","title":"The effect of collapsing multinomial data when assessing agreement","year":2000,"lang":"en","type":"article","venue":"International Journal of Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Multinomial distribution; Agreement; Medicine; Statistics; Econometrics; Mathematics","score_opus":0.2977449565525722,"score_gpt":0.4964184161530347,"score_spread":0.1986734596004625,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2134799834","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9645391,0.0010855594,0.006948484,0.01931542,0.0045059053,0.00018328855,0.000023767578,0.000003799851,0.0033946931],"genre_scores_gemma":[0.9926549,0.00023808223,0.0053713224,0.00047466197,0.0007799156,0.0000015874364,0.0000057267343,0.0000046843534,0.0004691181],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99196166,0.0033143342,0.00258677,0.0002949124,0.0015952358,0.00024709676],"domain_scores_gemma":[0.969726,0.027009657,0.0017080564,0.0006794995,0.00079037424,0.000086388485],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.056706358,0.00011753845,0.00053651753,0.0001291205,0.00012896757,0.0001025413,0.003433855,0.00006774171,0.0011872349],"category_scores_gemma":[0.047438722,0.000058746864,0.00020556613,0.000098059354,0.00028075042,0.00043160174,0.00025607433,0.00024790378,0.00005134287],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005800642,0.000049505692,0.1262987,0.000002520375,0.00023738082,0.000013292566,0.00014888801,0.0034646823,0.0006210772,0.00027447683,0.03722597,0.8310834],"study_design_scores_gemma":[0.0029541566,0.0011049228,0.09388031,0.0002509457,0.00010077625,0.00022075787,0.00043188318,0.013486652,0.0013069299,0.056742977,0.82929647,0.00022322654],"about_ca_topic_score_codex":0.000049801005,"about_ca_topic_score_gemma":0.000026830707,"teacher_disagreement_score":0.8308602,"about_ca_system_score_codex":0.00010341656,"about_ca_system_score_gemma":0.00011805015,"threshold_uncertainty_score":0.9997258},"labels":[],"label_agreement":null},{"id":"W2136733697","doi":"","title":"Estimating agreement coefficients from sample survey data","year":2012,"lang":"en","type":"article","venue":"Survey methodology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"","keywords":"Jackknife resampling; Statistics; Concordance; Mathematics; Cohen's kappa; Standard error; Correlation coefficient; Variance (accounting); Sampling (signal processing); Sample (material); Concordance correlation coefficient; Linearization; Computer science; Medicine; Nonlinear system; Accounting","score_opus":0.8726461020763974,"score_gpt":0.5551506380714631,"score_spread":0.3174954640049342,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2136733697","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.35562626,0.0002874872,0.6363247,0.00010997421,0.0046165916,0.00026945854,0.0024579326,0.00003000024,0.000277561],"genre_scores_gemma":[0.6076887,0.0000056937197,0.38948977,0.00039872422,0.00033315687,0.000011128115,0.0019001568,0.000017013237,0.00015562912],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.93870497,0.053953607,0.0018541134,0.0015297029,0.002826548,0.0011310799],"domain_scores_gemma":[0.83284336,0.16014433,0.00078210875,0.0048683113,0.0009122931,0.0004496143],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.27626795,0.00029913746,0.00079911907,0.00018851642,0.00030197974,0.00016900203,0.003276782,0.00018647555,0.0034358057],"category_scores_gemma":[0.313595,0.00022431945,0.00009042594,0.00095671904,0.00022941305,0.0005006731,0.0016465529,0.00031467358,0.0011995353],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010664633,0.00025117066,0.9368598,0.0000029490698,0.00005658237,5.6634786e-7,0.0004150082,0.0006549975,0.00027240696,0.00003978461,0.011169594,0.05017048],"study_design_scores_gemma":[0.00036641624,0.000055115874,0.9806669,0.000008128782,0.000020158941,9.480149e-7,0.0001664747,0.0063349716,0.0001991232,0.004805822,0.007109928,0.00026605593],"about_ca_topic_score_codex":0.027789982,"about_ca_topic_score_gemma":0.008170281,"teacher_disagreement_score":0.25206247,"about_ca_system_score_codex":0.00008274973,"about_ca_system_score_gemma":0.000108289336,"threshold_uncertainty_score":0.9995781},"labels":[],"label_agreement":null},{"id":"W2138454932","doi":"10.1111/1467-9884.00331","title":"An exact bootstrap confidence interval for kappa in small samples","year":2002,"lang":"en","type":"article","venue":"Journal of the Royal Statistical Society Series D (The Statistician)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cancer Care Ontario","funders":"","keywords":"Confidence interval; Statistics; Mathematics; Sample size determination; Coverage probability; CDF-based nonparametric confidence interval; Robust confidence intervals; Monte Carlo method; Exact statistics; Binary number; Sample (material); Distribution (mathematics); Mathematical analysis; Arithmetic; Physics","score_opus":0.20950103259085956,"score_gpt":0.3675185352347408,"score_spread":0.15801750264388126,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2138454932","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.05686508,0.00024601578,0.93083864,0.008962886,0.0013089055,0.0007269779,0.0007117027,0.000012644125,0.00032716844],"genre_scores_gemma":[0.93710893,0.000052845815,0.060430117,0.0012050269,0.00034579824,0.000020396388,0.0000034023515,0.000020271136,0.00081320043],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99504447,0.0007973628,0.001679104,0.0003597267,0.0016194949,0.0004998565],"domain_scores_gemma":[0.9913616,0.0063773524,0.00082332594,0.0005541267,0.0006796557,0.00020389294],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.006324474,0.0002527129,0.00057079317,0.00003741078,0.00048462133,0.0004713397,0.0018693885,0.00009832347,0.0017337834],"category_scores_gemma":[0.0087715285,0.00012830901,0.00041530674,0.00032551048,0.0008073517,0.0002739123,0.00012465022,0.0005758074,0.000027688639],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015045428,0.0021869938,0.028769974,0.00024978147,0.00057003443,0.00007163464,0.0221584,0.022877263,0.0006848927,0.2653466,0.39224115,0.26333874],"study_design_scores_gemma":[0.0018775169,0.002392736,0.16403452,0.00022623387,0.00021852256,0.000044742923,0.0132927215,0.088731915,0.00021636357,0.6742604,0.054128952,0.0005753925],"about_ca_topic_score_codex":0.00014839011,"about_ca_topic_score_gemma":0.00054566737,"teacher_disagreement_score":0.88024384,"about_ca_system_score_codex":0.00018351672,"about_ca_system_score_gemma":0.00009491983,"threshold_uncertainty_score":0.999578},"labels":[],"label_agreement":null},{"id":"W2138928670","doi":"10.1177/0163278711425041","title":"A Comparison of Empirical Ranking Methods of Frequency and Severity Ratings of Clinical Presentations","year":2011,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; University of Calgary","funders":"","keywords":"Ranking (information retrieval); Statistics; Psychology; Medicine; MEDLINE; Econometrics; Clinical psychology; Computer science; Mathematics; Artificial intelligence; Biology","score_opus":0.893414992052289,"score_gpt":0.7377671514789872,"score_spread":0.15564784057330183,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2138928670","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.942885,0.0005181855,0.045001876,0.0074090133,0.00062767084,0.0018351522,0.00001205843,0.000008973667,0.0017020728],"genre_scores_gemma":[0.93486094,0.000025674077,0.06479463,0.00019253447,0.000022807404,0.000053475265,0.0000024313015,0.0000035293622,0.00004397219],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.97668105,0.01641747,0.0038778754,0.00036178876,0.0024634663,0.00019836039],"domain_scores_gemma":[0.98884404,0.0066298936,0.0023187809,0.0006193613,0.0014905373,0.000097406584],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.082388155,0.00008942387,0.00060886523,0.000109406516,0.00033910823,0.0000073848732,0.00040872142,0.0000880565,0.00059499],"category_scores_gemma":[0.02009672,0.000050598,0.00012595126,0.00058319053,0.0003834685,0.00014328324,0.00014730376,0.00029516907,0.0000033604304],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006981074,0.0005878839,0.88104934,0.00007019439,0.000024034236,1.4745581e-8,0.036461543,0.00003064075,0.0003465058,0.0012111499,0.0024229118,0.077725984],"study_design_scores_gemma":[0.0005698503,0.00030614002,0.89619064,0.00013215934,0.00004996523,2.3153912e-7,0.011203008,0.012102186,0.00082329934,0.07843019,0.00014103403,0.00005127185],"about_ca_topic_score_codex":0.00021203354,"about_ca_topic_score_gemma":0.00007918526,"teacher_disagreement_score":0.07767472,"about_ca_system_score_codex":0.0000206596,"about_ca_system_score_gemma":0.0008566431,"threshold_uncertainty_score":0.98815745},"labels":[],"label_agreement":null},{"id":"W2140519023","doi":"10.1002/9781118445112.stat05255","title":"Receiver Operating Characteristic (<scp>ROC</scp>) Curves","year":2014,"lang":"en","type":"other","venue":"Wiley StatsRef: Statistics Reference Online","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Receiver operating characteristic; Fraction (chemistry); Boundary (topology); Mathematics; Statistics; Arbitrariness; Mathematical analysis; Chromatography","score_opus":0.14698234004203004,"score_gpt":0.38344374048368907,"score_spread":0.23646140044165903,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2140519023","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0013315154,0.011774194,0.4011115,0.001347668,0.007572323,0.0049245013,0.14182048,0.0011873889,0.42893043],"genre_scores_gemma":[0.0008927052,0.018375162,0.09989526,0.0025842478,0.0014756052,0.00012696718,0.010020144,0.00071168493,0.8659182],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9890569,0.0010576611,0.0023048206,0.0019425311,0.0046011424,0.0010369502],"domain_scores_gemma":[0.99067026,0.003419924,0.0019596557,0.0021924572,0.0012349541,0.0005227502],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0036640856,0.0010261722,0.001711817,0.0006109214,0.00030010776,0.00058954174,0.0023323142,0.00053434377,0.0113399215],"category_scores_gemma":[0.018774481,0.0007828132,0.00016950852,0.0007227904,0.00046159545,0.00018091185,0.00045066068,0.0011584878,0.0039862073],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000007775878,0.00031703993,0.0006814643,0.00042247085,0.00009622963,0.000028202176,0.00010659612,0.000019241686,0.00009424642,0.0019628499,0.9655343,0.030729553],"study_design_scores_gemma":[0.00049894146,0.00033468535,0.0025244537,0.0028946493,0.00013391043,0.0000058310507,0.00022183573,0.0012606634,0.000011930235,0.0074419207,0.984187,0.00048416227],"about_ca_topic_score_codex":0.00023784248,"about_ca_topic_score_gemma":0.00089580024,"teacher_disagreement_score":0.4369878,"about_ca_system_score_codex":0.00017288749,"about_ca_system_score_gemma":0.00046295597,"threshold_uncertainty_score":0.9994623},"labels":[],"label_agreement":null},{"id":"W2142200036","doi":"10.1002/sim.3110","title":"Reliability analysis for continuous measurements: Equivalence test for agreement","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Western Hospital; University Health Network; Memorial University of Newfoundland; Canadian Blood Services; University of Toronto","funders":"","keywords":"Repeatability; Equivalence (formal languages); Reliability (semiconductor); Inter-rater reliability; Computer science; Reliability engineering; Intra-rater reliability; Statistics; Test (biology); Consistency (knowledge bases); Mathematics; Rating scale; Artificial intelligence; Engineering","score_opus":0.20769671851629035,"score_gpt":0.4449493606354994,"score_spread":0.23725264211920905,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2142200036","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010049454,0.0001833444,0.9830931,0.0015159923,0.0008430755,0.002070747,0.00048037284,0.000020824322,0.0017431062],"genre_scores_gemma":[0.8380165,0.00003212174,0.1591485,0.0007489221,0.00030672463,0.000198937,0.00009821016,0.000015350226,0.0014347372],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99248743,0.00016993032,0.0023654676,0.0009747833,0.003288802,0.0007135741],"domain_scores_gemma":[0.9791664,0.016481029,0.00057605945,0.0010637984,0.00248739,0.00022535429],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.051765397,0.00027481257,0.00092440966,0.00063164433,0.00020334753,0.000058914266,0.0008814587,0.000100455356,0.0006228099],"category_scores_gemma":[0.0931019,0.00019376467,0.00017811779,0.0016629042,0.00040737854,0.00011252827,0.00007777477,0.00016861067,0.000020313939],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001004784,0.001353057,0.7889188,0.00028767646,0.00040535623,0.000012834078,0.0016732379,0.0033454993,0.004115014,0.011130948,0.08845501,0.09929781],"study_design_scores_gemma":[0.007352343,0.0039929515,0.5142854,0.00023428233,0.0011701988,0.0000012220845,0.00407656,0.01746142,0.0012203467,0.3865453,0.062896974,0.00076299394],"about_ca_topic_score_codex":0.00013186247,"about_ca_topic_score_gemma":0.001169081,"teacher_disagreement_score":0.82796705,"about_ca_system_score_codex":0.00035074487,"about_ca_system_score_gemma":0.000088698864,"threshold_uncertainty_score":0.97640705},"labels":[],"label_agreement":null},{"id":"W2142892303","doi":"10.2466/pr0.2000.87.3f.1171","title":"Creating Comparability among Reliability Coefficients: The Case of Cronbach Alpha and Cohen Kappa","year":2000,"lang":"en","type":"article","venue":"Psychological Reports","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":44,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Winnipeg","funders":"","keywords":"Cronbach's alpha; Comparability; Kappa; Psychology; Reliability (semiconductor); Alpha (finance); Estimator; Statistics; Metric (unit); Cohen's kappa; Contrast (vision); Complement (music); Psychometrics; Social psychology; Clinical psychology; Mathematics; Computer science; Artificial intelligence; Combinatorics; Chemistry; Engineering","score_opus":0.12349732271461297,"score_gpt":0.4136398494609489,"score_spread":0.2901425267463359,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2142892303","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97277987,0.00014078864,0.00046255617,0.0006268174,0.00021744793,0.00070091226,0.0000068768136,0.000038699152,0.025026044],"genre_scores_gemma":[0.998752,0.000016865702,0.0005081419,0.00020540929,0.000036131776,0.00003271312,0.0000016342287,0.000005089743,0.00044200246],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99432284,0.00095046876,0.0019112325,0.001173495,0.0012899202,0.0003520704],"domain_scores_gemma":[0.9945118,0.0024362444,0.00056744553,0.0018860813,0.00040537838,0.00019303942],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.017879734,0.0002156262,0.00051862537,0.000042517557,0.0004156884,0.00016069428,0.00044649706,0.00012228123,0.0030606024],"category_scores_gemma":[0.008336258,0.00010877384,0.00019127125,0.00055174343,0.0014239919,0.00016157095,0.00012789661,0.00031927286,0.000029839164],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015298926,0.001531631,0.78409535,0.000017163864,0.000019700054,0.00078705215,0.00074934436,0.0012058124,0.00015662196,0.00010359552,0.0029284342,0.2082523],"study_design_scores_gemma":[0.0002777298,0.0002716358,0.95814455,0.000023334453,0.000025262196,0.0009881625,0.0005281726,0.00084035104,0.00010202701,0.03252944,0.0060863234,0.00018299789],"about_ca_topic_score_codex":0.0002522452,"about_ca_topic_score_gemma":0.000045263645,"teacher_disagreement_score":0.20806931,"about_ca_system_score_codex":0.00004075881,"about_ca_system_score_gemma":0.000017082568,"threshold_uncertainty_score":0.9979882},"labels":[],"label_agreement":null},{"id":"W2143965974","doi":"10.1093/ije/dyp370","title":"Forest plots in reports of systematic reviews: a cross-sectional study reviewing current practice","year":2010,"lang":"en","type":"article","venue":"International Journal of Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":87,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"Cancer Research UK; University of Ottawa","keywords":"Cross-sectional study; Current (fluid); Medicine; MEDLINE; Systematic review; Environmental health; Geography; Political science; Engineering; Pathology","score_opus":0.5444736336084718,"score_gpt":0.5819775103364283,"score_spread":0.03750387672795652,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2143965974","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9765968,0.007376542,0.0025696715,0.0015387712,0.01056196,0.0010611647,0.0000029212993,0.0000037172053,0.0002884869],"genre_scores_gemma":[0.9956835,0.00088735635,0.002594403,0.0003123213,0.00045420523,0.000033097,0.0000010973528,0.0000051952297,0.000028831195],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98003614,0.0055925003,0.0111638345,0.0004576092,0.0025023234,0.00024759865],"domain_scores_gemma":[0.9625373,0.022494918,0.009888959,0.00061689055,0.00434673,0.000115220864],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.15785737,0.00015018143,0.0012963117,0.00049520395,0.000038855327,0.00006031524,0.0011787801,0.00008551392,0.00018588385],"category_scores_gemma":[0.50192547,0.00009108941,0.00041912097,0.0002845149,0.00013410869,0.00061368063,0.00014933832,0.0007360405,0.000030240812],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008294608,0.00069043273,0.9940318,0.00029408306,0.00008050904,0.00006757125,0.0001960741,0.00065819215,0.00013070519,0.0007980984,0.00068788626,0.0022817056],"study_design_scores_gemma":[0.00052446657,0.00032914415,0.9345525,0.002676393,0.000052289914,0.0011874755,0.00022246232,0.000202508,0.000010375327,0.033807453,0.026310299,0.00012462253],"about_ca_topic_score_codex":0.000046633326,"about_ca_topic_score_gemma":0.00011928858,"teacher_disagreement_score":0.34406808,"about_ca_system_score_codex":0.00012272527,"about_ca_system_score_gemma":0.00014702821,"threshold_uncertainty_score":0.86716306},"labels":[],"label_agreement":null},{"id":"W2146976162","doi":"10.1002/(sici)1097-0258(20000215)19:3<373::aid-sim337>3.0.co;2-y","title":"Testing the equality of two dependent kappa statistics","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":89,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph; Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Kappa; Correlation; Monte Carlo method; Mathematics; Cohen's kappa; Goodness of fit; Dependency (UML); Sample size determination; Variable (mathematics); Computer science; Artificial intelligence","score_opus":0.2670104097609483,"score_gpt":0.4556427207423286,"score_spread":0.18863231098138028,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2146976162","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18723473,0.00054637075,0.74531734,0.0034487566,0.0013127177,0.0012227242,0.001363933,0.000038839516,0.059514623],"genre_scores_gemma":[0.922231,0.00005878457,0.07571973,0.0005517206,0.00016818439,0.000014503446,0.00001900848,0.000010077611,0.0012269821],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99383336,0.0007102268,0.0015993909,0.0003923825,0.0031550494,0.00030958533],"domain_scores_gemma":[0.9887842,0.009418523,0.00034098257,0.00078289735,0.000582,0.00009143108],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.014635624,0.00015965826,0.00043650178,0.00010610052,0.00012871117,0.00003519339,0.0008215466,0.00003758503,0.0054663597],"category_scores_gemma":[0.016682744,0.00008785782,0.000021917085,0.0007223649,0.0006210688,0.000054083335,0.00007111597,0.0002935005,0.00010873619],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009540518,0.00028521835,0.09400722,0.00006967088,0.000030911684,0.00004357227,0.0032044826,0.00447572,0.0011688197,0.09661052,0.04216598,0.7578425],"study_design_scores_gemma":[0.0018952189,0.00043297387,0.1491368,0.00017189271,0.000052426283,0.000006656131,0.0015842917,0.012749433,0.00014007238,0.8244635,0.00915303,0.00021370706],"about_ca_topic_score_codex":0.0010010153,"about_ca_topic_score_gemma":0.0006780621,"teacher_disagreement_score":0.7576288,"about_ca_system_score_codex":0.00006649305,"about_ca_system_score_gemma":0.00009942054,"threshold_uncertainty_score":0.9954428},"labels":[],"label_agreement":null},{"id":"W2149750314","doi":"10.1136/jech.2003.010546","title":"A simple model for potential use with a misclassified binary outcome in epidemiology","year":2004,"lang":"en","type":"article","venue":"Journal of Epidemiology & Community Health","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Medicine; Epidemiology; Spurious relationship; Context (archaeology); Statistics; Outcome (game theory); Logistic regression; Odds ratio; Disease; Econometrics; Pathology; Internal medicine; Mathematics","score_opus":0.6286775932544271,"score_gpt":0.5194982981963062,"score_spread":0.10917929505812085,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2149750314","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6382628,0.00042112367,0.32058156,0.03995938,0.00022803486,0.00047654752,0.00003160259,0.000010051606,0.000028916433],"genre_scores_gemma":[0.89106625,0.00029281768,0.09282991,0.015628269,0.00008290143,0.000018152075,0.000011569838,0.000015180284,0.000054962675],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9756893,0.015842233,0.0062438254,0.00047407357,0.0006186475,0.0011319158],"domain_scores_gemma":[0.9566444,0.0364305,0.003965888,0.0015434078,0.0008574678,0.0005583463],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1449267,0.0003007047,0.0024634406,0.00061223947,0.00060018693,0.000023597193,0.0015240347,0.00028489463,0.000025422727],"category_scores_gemma":[0.08756481,0.00018701798,0.0004891327,0.00047086232,0.00048857444,0.0004723911,0.00021126191,0.002169756,0.000007667417],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010127041,0.0007300973,0.54498833,0.00004583614,0.000047080335,0.000006326836,0.0006104266,0.44293344,0.000035270423,0.0038292352,0.004463903,0.0012973647],"study_design_scores_gemma":[0.004082144,0.0037641365,0.49642286,0.00018482651,0.000030232628,0.00009976851,0.001286415,0.053696536,0.0000027032781,0.43758112,0.0026180393,0.00023120837],"about_ca_topic_score_codex":0.0032951513,"about_ca_topic_score_gemma":0.0065289345,"teacher_disagreement_score":0.43375188,"about_ca_system_score_codex":0.00081787107,"about_ca_system_score_gemma":0.0012771302,"threshold_uncertainty_score":0.9426631},"labels":[],"label_agreement":null},{"id":"W2151442672","doi":"10.1002/nur.20095","title":"A hybrid qualitative method for pretesting questionnaires: The example of a questionnaire to caregivers of Alzheimer disease patients","year":2005,"lang":"en","type":"article","venue":"Research in Nursing & Health","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Joseph’s Healthcare Hamilton; McGill University; McMaster University; Jewish General Hospital","funders":"","keywords":"Interview; Disease; Psychology; Clinical psychology; Cognition; Family medicine; Medicine; Gerontology; Psychiatry","score_opus":0.4164801381706226,"score_gpt":0.5917608903707678,"score_spread":0.17528075220014522,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2151442672","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7446231,0.0044669164,0.13403329,0.10613996,0.00060769665,0.009311998,0.00025101416,0.000043243195,0.0005227621],"genre_scores_gemma":[0.9748638,0.000017419876,0.02473848,0.00014157218,0.000058589805,0.00014156383,0.0000050023823,0.000010155268,0.000023417464],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9889803,0.005637704,0.0011511921,0.0005580291,0.0030839522,0.00058885815],"domain_scores_gemma":[0.9890979,0.0071619586,0.00039692814,0.00072908425,0.002251367,0.0003627576],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.046373725,0.00012151336,0.00035752772,0.0004966607,0.00036371604,0.00005107451,0.0006215476,0.000029232238,0.00001024305],"category_scores_gemma":[0.020979319,0.0000835828,0.00010538979,0.0010534354,0.00036145703,0.00019613052,0.000071465656,0.00024311776,0.000006072949],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006770679,0.0010963302,0.01506801,0.0002442191,0.000021550873,4.354914e-7,0.10546377,0.0073348894,0.00006123137,0.009014485,0.013709187,0.8473088],"study_design_scores_gemma":[0.004690727,0.0063852537,0.33134615,0.02415267,0.00006955779,0.0000016357588,0.053774398,0.05196789,0.0023323812,0.5119494,0.012549142,0.0007807433],"about_ca_topic_score_codex":0.0035817237,"about_ca_topic_score_gemma":0.00022248164,"teacher_disagreement_score":0.84652805,"about_ca_system_score_codex":0.0005800212,"about_ca_system_score_gemma":0.00091760047,"threshold_uncertainty_score":0.9872674},"labels":[],"label_agreement":null},{"id":"W2152177373","doi":"10.2106/jbjs.h.01624","title":"Evaluating Agreement: Conducting a Reliability Study","year":2009,"lang":"en","type":"article","venue":"Journal of Bone and Joint Surgery","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":162,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Health Science Centre; University of Toronto; McMaster University","funders":"","keywords":"Reliability (semiconductor); Sample size determination; Computer science; Reliability engineering; Context (archaeology); Test (biology); Psychology; Statistics; Engineering; Mathematics; Power (physics)","score_opus":0.5353733677564685,"score_gpt":0.4567027472231148,"score_spread":0.07867062053335366,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2152177373","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9942325,0.0005769378,0.00027913952,0.0036489822,0.00059360143,0.00022600869,7.5195715e-7,0.0000068592835,0.00043523067],"genre_scores_gemma":[0.9982053,0.000051810013,0.001007356,0.0003651157,0.00023178468,0.0000012976265,1.5192016e-7,0.0000036001823,0.00013360233],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9920106,0.00096732716,0.0033979043,0.00034241402,0.0030172875,0.00026445586],"domain_scores_gemma":[0.99405074,0.0021980915,0.0019903765,0.00043823797,0.0011401228,0.00018240057],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07968286,0.00015959043,0.0010120135,0.00048233836,0.00023112656,0.00024350482,0.000095938805,0.000047664787,0.00027462313],"category_scores_gemma":[0.02232389,0.00010226741,0.0004188722,0.0004610953,0.000060458337,0.0005913446,0.00006207136,0.0003206627,0.000014097607],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004087916,0.0027258282,0.4889885,0.000039086008,0.000102337006,0.00019575711,0.003877291,0.0008973652,0.016710369,0.000052090276,0.016541164,0.46946144],"study_design_scores_gemma":[0.0015204004,0.0027281658,0.9408506,0.00036132958,0.00013341014,0.00020167598,0.012910839,0.0006483971,0.001311331,0.036512583,0.0024272602,0.000394006],"about_ca_topic_score_codex":0.000011262881,"about_ca_topic_score_gemma":0.0000031745428,"teacher_disagreement_score":0.46906742,"about_ca_system_score_codex":0.000060934824,"about_ca_system_score_gemma":0.00014848814,"threshold_uncertainty_score":0.9859115},"labels":[],"label_agreement":null},{"id":"W2155243985","doi":"10.20982/tqmp.08.1.p023","title":"Computing Inter-Rater Reliability for Observational Data: An Overview and Tutorial","year":2012,"lang":"en","type":"article","venue":"Tutorials in Quantitative Methods for Psychology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3862,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"National Institute on Alcohol Abuse and Alcoholism","keywords":"Observational study; Reliability (semiconductor); Computer science; Syntax; Consistency (knowledge bases); Class (philosophy); Statistics; Multiple comparisons problem; Inter-rater reliability; Econometrics; Mathematics; Artificial intelligence; Power (physics); Rating scale","score_opus":0.8909602875101015,"score_gpt":0.7046596688682544,"score_spread":0.1863006186418471,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2155243985","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.11712881,0.0014551728,0.853558,0.0013512988,0.024463832,0.0016948615,0.00020682828,0.000027351409,0.00011382193],"genre_scores_gemma":[0.02445561,0.00005041047,0.9720247,0.0006627998,0.0024844545,0.00015725139,0.00009646177,0.000020293217,0.000048019083],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9906228,0.0053009326,0.0017274407,0.0012452644,0.000524117,0.0005794559],"domain_scores_gemma":[0.97033554,0.02654318,0.00059928163,0.0015239429,0.0008366702,0.00016137164],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.10479097,0.00025873363,0.0008242232,0.00019436795,0.00020330562,0.00014280995,0.0010800565,0.0002083337,0.00008539781],"category_scores_gemma":[0.052031912,0.00019697413,0.00011874525,0.00040386888,0.00031196818,0.0011651961,0.00028576757,0.00019146557,0.000010857902],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003044948,0.0017159109,0.13171375,0.0002072445,0.00012618888,4.06666e-7,0.006938288,0.0001156606,0.013957978,0.5343839,0.013677175,0.29411855],"study_design_scores_gemma":[0.0020939033,0.0008371876,0.055337533,0.000034101864,0.000038224487,0.0000017694829,0.00049138453,0.0046858224,0.00026351286,0.4983064,0.43759313,0.000317039],"about_ca_topic_score_codex":0.00003845892,"about_ca_topic_score_gemma":0.000027277176,"teacher_disagreement_score":0.42391595,"about_ca_system_score_codex":0.000073924355,"about_ca_system_score_gemma":0.000065669585,"threshold_uncertainty_score":0.95595324},"labels":[],"label_agreement":null},{"id":"W2155433306","doi":"10.1590/s1413-35552006000200002","title":"Importance and clarification of measurement properties in rehabilitation","year":2006,"lang":"en","type":"article","venue":"Brazilian Journal of Physical Therapy","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":109,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Canadian Institutes of Health Research; Coordenação de Aperfeiçoamento de Pessoal de Nível Superior; University of Alberta","keywords":"Reliability (semiconductor); Rehabilitation; Quality (philosophy); Clinical Practice; Validity; Field (mathematics); Psychology; Applied psychology; Computer science; Management science; Psychometrics; Medicine; Power (physics); Physical therapy; Clinical psychology; Engineering; Mathematics; Epistemology","score_opus":0.1405243827265361,"score_gpt":0.32899155337758973,"score_spread":0.18846717065105362,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2155433306","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9967065,0.0009016003,0.00026167967,0.001711323,0.00005960467,0.0001928585,7.016931e-7,0.0000022645472,0.00016346948],"genre_scores_gemma":[0.9993198,0.000035074074,0.00049016246,0.00003283509,0.00010140642,0.0000037193556,1.15852814e-7,0.0000039809956,0.000012887414],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.99666184,0.00034040734,0.0009648737,0.0001704825,0.0017563538,0.00010606548],"domain_scores_gemma":[0.99809533,0.00021125171,0.0005915935,0.0002240087,0.0008390243,0.000038806218],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0056831264,0.00009158583,0.00030750877,0.00014569142,0.000033415876,0.0000432901,0.0002565675,0.00002792096,0.000012001775],"category_scores_gemma":[0.0008374867,0.000054981014,0.00012605084,0.0003275616,0.00015459037,0.00032117302,0.000010146343,0.00012781931,0.000002129911],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00059950224,0.0020231374,0.23994507,0.000035541976,0.000027989565,0.000001917044,0.002508861,0.0014873075,0.56878316,0.0024542552,0.00062814716,0.18150511],"study_design_scores_gemma":[0.0010482734,0.000920958,0.83440775,0.00013362132,0.000004984463,0.000001793555,0.00054505793,0.0003543655,0.033704326,0.12767541,0.0011107656,0.00009271057],"about_ca_topic_score_codex":0.000014767531,"about_ca_topic_score_gemma":0.00002060828,"teacher_disagreement_score":0.59446263,"about_ca_system_score_codex":0.000060506463,"about_ca_system_score_gemma":0.00006159822,"threshold_uncertainty_score":0.22420606},"labels":[],"label_agreement":null},{"id":"W2162264344","doi":"10.1002/9781118445112.stat05301","title":"Agreement, Measurement of","year":2014,"lang":"en","type":"other","venue":"Wiley StatsRef: Statistics Reference Online","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph","funders":"","keywords":"Categorical variable; Agreement; Unanimity; Measure (data warehouse); Interpretation (philosophy); Kappa; Statistics; Mathematics; Interval (graph theory); Monotonic function; Computer science; Data mining; Combinatorics; Mathematical analysis; Geometry","score_opus":0.2273606341632211,"score_gpt":0.39418978133845184,"score_spread":0.16682914717523073,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2162264344","genre_codex":"methods","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00006481171,0.002562881,0.5003544,0.00041329797,0.002181149,0.0017613699,0.02776156,0.0001966451,0.4647039],"genre_scores_gemma":[0.009366938,0.0049277926,0.2608642,0.00060919323,0.001121199,0.00011053142,0.0022536162,0.0008149431,0.7199316],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9835387,0.0007643091,0.0025343753,0.001393303,0.0110532455,0.0007160705],"domain_scores_gemma":[0.99186,0.0007017772,0.0025017152,0.0024557267,0.002126952,0.00035384364],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.005992513,0.0007675931,0.0015730449,0.0008076658,0.00010442462,0.00015091994,0.0021512306,0.000420074,0.020567775],"category_scores_gemma":[0.0054002767,0.0005746,0.00019305927,0.00065005117,0.00046540995,0.00007125984,0.00036503468,0.00059077033,0.0012802596],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004539737,0.0006276605,0.0005514553,0.00022877748,0.00017166958,0.000007101124,0.000049217626,0.00003729418,0.00027770188,0.011301603,0.9174485,0.06925362],"study_design_scores_gemma":[0.0007774363,0.0004696364,0.00064125244,0.00094759424,0.00014017089,0.0000012503507,0.0001600936,0.00042850213,0.0000782428,0.030694721,0.9650038,0.0006572824],"about_ca_topic_score_codex":0.00025296764,"about_ca_topic_score_gemma":0.0021611832,"teacher_disagreement_score":0.2552277,"about_ca_system_score_codex":0.00021535675,"about_ca_system_score_gemma":0.00054522726,"threshold_uncertainty_score":0.99967057},"labels":[],"label_agreement":null},{"id":"W2163148621","doi":"10.3138/jvme.37.4.377","title":"North American Veterinary Licensing Examination Pacing Study","year":2010,"lang":"en","type":"article","venue":"Journal of Veterinary Medical Education","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"","keywords":"Medicine; Test (biology); Psychology; Word (group theory); Presentation (obstetrics); Mathematics; Surgery","score_opus":0.2172615267746426,"score_gpt":0.46451331547140057,"score_spread":0.24725178869675796,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2163148621","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.993024,0.000026520089,0.0004920223,0.0018927219,0.0040073767,0.00019768492,4.9553705e-7,0.000010167917,0.00034904788],"genre_scores_gemma":[0.997055,0.00001430805,0.0015700605,0.00036484416,0.00091729144,0.0000055850396,0.0000011555834,0.000008896368,0.000062878076],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99377567,0.00082939415,0.0012767462,0.00027921423,0.003634947,0.00020403982],"domain_scores_gemma":[0.9964067,0.0007103928,0.0010514744,0.00044325873,0.000996504,0.00039169978],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.011211208,0.00014495382,0.00034450003,0.0003842649,0.00017244487,0.00016054735,0.00071139616,0.000054305598,0.00048567116],"category_scores_gemma":[0.012652491,0.000102857484,0.00012214917,0.0007303132,0.00017820392,0.0005703693,0.000108019565,0.0006799319,0.000036861868],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012995703,0.002337358,0.128368,0.000010004769,0.000024519704,0.00007056295,0.0028712586,0.000010908394,0.0055997423,0.000010020732,0.0014517179,0.85911596],"study_design_scores_gemma":[0.00042238907,0.004918181,0.96932447,0.000058157726,0.000025156862,0.00065737206,0.006724181,0.00023242756,0.000051601863,0.00017378015,0.017274292,0.00013800431],"about_ca_topic_score_codex":0.000033488846,"about_ca_topic_score_gemma":0.00009598347,"teacher_disagreement_score":0.858978,"about_ca_system_score_codex":0.0000913033,"about_ca_system_score_gemma":0.0007430947,"threshold_uncertainty_score":0.99566436},"labels":[],"label_agreement":null},{"id":"W2165514584","doi":"10.7202/900116ar","title":"Étude de fiabilité d’un instrument d’observation des comportements de l’élève en classe","year":2009,"lang":"fr","type":"article","venue":"Revue des sciences de l éducation","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"","keywords":"Humanities; Philosophy","score_opus":0.5535813751294997,"score_gpt":0.45142720773716905,"score_spread":0.10215416739233063,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2165514584","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9494568,0.0017782253,0.02887232,0.015498999,0.0012183408,0.0006628985,0.0000090720605,0.000051608364,0.0024517179],"genre_scores_gemma":[0.91316533,0.0009374029,0.08313472,0.0012897939,0.00032582492,0.000048212194,0.000011204048,0.000010141609,0.0010773689],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9934599,0.0016047643,0.0012412589,0.0009827553,0.0016035846,0.00110776],"domain_scores_gemma":[0.9969336,0.00082626793,0.0005616623,0.00070363417,0.000561054,0.00041378685],"candidate_categories":["metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.020416247,0.00034492151,0.00037404077,0.00029177766,0.0015361063,0.0008608114,0.0014645994,0.00024405571,0.00057144486],"category_scores_gemma":[0.0040169284,0.00032717787,0.00020802826,0.0019206706,0.0015046,0.0017161476,0.00010741648,0.0003112406,0.00017232266],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000041312538,0.0019741738,0.40777102,0.00025189543,0.000036544425,0.000008921507,0.057571422,0.011572855,0.012265801,0.013075793,0.001279712,0.49415055],"study_design_scores_gemma":[0.00034257985,0.000733818,0.6197635,0.0005294412,0.000068114416,0.000064008025,0.0087402,0.044066347,0.0034774847,0.31794283,0.003905587,0.00036607537],"about_ca_topic_score_codex":0.00088433095,"about_ca_topic_score_gemma":0.0006259127,"teacher_disagreement_score":0.4937845,"about_ca_system_score_codex":0.004602644,"about_ca_system_score_gemma":0.0013760881,"threshold_uncertainty_score":0.99991804},"labels":[],"label_agreement":null},{"id":"W2166290337","doi":"10.1186/1471-2288-3-18","title":"How does correlation structure differ between real and fabricated data-sets?","year":2003,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Population Health Research Institute; McMaster University","funders":"Shiraz University; Shiraz University of Medical Sciences; McMaster University","keywords":"Correlation; Data set; Set (abstract data type); Correlation coefficient; Statistics; Data mining; Pearson product-moment correlation coefficient; Misconduct; Computer science; Mathematics","score_opus":0.8378742148671262,"score_gpt":0.607124913089742,"score_spread":0.23074930177738417,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2166290337","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.59226984,0.00030566144,0.39537004,0.009644175,0.0006638045,0.0005914786,0.00008988853,0.000034983714,0.001030121],"genre_scores_gemma":[0.92327195,0.00034543814,0.07417823,0.00014868614,0.00028531576,0.000019364255,0.00007933628,0.000014262733,0.0016574428],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9432103,0.045661774,0.0008612231,0.0013922144,0.008049086,0.0008253783],"domain_scores_gemma":[0.84601533,0.14953129,0.00023829573,0.0022761258,0.00096664677,0.0009723126],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.12796389,0.0001786839,0.00062961463,0.0003621969,0.0003470515,0.00028238562,0.0018413641,0.00048391632,0.0021465875],"category_scores_gemma":[0.49407032,0.00009412664,0.00005690762,0.0009426007,0.001262025,0.0002897707,0.00081246486,0.0011527167,0.00004761652],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001428909,0.000085095744,0.6715367,0.000062518164,0.000059192294,0.00001991747,0.0004321679,0.0000063920697,0.001974785,0.012627829,0.01889576,0.29415676],"study_design_scores_gemma":[0.0013542962,0.00034425507,0.42518017,0.000050047947,0.000036707243,0.000020636991,0.002297207,0.0037063248,0.0014016273,0.46710184,0.09819328,0.00031358478],"about_ca_topic_score_codex":0.00015068817,"about_ca_topic_score_gemma":0.0006638402,"teacher_disagreement_score":0.45447403,"about_ca_system_score_codex":0.00006474154,"about_ca_system_score_gemma":0.0006000768,"threshold_uncertainty_score":0.9987656},"labels":[],"label_agreement":null},{"id":"W2167816954","doi":"10.2466/pr0.101.3.1001-1010","title":"Four Multi-Item Interrater Agreement Options: Comparisons and Outcomes","year":2007,"lang":"en","type":"article","venue":"Psychological Reports","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Inter-rater reliability; Intraclass correlation; Psychology; Agreement; Statistics; Psychometrics; Clinical psychology; Rating scale; Developmental psychology; Mathematics","score_opus":0.4302605755541647,"score_gpt":0.49161844742747823,"score_spread":0.06135787187331354,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2167816954","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.93371356,0.00016946034,0.05297678,0.002450367,0.0012644104,0.0004450097,0.0000018688861,0.000059074126,0.008919481],"genre_scores_gemma":[0.98401517,0.000022698165,0.012042088,0.0012278622,0.00004853228,0.000021610675,0.0000018976217,0.000005223526,0.0026149428],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99541265,0.00019004707,0.001553884,0.00092016574,0.0015082374,0.0004150136],"domain_scores_gemma":[0.99733084,0.0008487334,0.0004244425,0.00088950334,0.00024338886,0.0002630815],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.011390154,0.00021289283,0.00040713736,0.00015322522,0.00019228266,0.00021518457,0.00032786158,0.000113381786,0.0009971563],"category_scores_gemma":[0.0040592663,0.0001211985,0.0001719268,0.00029320727,0.00020472106,0.00015330513,0.00018821243,0.00024363157,0.00021586369],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000033480603,0.00069023774,0.94709575,0.0000026151254,0.00002532262,0.00030644774,0.00013668256,0.000012825053,0.0006065703,0.00020524328,0.013491679,0.03739312],"study_design_scores_gemma":[0.0002833764,0.0001492317,0.9604405,0.000013071267,0.000010041694,0.00012278077,0.00033981045,0.000072954674,0.00008718921,0.0065269065,0.031784166,0.00016993785],"about_ca_topic_score_codex":0.000013268749,"about_ca_topic_score_gemma":0.000062453335,"teacher_disagreement_score":0.050301597,"about_ca_system_score_codex":0.0000469582,"about_ca_system_score_gemma":0.0000072562852,"threshold_uncertainty_score":0.9999161},"labels":[],"label_agreement":null},{"id":"W2171762906","doi":"10.1016/j.jclinepi.2010.02.006","title":"The COSMIN study reached international consensus on taxonomy, terminology, and definitions of measurement properties for health-related patient-reported outcomes","year":2010,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4442,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Terminology; Taxonomy (biology); Delphi method; Confusion; Delphi; Consistency (knowledge bases); Management science; Medicine; Psychology; Applied psychology; Computer science; Artificial intelligence; Engineering; Linguistics; Ecology","score_opus":0.8170932433584581,"score_gpt":0.5349041107663476,"score_spread":0.28218913259211054,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2171762906","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9132114,0.00016675061,0.0002725753,0.081946425,0.0033409519,0.00089696195,0.000008605174,0.0000051671286,0.00015116922],"genre_scores_gemma":[0.995866,0.000117223084,0.002555033,0.001294475,0.0000896816,0.000046118457,5.359728e-7,0.0000051414136,0.000025807873],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98165023,0.005626863,0.010877745,0.00040365473,0.0011377146,0.00030376832],"domain_scores_gemma":[0.9297979,0.0577233,0.009249949,0.0006048247,0.0024000504,0.00022400494],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.11914837,0.00015795481,0.0016898619,0.00015018943,0.00022250338,0.000025479645,0.000605055,0.0001784293,0.000022042272],"category_scores_gemma":[0.40713936,0.00007446431,0.0004844949,0.000091933776,0.000882184,0.000051908795,0.00010140854,0.00081549096,0.000003880514],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012842019,0.0016583,0.8948935,0.0000067261776,0.0005431209,0.0000048633447,0.00022945566,0.00007202853,0.00010976105,0.0029336,0.010192911,0.088071495],"study_design_scores_gemma":[0.0035620073,0.0060702427,0.8864807,0.000069725815,0.0000838946,0.00006129876,0.001801442,0.00033894306,0.000030487563,0.079168476,0.022213036,0.00011974064],"about_ca_topic_score_codex":0.000035300724,"about_ca_topic_score_gemma":0.00021169493,"teacher_disagreement_score":0.287991,"about_ca_system_score_codex":0.00004022543,"about_ca_system_score_gemma":0.0003749423,"threshold_uncertainty_score":0.9070221},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2177903965","doi":"10.24095/hpcdp.32.1.05","title":"2008 Niday Perinatal Database quality audit: report of a quality assurance project","year":2011,"lang":"en","type":"article","venue":"Chronic diseases and injuries in Canada","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":71,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":true,"ca_institutions":"Ottawa Hospital; Ottawa Public Health; University of Ottawa; Champlain Regional College; Children's Hospital of Eastern Ontario; Ontario Stroke Network","funders":"","keywords":"Intraclass correlation; Quality assurance; Cohen's kappa; Data quality; Kappa; Audit; Database; Statistic; Statistics; Quality (philosophy); Medicine; Sample (material); Reliability (semiconductor); Mathematics; Computer science; Reproducibility; Accounting; Operations management; Physics; Business; Engineering","score_opus":0.12208470535853659,"score_gpt":0.3818689116768931,"score_spread":0.25978420631835647,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2177903965","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9910793,0.0037366266,0.00009229638,0.00028480784,0.00061122904,0.0003712107,0.0010538307,0.0000101802525,0.0027605398],"genre_scores_gemma":[0.99915564,0.0002335047,0.00017609402,0.000094304974,0.000051929022,0.000038995204,0.000016476566,0.0000060926745,0.00022698054],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9955185,0.00039737715,0.0013826698,0.0006457436,0.0017222831,0.0003334135],"domain_scores_gemma":[0.99737406,0.00056290557,0.0005342944,0.0010978964,0.0002666949,0.00016416555],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0031541486,0.00018723932,0.0004965353,0.00008136389,0.00011704403,0.00004446628,0.0005564489,0.000035655426,0.0007542467],"category_scores_gemma":[0.004015033,0.00014166012,0.00008955582,0.00039812224,0.000393236,0.0003789649,0.00024766778,0.00012794494,0.0000021252304],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00024386849,0.0002250375,0.9581692,0.00042763096,0.00003842975,0.00011574382,0.0003875525,0.000013019516,0.00008142735,0.0015438106,0.024003759,0.014750481],"study_design_scores_gemma":[0.00043670478,0.00010083645,0.966229,0.00011986516,0.0000146428765,0.000005176015,0.0015450453,0.000043311127,0.00044689953,0.002274009,0.02852862,0.00025586315],"about_ca_topic_score_codex":0.84291726,"about_ca_topic_score_gemma":0.892978,"teacher_disagreement_score":0.050060745,"about_ca_system_score_codex":0.00054990547,"about_ca_system_score_gemma":0.0068776635,"threshold_uncertainty_score":0.9987524},"labels":[],"label_agreement":null},{"id":"W2185742531","doi":"10.1002/j.0022-0337.2002.66.9.tb03570.x","title":"A Measure of Agreement Between Clinicians and a Computer‐Based Decision Support System for Planning Dental Treatment","year":2002,"lang":"en","type":"article","venue":"Journal of Dental Education","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; TRIUMF","funders":"","keywords":"Measure (data warehouse); Decision support system; Medical physics; Medicine; Psychology; Computer science; Artificial intelligence; Data mining","score_opus":0.19541446516598343,"score_gpt":0.41677564037561216,"score_spread":0.22136117520962872,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2185742531","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9701573,0.00049722014,0.0266108,0.00023144865,0.0017047223,0.0005173173,0.000026065634,0.00000473895,0.00025036416],"genre_scores_gemma":[0.99021316,0.000014307172,0.00914931,0.00005801303,0.00042822107,0.000010813111,0.000004382969,0.000007093026,0.00011470211],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9962471,0.00016400253,0.0015793954,0.0002494949,0.0015973547,0.00016262582],"domain_scores_gemma":[0.9968887,0.00086424703,0.0012574511,0.0002396302,0.0005829292,0.00016703947],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0030986895,0.00015418093,0.000475211,0.00034295573,0.00012595256,0.000119491,0.00031772122,0.00007227067,0.0001108492],"category_scores_gemma":[0.0003356937,0.00010687731,0.00027962003,0.00022120631,0.000051597704,0.00025132304,0.000028195684,0.00008488481,0.000019885585],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028873532,0.0015513139,0.51806855,0.0000753276,0.0001291502,0.000005455231,0.0017769538,0.00048134814,0.0005796968,0.000042811535,0.010067773,0.4669329],"study_design_scores_gemma":[0.023994358,0.029038897,0.7651385,0.0058968766,0.001745925,0.00062646525,0.078670256,0.030058673,0.020189142,0.0020251907,0.04127331,0.0013424249],"about_ca_topic_score_codex":0.000013712077,"about_ca_topic_score_gemma":0.000011708689,"teacher_disagreement_score":0.46559048,"about_ca_system_score_codex":0.00039902897,"about_ca_system_score_gemma":0.0001850393,"threshold_uncertainty_score":0.43583304},"labels":[],"label_agreement":null},{"id":"W2271963175","doi":"10.1302/2046-3758.412.2000433","title":"Spectrum bias, a common unrecognised issue in orthopaedic agreement studies","year":2015,"lang":"en","type":"article","venue":"Bone and Joint Research","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"ZonMw; NuVasive; Achmea; Amgen; Stryker; Sanofi; Eli Lilly and Company","keywords":"Medicine; Radiography; Radiology; Kappa; Population; Nuclear medicine; Orthodontics; Surgery; Mathematics; Geometry","score_opus":0.7498127583013002,"score_gpt":0.527483675786595,"score_spread":0.22232908251470518,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2271963175","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9559276,0.0060004564,0.000024946876,0.025000911,0.00027609337,0.0007910292,0.000004599396,0.000017450624,0.011956885],"genre_scores_gemma":[0.9927696,0.0009962925,0.00023191144,0.00026318972,0.0001722242,0.00006930794,0.0000022372558,0.000007473144,0.00548774],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9916372,0.0015437765,0.0010348733,0.00065373856,0.004500933,0.00062951905],"domain_scores_gemma":[0.99698716,0.0011939404,0.0001215614,0.00064989424,0.0007288771,0.00031857987],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.052325714,0.00016207283,0.0005524143,0.00080216664,0.00021546918,0.00026533162,0.00028644115,0.000071395145,0.0003532278],"category_scores_gemma":[0.008576917,0.000106767235,0.000088451954,0.0012965485,0.00038291,0.00021035004,0.00059861917,0.00047646716,0.0008409419],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005543593,0.0011547495,0.3652088,0.00013974847,0.00009103648,0.00035865302,0.010044583,0.00008872539,0.00093592657,0.0017716056,0.4228255,0.19682634],"study_design_scores_gemma":[0.005462528,0.0024964313,0.16982065,0.0005181017,0.000018111308,0.000026602318,0.06112365,0.0011850673,0.002990784,0.3351632,0.42056066,0.000634197],"about_ca_topic_score_codex":0.00064779364,"about_ca_topic_score_gemma":0.003275117,"teacher_disagreement_score":0.3333916,"about_ca_system_score_codex":0.00020208674,"about_ca_system_score_gemma":0.00020175388,"threshold_uncertainty_score":0.999937},"labels":[],"label_agreement":null},{"id":"W2314842795","doi":"10.1002/j.1538-9235.2001.tb03567.x","title":"INTER-RATER RELIABILITY AMONG OPTOMETRY CLINICAL INSTRUCTORS.","year":2001,"lang":"en","type":"article","venue":"Optometry and Vision Science","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Optometry; Cornea; Reliability (semiconductor); Medicine; Inter-rater reliability; Ophthalmology; Psychology; Rating scale","score_opus":0.10253704012357771,"score_gpt":0.5337989571076218,"score_spread":0.4312619169840441,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2314842795","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9850471,0.00007891847,0.004927035,0.00088118133,0.0018934229,0.0002907582,0.0000050628505,0.000056919034,0.0068195993],"genre_scores_gemma":[0.99553555,0.00012969795,0.0026418779,0.00045631672,0.00015263655,0.000006891742,8.162195e-7,0.0000079010815,0.0010682957],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99110234,0.0005556046,0.0017112403,0.0018609597,0.0041070106,0.00066286477],"domain_scores_gemma":[0.9944043,0.0017413313,0.00041395068,0.0017604901,0.0010129208,0.0006670069],"candidate_categories":["metaresearch","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.040964995,0.00028332134,0.0005499993,0.0010963958,0.0008103185,0.0013440665,0.0021584139,0.00018283437,0.0011172278],"category_scores_gemma":[0.014800743,0.00018218809,0.00020719942,0.0072733904,0.003645649,0.0024598474,0.0011117808,0.00049623527,0.00037872678],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000054134114,0.00017527705,0.8988363,0.0000062594922,0.00000340588,0.000004075006,0.000098639444,0.000010775855,0.0005412746,0.00006074087,0.0006458748,0.09956321],"study_design_scores_gemma":[0.00048445276,0.00029197344,0.97358483,0.000053196396,0.000007025841,0.000011006558,0.0004972236,0.002091033,0.0006075664,0.0020932064,0.020013593,0.00026491113],"about_ca_topic_score_codex":0.000016059917,"about_ca_topic_score_gemma":0.0000019137792,"teacher_disagreement_score":0.0992983,"about_ca_system_score_codex":0.00009412441,"about_ca_system_score_gemma":0.00011474219,"threshold_uncertainty_score":0.9997959},"labels":[],"label_agreement":null},{"id":"W2317471821","doi":"10.1021/es301320n","title":"Negative Consequences of Using α = 0.05 for Environmental Monitoring Decisions: A Case Study from a Decade of Canada’s Environmental Effects Monitoring Program","year":2012,"lang":"en","type":"article","venue":"Environmental Science & Technology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Canadian Water Network; University of New Brunswick","funders":"Natural Resources Canada; Canadian Forest Service; Natural Sciences and Engineering Research Council of Canada; Canadian Rivers Institute, University of New Brunswick","keywords":"Type I and type II errors; Statistics; Null hypothesis; Set (abstract data type); Econometrics; Approximation error; Mathematics; Computer science","score_opus":0.08694077720667744,"score_gpt":0.35924353480984633,"score_spread":0.2723027576031689,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2317471821","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99536836,0.00080055976,0.00039151072,0.000054688735,0.00085345964,0.00234606,0.00013809548,0.000028980257,0.000018265484],"genre_scores_gemma":[0.98605424,0.000036239828,0.013558124,0.000008037759,0.00007915326,0.00022421643,0.0000016299784,0.00001841823,0.000019914967],"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.99358976,0.00021408858,0.0012394339,0.001010703,0.0029965558,0.0009494719],"domain_scores_gemma":[0.9964358,0.0014897282,0.00078564044,0.0009635261,0.000018395354,0.00030687984],"candidate_categories":["metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.0030205152,0.00038464443,0.00064346433,0.00056638627,0.00077272294,0.0000533508,0.0014606252,0.0001670629,0.00007545642],"category_scores_gemma":[0.001201431,0.0003180545,0.00014243012,0.0007918898,0.003510987,0.0007411965,0.0008218723,0.00027746268,0.000005241747],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000028440963,0.001080395,0.5689798,0.0000023361604,0.000031766827,0.00004036765,0.0010301205,0.00016469082,0.38626134,0.00000560634,0.0000030863987,0.04237203],"study_design_scores_gemma":[0.0009028293,0.00083893235,0.22617745,0.000062324616,0.00007946476,0.0001261302,0.07505595,0.00022401614,0.69533074,0.00072673603,0.00014642297,0.00032899008],"about_ca_topic_score_codex":0.010080098,"about_ca_topic_score_gemma":0.00090910273,"teacher_disagreement_score":0.34280235,"about_ca_system_score_codex":0.0015487883,"about_ca_system_score_gemma":0.00022735717,"threshold_uncertainty_score":0.99992716},"labels":[],"label_agreement":null},{"id":"W2319535879","doi":"10.1097/jsm.0000000000000047","title":"Interrater Agreement of an Observational Tool to Code Knockouts and Technical Knockouts in Mixed Martial Arts","year":2013,"lang":"en","type":"article","venue":"Clinical Journal of Sport Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital; University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Inter-rater reliability; Martial arts; Situational ethics; Context (archaeology); Gene knockout; Applied psychology; Medicine; Psychology; Statistics; Social psychology; Mathematics; Developmental psychology; Genetics; History","score_opus":0.38690116381851114,"score_gpt":0.48134862574224013,"score_spread":0.09444746192372899,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2319535879","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9812868,0.00007088603,0.0005178568,0.016250744,0.0011035176,0.0003963994,0.000003013247,0.000004172493,0.00036656388],"genre_scores_gemma":[0.99307555,0.000039660295,0.00394829,0.0022105395,0.00045061467,0.0000075048806,0.0000018891245,0.0000066501098,0.00025931527],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99215317,0.00013268954,0.004678429,0.00039613483,0.0023901856,0.00024940041],"domain_scores_gemma":[0.9960827,0.00073952304,0.0010657886,0.00047224926,0.0011927672,0.0004469627],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.016712306,0.00017617414,0.0009830006,0.00031424357,0.000035381378,0.00003798989,0.00070032617,0.00013885157,0.0023826389],"category_scores_gemma":[0.0044948393,0.000103316415,0.00015347489,0.00039202932,0.00036532813,0.0004061471,0.00017150653,0.00043193804,0.00003546976],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00060786825,0.0009162224,0.8984547,0.000015820757,0.000037203572,0.0000615755,0.00028154763,0.000101631755,0.006688641,0.0002560887,0.049227346,0.043351315],"study_design_scores_gemma":[0.0019109446,0.0024502892,0.9725328,0.0004038092,0.000029880814,0.000040849154,0.00040603054,0.0001722112,0.0003554888,0.0069023957,0.014673476,0.00012180777],"about_ca_topic_score_codex":0.000030398704,"about_ca_topic_score_gemma":0.00010534096,"teacher_disagreement_score":0.074078076,"about_ca_system_score_codex":0.00005236594,"about_ca_system_score_gemma":0.00012868606,"threshold_uncertainty_score":0.9985293},"labels":[],"label_agreement":null},{"id":"W2353892345","doi":"10.1111/hir.12140","title":"Inter‐rater reliability of h‐index scores calculated by Web of Science and Scopus for clinical epidemiology scientists","year":2016,"lang":"en","type":"article","venue":"Health Information & Libraries Journal","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":42,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Ottawa Hospital","funders":"Ottawa Hospital Research Institute","keywords":"Scopus; Rank correlation; Bibliometrics; Reliability (semiconductor); Web of science; Statistics; Spearman's rank correlation coefficient; Index (typography); Inter-rater reliability; Psychology; Medicine; Meta-analysis; MEDLINE; Mathematics; Library science; Computer science; World Wide Web; Physics; Political science; Internal medicine","score_opus":0.25026763577986333,"score_gpt":0.4679676814020811,"score_spread":0.21770004562221779,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2353892345","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.85069484,0.00053175533,0.11872404,0.026802642,0.002076392,0.0007543032,0.00015313188,0.000017112568,0.00024576194],"genre_scores_gemma":[0.99257123,0.00039045129,0.0051541114,0.0017712796,0.00004840301,0.0000068382164,0.0000028003192,0.0000030694982,0.00005183873],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9921054,0.00088849367,0.0047260188,0.00029405163,0.0015316927,0.0004543304],"domain_scores_gemma":[0.9893441,0.004225103,0.0029695013,0.00049583934,0.0025534,0.0004120151],"candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07015967,0.0001262578,0.0006458722,0.0004525889,0.0004965671,0.00019522825,0.0007578423,0.0001006568,0.00014651768],"category_scores_gemma":[0.051510133,0.000066812005,0.00012599025,0.0007205642,0.00366928,0.004899823,0.0002041469,0.00018572145,0.0000099882745],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004954676,0.00009902647,0.70976585,0.00016124094,0.000012446481,8.478492e-8,0.0006958957,0.000036344307,0.00018967861,0.0055535347,0.09810676,0.18488367],"study_design_scores_gemma":[0.0042908126,0.002349595,0.5953082,0.0007516743,0.000012918284,0.000039861945,0.0010573068,0.010717493,0.0022141205,0.10604245,0.27689672,0.00031883764],"about_ca_topic_score_codex":0.000013967742,"about_ca_topic_score_gemma":0.0000028769448,"teacher_disagreement_score":0.18456483,"about_ca_system_score_codex":0.00010466871,"about_ca_system_score_gemma":0.002312075,"threshold_uncertainty_score":0.99904215},"labels":[],"label_agreement":null},{"id":"W2414698452","doi":"","title":"Comparability of self-reported medication use and pharmacy claims data.","year":2013,"lang":"en","type":"article","venue":"PubMed","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Medicine; Pharmacy; Kappa; Family medicine; Logistic regression; Comparability; Pharmacoepidemiology; Cohen's kappa; Medicare Part D; Medical prescription; Prescription drug; Statistics; Internal medicine; Nursing","score_opus":0.4920086509050939,"score_gpt":0.39880317338084575,"score_spread":0.09320547752424813,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2414698452","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9950074,0.00007650566,0.0004100088,0.00246365,0.00023951883,0.0011956891,0.000015547175,0.000034560817,0.0005570945],"genre_scores_gemma":[0.9986411,0.000047994097,0.0007482408,0.00015367194,0.00003432589,0.00026739182,0.000015421378,0.0000030655544,0.00008884127],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9966373,0.00028928445,0.0008746634,0.00051126286,0.0014778429,0.0002096503],"domain_scores_gemma":[0.996591,0.0009700966,0.0003778999,0.0013952173,0.00048209704,0.00018373101],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008088523,0.000092248105,0.00022960306,0.00007539456,0.00006511166,0.00016930404,0.00071624696,0.000031414558,0.00030814865],"category_scores_gemma":[0.0039055014,0.000062356,0.000032503907,0.00028444204,0.00016896235,0.0010898577,0.00025636484,0.000089678375,0.00004350272],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000014126977,0.00026900714,0.7961787,0.000014780076,0.000055580746,3.4970424e-7,0.00014319562,0.0000064336405,0.00018423825,0.000071858085,0.020592634,0.18246908],"study_design_scores_gemma":[0.00037965868,0.000005709098,0.97160476,0.000002356586,0.000030857547,0.0000013744684,0.000054571694,0.0025495847,0.00050139084,0.00625074,0.018549228,0.000069785034],"about_ca_topic_score_codex":0.00030727612,"about_ca_topic_score_gemma":0.000023390505,"teacher_disagreement_score":0.1823993,"about_ca_system_score_codex":0.000028417999,"about_ca_system_score_gemma":0.000034208355,"threshold_uncertainty_score":0.4675532},"labels":[],"label_agreement":null},{"id":"W2463810109","doi":"","title":"Use of the limits of agreement approach in periodontology.","year":2007,"lang":"en","type":"article","venue":"PubMed","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Addiction and Mental Health","funders":"","keywords":"Medicine; Periodontology; Cohen's kappa; Statistics; Correlation coefficient; Kappa; Standard deviation; Linear regression; Correlation; Limits of agreement; Standard error; Pearson product-moment correlation coefficient; Sample size determination; Medical physics; Dentistry; Orthodontics; Mathematics; Nuclear medicine","score_opus":0.3531444023639929,"score_gpt":0.3407668189419939,"score_spread":0.012377583421999017,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2463810109","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98915434,0.000095537594,0.0006640723,0.00061275426,0.00027053902,0.0008288336,0.0000032826663,0.0000032041048,0.008367447],"genre_scores_gemma":[0.9985809,0.0000062320037,0.0005979604,0.00011112112,0.000015442829,0.000078822064,2.6816826e-7,0.0000022933639,0.0006069734],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99684155,0.00024044211,0.0008946995,0.00026469643,0.0014672587,0.00029134992],"domain_scores_gemma":[0.998149,0.0006216566,0.00031077446,0.0006674806,0.00019817546,0.00005291747],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010772531,0.00008143374,0.00024483094,0.00014514687,0.000033533448,0.00002399153,0.00068839896,0.00005988697,0.000033707234],"category_scores_gemma":[0.0027769308,0.000044753964,0.00012017135,0.000689856,0.0001971056,0.00012112186,0.00011429856,0.00010284097,0.0000033907013],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000059210564,0.00036018583,0.833321,0.000012026448,0.000011948768,5.4353035e-7,0.0005662621,0.0003158784,0.00028442187,0.0009732719,0.0007258304,0.16336945],"study_design_scores_gemma":[0.00020541965,0.000013723833,0.99011284,0.000005302317,0.0000052345795,5.828445e-7,0.0003870694,0.000056555316,0.0038777927,0.00095818547,0.004329494,0.000047798167],"about_ca_topic_score_codex":0.00008385084,"about_ca_topic_score_gemma":0.00031084,"teacher_disagreement_score":0.16332166,"about_ca_system_score_codex":0.000050830204,"about_ca_system_score_gemma":0.000022636379,"threshold_uncertainty_score":0.37335658},"labels":[],"label_agreement":null},{"id":"W2467932880","doi":"10.1177/0013164416654740","title":"An Unbiased Estimate of Global Interrater Agreement","year":2016,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université du Québec à Trois-Rivières; University of Ottawa","funders":"","keywords":"Inter-rater reliability; Statistics; Agreement; Econometrics; Psychology; Mathematics; Rating scale; Linguistics","score_opus":0.40228873007278076,"score_gpt":0.47602807594491,"score_spread":0.07373934587212921,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2467932880","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9668057,0.00016877758,0.0062217433,0.02041754,0.0008634564,0.0002825336,0.000031291645,0.000014046925,0.005194914],"genre_scores_gemma":[0.9975572,0.000011789124,0.0016178658,0.00047623247,0.00012474941,0.000040532104,0.0000017642567,0.00000279572,0.00016704097],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99566996,0.00036404424,0.00076895545,0.00063515257,0.0023242258,0.00023767653],"domain_scores_gemma":[0.99799365,0.00023398563,0.00022089656,0.0005532801,0.00077174685,0.00022644755],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00468477,0.00016850828,0.00024208507,0.000048808477,0.000097328986,0.0000705069,0.00053024787,0.00006647132,0.0040821107],"category_scores_gemma":[0.0012375994,0.0000809371,0.00008718234,0.00023945885,0.00024096154,0.0001792417,0.000049381568,0.000053947435,0.00022056495],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004415165,0.0051180273,0.60805285,0.000017311946,0.0000770604,0.0000015019536,0.00024916933,0.000024598812,0.060346335,0.030824706,0.026554117,0.26829278],"study_design_scores_gemma":[0.00054767285,0.0005817458,0.8357299,0.00006297764,0.000010246678,0.0000039890397,0.000090140544,0.000010932419,0.00066665065,0.15925543,0.0029006568,0.00013966886],"about_ca_topic_score_codex":0.000012254346,"about_ca_topic_score_gemma":0.000025491332,"teacher_disagreement_score":0.2681531,"about_ca_system_score_codex":0.00013589315,"about_ca_system_score_gemma":0.000059725386,"threshold_uncertainty_score":0.9968283},"labels":[],"label_agreement":null},{"id":"W2470750097","doi":"","title":"Enhancing retrieval of best evidence for health care from bibliographic databases: calibration of the hand search of the literature.","year":2001,"lang":"en","type":"article","venue":"PubMed","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":102,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McMaster University; Hamilton Health Sciences","funders":"","keywords":"Cohen's kappa; Statistic; Information retrieval; Reliability (semiconductor); Computer science; MEDLINE; Health care; Test (biology); Medical education; Database; Medicine; Statistics; Machine learning; Mathematics","score_opus":0.267435042832476,"score_gpt":0.40163266319868807,"score_spread":0.13419762036621208,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2470750097","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98791325,0.0048234034,0.0023883015,0.0025924118,0.00037428577,0.0015795544,0.0002995568,0.000002784444,0.000026446394],"genre_scores_gemma":[0.9990651,0.00040759478,0.00021164426,0.00013253164,0.000056295077,0.000059694034,0.0000065638524,0.0000037732327,0.000056781406],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9962166,0.00056338316,0.00079050206,0.0002990311,0.0019146579,0.00021582984],"domain_scores_gemma":[0.99630713,0.001481035,0.00047358306,0.000929822,0.00075130013,0.000057151083],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00624426,0.00008429028,0.00024160469,0.00038130768,0.00017743364,0.00009243298,0.0008583294,0.000039462564,0.000016928392],"category_scores_gemma":[0.00499357,0.000041403833,0.00019494977,0.005210837,0.00023314013,0.00034987624,0.00016919602,0.00011317113,2.2637106e-7],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010877064,0.00040577233,0.91260666,0.00069135666,0.000084602434,3.4458512e-7,0.013511912,0.00052038766,0.027224293,0.00027642207,0.00419822,0.039392333],"study_design_scores_gemma":[0.00053405913,0.00011281184,0.7333352,0.0008832099,0.00003812107,5.043844e-7,0.002758601,0.00017895969,0.25928566,0.0014472824,0.0013291436,0.00009647669],"about_ca_topic_score_codex":0.00040083146,"about_ca_topic_score_gemma":0.0010007661,"teacher_disagreement_score":0.23206136,"about_ca_system_score_codex":0.000030192277,"about_ca_system_score_gemma":0.00015936416,"threshold_uncertainty_score":0.597813},"labels":[],"label_agreement":null},{"id":"W2500917129","doi":"10.1016/j.diii.2016.05.014","title":"Agreement studies in radiology research","year":2016,"lang":"en","type":"article","venue":"Diagnostic and Interventional Imaging","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":48,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta Hospital; Université de Montréal; Hôpital Notre-Dame; Health Sciences Centre; Centre Hospitalier de l’Université de Montréal","funders":"","keywords":"Medicine; Confidence interval; Medical physics; Reliability (semiconductor); Sample size determination; Nuclear medicine; Statistics; Internal medicine","score_opus":0.512024444147994,"score_gpt":0.5438569514778593,"score_spread":0.03183250732986531,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2500917129","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9513464,0.01385302,0.0044255694,0.026109155,0.00091881165,0.00036203928,0.000011901851,0.000013842257,0.0029592519],"genre_scores_gemma":[0.997516,0.0006429974,0.00015855124,0.00015738037,0.00008025744,0.000057950638,7.861142e-7,0.000003096834,0.0013829825],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99694633,0.0006546952,0.0006305856,0.00043927753,0.0010368827,0.00029224882],"domain_scores_gemma":[0.9868995,0.012226624,0.00007974542,0.00023562787,0.0004949858,0.00006354898],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.011372222,0.00008949568,0.00019705863,0.00033965235,0.00011983747,0.0000683927,0.00029862992,0.000020236921,0.0008703622],"category_scores_gemma":[0.030501973,0.000049245973,0.00008179222,0.0002554522,0.0005667501,0.0002633537,0.00028447722,0.00010792196,0.00031599507],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000022140122,0.0002192602,0.75686044,0.000018524657,0.000029327508,0.000038816528,0.00038446352,0.000005895562,0.00047431883,0.01737137,0.034824718,0.18975075],"study_design_scores_gemma":[0.0010573203,0.00018347813,0.51717293,0.0010541221,0.000006353092,0.000022649703,0.0030781014,0.00012189906,0.00028733973,0.4556624,0.021195803,0.00015757504],"about_ca_topic_score_codex":0.000026229394,"about_ca_topic_score_gemma":0.00004732952,"teacher_disagreement_score":0.43829104,"about_ca_system_score_codex":0.00011854555,"about_ca_system_score_gemma":0.000022003826,"threshold_uncertainty_score":0.97766453},"labels":[],"label_agreement":null},{"id":"W2525241236","doi":"10.5210/ojphi.v8i2.6720","title":"Development and Validation of a Standardized Tool for Prioritization of Information Sources","year":2016,"lang":"en","type":"article","venue":"Online Journal of Public Health Informatics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Agency of Canada; Canadian Food Inspection Agency","funders":"Canadian Food Inspection Agency","keywords":"Concordance; Cohen's kappa; Kappa; Prioritization; Reliability (semiconductor); Macro; Medicine; Data mining; Medical physics; Computer science; Statistics; Management science; Mathematics; Machine learning; Engineering","score_opus":0.18496991497549503,"score_gpt":0.40245150118308376,"score_spread":0.21748158620758873,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2525241236","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5285797,0.000029282242,0.4683606,0.00262558,0.00010244972,0.00022444433,0.000054501776,0.0000022774636,0.00002113711],"genre_scores_gemma":[0.5787143,0.00032588223,0.4204844,0.0003509531,0.00006786989,0.0000047175563,0.000021163409,0.0000045647853,0.000026129419],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9935278,0.00013350924,0.004394292,0.00004101374,0.0017328534,0.00017050075],"domain_scores_gemma":[0.99146646,0.0006051052,0.0045430954,0.00014987639,0.0031124812,0.000123004],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02260041,0.000081256694,0.0003842734,0.0004554753,0.00007990259,0.000083865576,0.0002469264,0.00004958959,0.000020170743],"category_scores_gemma":[0.009072263,0.000045195018,0.000067336776,0.00030390354,0.0000566872,0.0021795256,0.00004169061,0.00005546543,9.660585e-7],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007064148,0.00011219369,0.0091561945,0.0004343507,0.000033105796,1.7718936e-8,0.009837415,0.0001340048,0.0001061122,0.0010196611,0.0010613959,0.9780349],"study_design_scores_gemma":[0.007920998,0.001863623,0.030054543,0.0011208515,0.000028696648,0.000019257,0.018503182,0.0031103033,0.009714631,0.0069137085,0.9204611,0.00028910808],"about_ca_topic_score_codex":0.0000011871196,"about_ca_topic_score_gemma":0.0000024779597,"teacher_disagreement_score":0.9777458,"about_ca_system_score_codex":0.00013142233,"about_ca_system_score_gemma":0.0013971757,"threshold_uncertainty_score":0.99927473},"labels":[],"label_agreement":null},{"id":"W2556459859","doi":"","title":"A comparison of antibiotic disks from different sources on Quicolor and Mueller-Hinton agar media in evaluation of antibacterial susceptibility testing.","year":2016,"lang":"en","type":"article","venue":"PubMed","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Economics","funders":"","keywords":"Agar; Agar diffusion test; Antibiotics; Microbiology; Enterobacter; Shigella; Biology; Antibacterial activity; Escherichia coli; Bacteria; Salmonella","score_opus":0.36786156895707584,"score_gpt":0.3919600067517543,"score_spread":0.024098437794678462,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2556459859","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99787945,0.00008375485,0.000029434097,0.0003081119,0.00044161573,0.0009791743,0.00004213095,0.0000090553385,0.00022728014],"genre_scores_gemma":[0.9998064,0.00000711434,0.00006656554,0.000008449443,0.0000587654,0.000036610927,0.0000034842615,0.0000048708102,0.00000771859],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99491155,0.00092855876,0.0010937537,0.0005419251,0.0022789582,0.00024525227],"domain_scores_gemma":[0.9945413,0.0038853628,0.0005627517,0.00052381604,0.00039119498,0.00009559384],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.00883062,0.0001528637,0.0005510908,0.00017091232,0.00003775462,0.000044181772,0.00034931148,0.000087104345,0.00013510068],"category_scores_gemma":[0.016436605,0.00008268746,0.000066793094,0.00024882855,0.00027600353,0.0001250827,0.00011555786,0.00007675522,0.0000049232044],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016764668,0.0004498235,0.71699613,0.000009310288,0.000009002759,8.85964e-8,0.00060310034,0.000022343984,0.018453583,0.000006417029,0.000029941537,0.2632526],"study_design_scores_gemma":[0.0013196679,0.00012877613,0.93559366,0.000081863174,0.00003096076,4.3924175e-8,0.00028176385,0.00054568547,0.05867148,0.0032372503,0.000014823441,0.000094026946],"about_ca_topic_score_codex":0.00019313193,"about_ca_topic_score_gemma":0.0010132642,"teacher_disagreement_score":0.26315856,"about_ca_system_score_codex":0.00010075331,"about_ca_system_score_gemma":0.000029373065,"threshold_uncertainty_score":0.99184835},"labels":[],"label_agreement":null},{"id":"W2580371235","doi":"10.1177/0146621616684584","title":"An Evaluation of Interrater Reliability Measures on Binary Tasks Using <i>d-Prime</i>","year":2016,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland","funders":"","keywords":"Inter-rater reliability; Kappa; Prime (order theory); Psychology; Statistics; Reliability (semiconductor); Cohen's kappa; Binary number; Agreement; Psychometrics; Social psychology; Mathematics; Combinatorics; Arithmetic; Rating scale; Linguistics","score_opus":0.5204254440661797,"score_gpt":0.4693431790011876,"score_spread":0.05108226506499214,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2580371235","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96720684,0.00007349334,0.017677976,0.0012774194,0.0008208781,0.0014641808,0.000016154849,0.00007133694,0.011391706],"genre_scores_gemma":[0.9975315,0.000009064229,0.0014612873,0.0006964535,0.00012401963,0.00014329534,0.0000015918083,0.000017438788,0.000015391912],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.98265845,0.0021804634,0.0017171813,0.0016102736,0.011309546,0.0005240708],"domain_scores_gemma":[0.99381554,0.00044840586,0.00060400134,0.0024042442,0.0024629426,0.0002648791],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.051340144,0.00039420204,0.0006055025,0.0002234392,0.00018801454,0.00009159016,0.0013565761,0.00024389292,0.001702357],"category_scores_gemma":[0.0034151927,0.00019960388,0.00024347505,0.00056700216,0.00039500234,0.00022629647,0.000107899,0.00023449621,0.00030538128],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00080110953,0.002245128,0.0036005555,0.000005222885,0.000037854956,7.9126335e-7,0.00013777122,0.0013157636,0.77225673,0.0006300569,0.0021369685,0.21683203],"study_design_scores_gemma":[0.010008819,0.0051908647,0.31831816,0.00042442497,0.0003987158,0.0000070522187,0.0008105951,0.0020375578,0.38310438,0.2652449,0.012669214,0.0017853181],"about_ca_topic_score_codex":0.000010817869,"about_ca_topic_score_gemma":0.0000076647375,"teacher_disagreement_score":0.38915235,"about_ca_system_score_codex":0.0005671026,"about_ca_system_score_gemma":0.00009329298,"threshold_uncertainty_score":0.99921024},"labels":[],"label_agreement":null},{"id":"W2592866053","doi":"10.1186/s13643-017-0441-7","title":"Effect of standardized training on the reliability of the Cochrane risk of bias assessment tool: a prospective study","year":2017,"lang":"en","type":"article","venue":"Systematic Reviews","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":65,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta; St. Michael's Hospital; University of Toronto","funders":"","keywords":"Medicine; Kappa; Reliability (semiconductor); Physical therapy; Randomized controlled trial; Cohen's kappa; Risk assessment; MEDLINE; Clinical psychology; Statistics; Surgery","score_opus":0.23073563764269314,"score_gpt":0.4609718262253066,"score_spread":0.23023618858261344,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2592866053","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96999407,0.00041190436,0.00036850167,0.00025651456,0.00040604977,0.025762752,0.000029415218,0.000004742484,0.0027660292],"genre_scores_gemma":[0.9987981,0.00010199792,0.00014760908,0.000009002571,0.000016861795,0.00083200366,1.0199758e-7,0.0000064271444,0.000087942455],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9651013,0.025180621,0.00429227,0.0005036243,0.004734975,0.00018720925],"domain_scores_gemma":[0.96891856,0.016641276,0.008014901,0.0055269925,0.00086327095,0.000034987275],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.23339024,0.00026013103,0.0029606253,0.000052009305,0.00040666413,0.00011574498,0.002209717,0.000060807382,0.00008590473],"category_scores_gemma":[0.19908556,0.00008028286,0.0008971494,0.00032057584,0.00037379874,0.00011469319,0.0002303493,0.0002838175,0.000012949934],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005409111,0.0017808733,0.85516495,0.10447537,0.00079047185,0.0000012765254,0.021434106,0.0003538141,0.0010404577,0.0027621412,0.0010270724,0.010628542],"study_design_scores_gemma":[0.006353878,0.009482225,0.8459819,0.09302914,0.0026276833,0.0000034103275,0.015129975,0.0012669215,0.009308466,0.01538805,0.000807728,0.0006206254],"about_ca_topic_score_codex":0.000060713635,"about_ca_topic_score_gemma":0.000028306555,"teacher_disagreement_score":0.03430468,"about_ca_system_score_codex":0.000099411904,"about_ca_system_score_gemma":0.000116720454,"threshold_uncertainty_score":0.8076609},"labels":[],"label_agreement":null},{"id":"W2593472913","doi":"10.1002/cyto.b.21522","title":"Choosing a new CD4 technology: Can statistical method comparison tools influence the decision?","year":2017,"lang":"en","type":"article","venue":"Cytometry Part B Clinical Cytometry","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Grand Challenges Canada; Thailand Research Fund","keywords":"Concordance; Concordance correlation coefficient; Similarity (geometry); Bland–Altman plot; Sensitivity (control systems); Standard deviation; Statistics; Accuracy and precision; Computer science; Data mining; Mathematics; Sample (material); Correlation coefficient; Artificial intelligence; Limits of agreement; Nuclear medicine; Medicine; Chemistry; Image (mathematics); Engineering","score_opus":0.4041854619677802,"score_gpt":0.568325344249743,"score_spread":0.1641398822819628,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2593472913","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.61388564,0.0005106326,0.36299983,0.015519374,0.003190836,0.0010182385,0.00014046053,0.0002065943,0.002528429],"genre_scores_gemma":[0.9307634,0.000086794615,0.06557327,0.001891891,0.00077693636,0.00003834295,0.0000069530392,0.00003609415,0.0008262706],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.985503,0.0014868027,0.0044699796,0.0021380505,0.0052195177,0.0011826336],"domain_scores_gemma":[0.9494785,0.03879601,0.002257835,0.007229259,0.001182675,0.0010557408],"candidate_categories":["metaresearch","metaepi_narrow","sts","scholarly_communication","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.038128756,0.00058619113,0.0017648635,0.0010792447,0.0023970471,0.0033728697,0.008015796,0.0007390038,0.0012922408],"category_scores_gemma":[0.27549097,0.00035886702,0.00059852964,0.003751575,0.0020190647,0.00087890413,0.0022353414,0.0022356762,0.0019797396],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012559349,0.0002980807,0.5787051,0.0000064453457,0.000077743505,0.000021018699,0.000044718716,0.00009696701,0.000110274304,0.0013442851,0.030451883,0.3887179],"study_design_scores_gemma":[0.0014695511,0.0004464608,0.77737707,0.00014929287,0.00011512213,0.00001858784,0.0005653539,0.0010705412,0.0006623944,0.06177862,0.15578364,0.0005633928],"about_ca_topic_score_codex":0.00008844706,"about_ca_topic_score_gemma":0.00008900638,"teacher_disagreement_score":0.3881545,"about_ca_system_score_codex":0.00019392178,"about_ca_system_score_gemma":0.0005763629,"threshold_uncertainty_score":0.99988633},"labels":[],"label_agreement":null},{"id":"W2606094999","doi":"10.1186/s12889-017-4259-y","title":"The development and validation of an instrument to measure the quality of health research reports in the lay media","year":2017,"lang":"en","type":"article","venue":"BMC Public Health","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":31,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University; Impact; McMaster University","funders":"QIMR Berghofer Medical Research Institute","keywords":"Construct validity; Face validity; Public health; Reliability (semiconductor); Content validity; Biostatistics; Quality (philosophy); Validity; Medicine; Scale (ratio); Applied psychology; Psychology; Psychometrics; Nursing; Clinical psychology","score_opus":0.7828744978122634,"score_gpt":0.5599816879127737,"score_spread":0.22289280989948967,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2606094999","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8772449,0.00017202874,0.0004154862,0.12016657,0.000211107,0.0012300281,0.0000039613205,0.000003734243,0.0005521886],"genre_scores_gemma":[0.9983169,0.00003741974,0.0010663109,0.00044049334,0.00003379773,0.00007429422,0.0000025387142,0.0000035398932,0.000024726389],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.94759065,0.025552683,0.0062855813,0.0011811772,0.018028615,0.0013612697],"domain_scores_gemma":[0.9905643,0.0037959362,0.0015652416,0.002551807,0.0012773616,0.00024537777],"candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.44134077,0.00009942904,0.00033244523,0.00015096362,0.0018826873,0.0005507534,0.0016186911,0.00004334353,0.000010981528],"category_scores_gemma":[0.04539233,0.000041531628,0.00004034035,0.0005298954,0.00035834967,0.0002943289,0.00030027563,0.0002632071,0.0000030582016],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004457605,0.00036284898,0.33095258,0.00008688831,0.000009254072,3.3964963e-7,0.045788374,0.000020481224,0.000022368666,0.0071997126,0.002906834,0.61260575],"study_design_scores_gemma":[0.00018203173,0.0001422458,0.94400495,0.000043191365,3.9915798e-7,0.0000013288509,0.016504815,0.000015711523,0.00012775598,0.0070330286,0.031897306,0.00004725549],"about_ca_topic_score_codex":0.0025674715,"about_ca_topic_score_gemma":0.021820305,"teacher_disagreement_score":0.61305237,"about_ca_system_score_codex":0.0002932699,"about_ca_system_score_gemma":0.0040032133,"threshold_uncertainty_score":0.9994167},"labels":[],"label_agreement":null},{"id":"W2606619291","doi":"10.1177/1536867x1701700111","title":"Biasplot: A Package to Effective Plots to Assess Bias and Precision in Method Comparison Studies","year":2017,"lang":"en","type":"article","venue":"The Stata Journal Promoting communications on statistics and Stata","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Plot (graphics); Statistics; Computer science; Accuracy and precision; Data mining; Scatter plot; Mathematics","score_opus":0.6425845793989605,"score_gpt":0.5792506941073572,"score_spread":0.06333388529160333,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2606619291","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5223638,0.0011364118,0.44999245,0.02234395,0.00035695065,0.0023215166,0.0008434519,0.000028391045,0.0006130842],"genre_scores_gemma":[0.8300692,0.0006537357,0.16893739,0.00019304384,0.000025015761,0.000046765254,0.000004582099,0.000014833595,0.000055455614],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99409956,0.0028085231,0.0010820801,0.00045209928,0.0012265107,0.00033124723],"domain_scores_gemma":[0.9796349,0.015700003,0.0007480275,0.0028999741,0.00076313945,0.0002539813],"candidate_categories":["metaresearch","sts","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.026647221,0.00022128438,0.0005333433,0.00029041362,0.0029311785,0.0018429703,0.0027248752,0.00003660729,0.000006908124],"category_scores_gemma":[0.03660919,0.00013347526,0.000036129008,0.00029474712,0.0003013283,0.0002950076,0.0022109603,0.0006609109,0.000024400506],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00027375025,0.00044503552,0.024652958,0.000050181756,0.00014107865,0.000015234707,0.03486708,0.00049314974,0.00039569798,0.01731669,0.008134353,0.9132148],"study_design_scores_gemma":[0.0025626791,0.0033725563,0.619732,0.0022483554,0.00018245967,0.00007171032,0.05088929,0.07113797,0.0005865076,0.2258365,0.022459533,0.0009204246],"about_ca_topic_score_codex":0.00012077231,"about_ca_topic_score_gemma":0.0007499783,"teacher_disagreement_score":0.9122944,"about_ca_system_score_codex":0.00012491288,"about_ca_system_score_gemma":0.000060482238,"threshold_uncertainty_score":0.9991932},"labels":[],"label_agreement":null},{"id":"W2613103431","doi":"10.1177/0962280217702540","title":"Comparing heteroscedastic measurement systems with the probability of agreement","year":2017,"lang":"en","type":"article","venue":"Statistical Methods in Medical Research","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Actua; University of Waterloo","funders":"","keywords":"Homoscedasticity; Heteroscedasticity; Metric (unit); Observational error; Computer science; Statistics; Mathematics; Engineering","score_opus":0.7493861067260326,"score_gpt":0.6289003428450651,"score_spread":0.12048576388096743,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2613103431","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.039895643,0.00023717505,0.93705666,0.0075632255,0.0004715428,0.0016615908,0.000013377774,0.000009726415,0.013091033],"genre_scores_gemma":[0.9373242,0.000016281381,0.062255096,0.00003607123,0.00007110191,0.00019582015,5.419494e-7,0.00000866726,0.00009219721],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9526764,0.016344182,0.00191822,0.0011231829,0.026860977,0.0010770345],"domain_scores_gemma":[0.96345603,0.029645495,0.00044608655,0.0030822563,0.0027716586,0.00059844257],"candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.24910049,0.00019637149,0.0007489277,0.00019305073,0.0007492167,0.0005378898,0.0037057395,0.0001293468,0.0009842395],"category_scores_gemma":[0.26424545,0.000093811104,0.000072582276,0.00048932,0.0038596678,0.00012938419,0.0009970277,0.0013143372,0.00005060629],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010675256,0.0018211257,0.18604782,0.00069313135,0.00014259764,0.00011460665,0.0010290885,0.0009232444,0.0013548638,0.12632635,0.0065151094,0.67396456],"study_design_scores_gemma":[0.002358788,0.0014621908,0.6582694,0.0012801354,0.000033464425,0.000008780909,0.0023369289,0.080041446,0.0007546898,0.24301313,0.01004713,0.00039391764],"about_ca_topic_score_codex":0.00081028964,"about_ca_topic_score_gemma":0.0008219313,"teacher_disagreement_score":0.8974286,"about_ca_system_score_codex":0.00034370046,"about_ca_system_score_gemma":0.0006091844,"threshold_uncertainty_score":0.999929},"labels":[],"label_agreement":null},{"id":"W2620680567","doi":"10.3899/jrheum.170315","title":"Dr. Christiansen, <i>et al,</i> reply","year":2017,"lang":"he","type":"letter","venue":"The Journal of Rheumatology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"","keywords":"Medicine; Minor (academic); Statistics; Demography; Humanities; Mathematics","score_opus":0.09729109516209565,"score_gpt":0.3599983043807382,"score_spread":0.26270720921864255,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2620680567","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.039405722,0.006286686,0.0017165102,0.93924594,0.010668922,0.00041239415,0.000032224765,0.000012191555,0.00221941],"genre_scores_gemma":[0.06322927,0.032750323,0.0005516149,0.8942781,0.00044431942,0.0000054943903,0.000007461402,0.00008374544,0.008649648],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98285735,0.006052875,0.004594221,0.00070690707,0.004799057,0.0009895643],"domain_scores_gemma":[0.9751189,0.008746897,0.010079444,0.0037009406,0.0021060912,0.00024773035],"candidate_categories":["metaresearch","metaepi_narrow","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.042092353,0.00073544023,0.0024539435,0.00047130542,0.001164222,0.0006926704,0.0085651055,0.0005159981,0.0025053432],"category_scores_gemma":[0.012895856,0.0003905081,0.0010614797,0.00027856714,0.0017773346,0.00051026343,0.0006693109,0.004964286,0.0018908802],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032505512,0.0000857563,0.012953733,0.00006429029,0.00038758497,0.0002133213,0.0009881257,0.000091103066,0.00005863604,0.000077034456,0.9817886,0.002966802],"study_design_scores_gemma":[0.0009765248,0.0005843302,0.0008122656,0.0009946738,0.0002590496,0.012991612,0.00045530873,0.000060336104,0.00003184581,0.0075092814,0.9749234,0.00040137637],"about_ca_topic_score_codex":0.000673283,"about_ca_topic_score_gemma":0.00011633677,"teacher_disagreement_score":0.044967823,"about_ca_system_score_codex":0.00016081436,"about_ca_system_score_gemma":0.0013000873,"threshold_uncertainty_score":0.9998547},"labels":[],"label_agreement":null},{"id":"W2626903902","doi":"10.17705/1jais.00458","title":"Heuristic Principles and Differential Judgments in the Assessment of Information Quality","year":2017,"lang":"en","type":"article","venue":"Journal of the Association for Information Systems","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Completeness (order theory); Heuristic; Objectivity (philosophy); Psychology; Cognitive psychology; Computer science; Quality (philosophy); Set (abstract data type); Artificial intelligence; Social psychology; Epistemology; Mathematics","score_opus":0.1708040051470891,"score_gpt":0.4194472045691902,"score_spread":0.2486431994221011,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2626903902","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97164756,0.000016275602,0.013291068,0.0035779795,0.0028712312,0.0012287231,0.00008475328,0.0000031961952,0.007279198],"genre_scores_gemma":[0.99965966,0.000009861552,0.00009241275,0.00007146914,0.00005711764,0.000012731234,0.000002632117,9.94884e-7,0.00009311512],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9944175,0.00048184823,0.0022584815,0.000043817006,0.0026907348,0.00010761588],"domain_scores_gemma":[0.98864853,0.000966948,0.00866686,0.00038998443,0.0013022725,0.00002537554],"candidate_categories":["metaresearch","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.023742812,0.00007306302,0.00026964053,0.00014842008,0.0003731284,0.0010723966,0.000932812,0.00006495482,0.0000026648313],"category_scores_gemma":[0.010741828,0.00003552927,0.00014378746,0.00010799092,0.000032031294,0.0028382626,0.000086103544,0.00014216518,0.0000029854662],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000059606034,0.00008052791,0.957734,0.00014673802,0.00007686609,2.9978967e-8,0.0047608493,0.0037379295,0.000050251932,0.026768338,0.0030542137,0.0035306884],"study_design_scores_gemma":[0.00095713825,0.000053524276,0.96730417,0.00007037674,0.000019911871,0.0000014720287,0.0024685105,0.003955325,0.000024744177,0.0015170141,0.02358275,0.00004506098],"about_ca_topic_score_codex":0.00003747754,"about_ca_topic_score_gemma":0.0000129283,"teacher_disagreement_score":0.028012091,"about_ca_system_score_codex":0.00024868976,"about_ca_system_score_gemma":0.00009288954,"threshold_uncertainty_score":0.9999646},"labels":[],"label_agreement":null},{"id":"W2768731979","doi":"10.4103/jpi.jpi_50_17","title":"Next Generation Quality: Assessing the Physician in Clinical History Completeness and Diagnostic Interpretations Using Funnel Plots and Normalized Deviations Plots in 3,854 Prostate Biopsies","year":2017,"lang":"en","type":"article","venue":"Journal of Pathology Informatics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Juravinski Cancer Centre; McMaster University; St. Joseph’s Healthcare Hamilton","funders":"","keywords":"Funnel plot; Medicine; Audit; Categorization; Medical physics; Statistics; Pathology; Artificial intelligence; Computer science; Publication bias; Meta-analysis; Mathematics","score_opus":0.5222789496249143,"score_gpt":0.49371059623875124,"score_spread":0.028568353386163037,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2768731979","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99008644,0.00024327624,0.007926153,0.00087610504,0.0005751304,0.0002026808,0.000004901326,0.0000026293665,0.00008270947],"genre_scores_gemma":[0.9947037,0.00021953665,0.0043767253,0.00061979174,0.00006594378,0.00000456152,0.0000014585177,0.0000038109717,0.00000446949],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9960147,0.00080176775,0.0023970164,0.00010288528,0.00052367314,0.0001599463],"domain_scores_gemma":[0.994365,0.0025739544,0.0022779969,0.0003217327,0.00040615504,0.000055166616],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.011845001,0.000110708876,0.00042230816,0.00025320487,0.00035308884,0.00047101578,0.00036559047,0.000080632315,0.0000040738823],"category_scores_gemma":[0.012658321,0.00007105213,0.00006501299,0.00010300809,0.00048735001,0.0020740805,0.00014139141,0.0003490362,0.000002533766],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003397643,0.00019539674,0.9309102,0.000052419284,0.000026769845,0.00003262264,0.026320938,0.0041816323,0.001918834,0.00056330813,0.00015158718,0.035612293],"study_design_scores_gemma":[0.000946101,0.000083791376,0.90095425,0.00017060024,0.000032792108,0.00005653927,0.009796003,0.084523454,0.000018126953,0.0030444865,0.00026898607,0.00010485701],"about_ca_topic_score_codex":0.000019917992,"about_ca_topic_score_gemma":0.00019882855,"teacher_disagreement_score":0.08034182,"about_ca_system_score_codex":0.00011989408,"about_ca_system_score_gemma":0.00017796029,"threshold_uncertainty_score":0.99565846},"labels":[],"label_agreement":null},{"id":"W2774785678","doi":"10.1177/0194599817739260","title":"To <i>P</i> or Not to <i>P</i>: Backing Bayesian Statistics","year":2017,"lang":"en","type":"review","venue":"Otolaryngology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"BC Children's Hospital; University of British Columbia","funders":"","keywords":"Null hypothesis; Bayesian probability; Null (SQL); Statistics; Statistical hypothesis testing; Bayesian statistics; Null model; Econometrics; Posterior probability; Alternative hypothesis; Population; Mathematics; Computer science; Bayesian inference; Data mining; Sociology; Demography","score_opus":0.42737667020850845,"score_gpt":0.5076963095392637,"score_spread":0.08031963933075525,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2774785678","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000014545833,0.9008247,0.07373378,0.0019101781,0.0055933436,0.005895189,0.0012270571,0.00011712761,0.01068407],"genre_scores_gemma":[0.000016164948,0.95304024,0.025286438,0.004583015,0.0006728578,0.00039768868,0.00005297193,0.00009421449,0.01585638],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.991222,0.0012782592,0.0023685724,0.0018978007,0.0022671088,0.0009662369],"domain_scores_gemma":[0.9906221,0.0033172218,0.0010517496,0.0037648631,0.0005332955,0.0007107849],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.007652546,0.00075360906,0.0035486487,0.0007393151,0.0005393205,0.00076726376,0.004984145,0.0005700178,0.0023374024],"category_scores_gemma":[0.016702633,0.0004772484,0.0005347416,0.0007991489,0.00015705333,0.00017816482,0.0012600078,0.00067553506,0.012787249],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000048230628,0.00006264945,0.000046628556,0.00088082196,0.000059676964,0.00013144057,0.0000990619,0.0000051427455,0.0000020637326,0.0006574509,0.10837991,0.8896269],"study_design_scores_gemma":[0.00013409696,0.00034785274,0.00011529629,0.0021678659,0.00018623308,0.00005566089,0.000015980371,0.000006594224,0.0000021464332,0.0007753379,0.99559546,0.00059748266],"about_ca_topic_score_codex":0.000049707163,"about_ca_topic_score_gemma":0.00047690558,"teacher_disagreement_score":0.88902944,"about_ca_system_score_codex":0.00020242005,"about_ca_system_score_gemma":0.00080502045,"threshold_uncertainty_score":0.9997679},"labels":[],"label_agreement":null},{"id":"W2782320564","doi":"10.1007/s40037-017-0391-8","title":"Detecting rater bias using a&amp;nbsp;person-fit statistic: a&amp;nbsp;Monte Carlo simulation study","year":2018,"lang":"en","type":"article","venue":"Perspectives on Medical Education","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; Université de Sherbrooke","funders":"Fonds de Recherche du Québec-Société et Culture","keywords":"Monte Carlo method; Statistic; Medical physics; Computer science; Psychology; Medicine; Statistics; Mathematics","score_opus":0.4676450050714238,"score_gpt":0.5170803826767165,"score_spread":0.04943537760529271,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2782320564","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9720289,0.00014434404,0.021937348,0.0017328287,0.0020502696,0.000911088,0.000007035251,0.0000713475,0.0011168064],"genre_scores_gemma":[0.993521,0.000008569723,0.0035123723,0.00053697824,0.0013455715,0.000045916007,0.0000031145748,0.000026067231,0.0010004576],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.991454,0.0013165112,0.00093647104,0.0012226147,0.0045979754,0.00047240016],"domain_scores_gemma":[0.9940164,0.0022014799,0.00048318118,0.0009988195,0.0018583563,0.00044177627],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.008527765,0.00032657315,0.0004216501,0.00046455237,0.00075516745,0.00043434853,0.00067232613,0.00016370103,0.00420603],"category_scores_gemma":[0.048720956,0.00024596287,0.00014062528,0.0010218829,0.00039818618,0.00043704876,0.00009132592,0.00047522935,0.00076765125],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011007015,0.014728379,0.15342358,0.00006461286,0.00036229033,0.0000095374,0.45213386,0.021636467,0.0015111401,0.00078228716,0.011472971,0.34277418],"study_design_scores_gemma":[0.0030316834,0.002361671,0.21257614,0.00051519333,0.00025295233,0.000018293777,0.6020955,0.14913248,0.00029390058,0.0059868977,0.022197906,0.0015373946],"about_ca_topic_score_codex":0.00090624223,"about_ca_topic_score_gemma":0.0010291126,"teacher_disagreement_score":0.34123677,"about_ca_system_score_codex":0.00071715366,"about_ca_system_score_gemma":0.0008223636,"threshold_uncertainty_score":0.9999993},"labels":[],"label_agreement":null},{"id":"W2784251105","doi":"10.1016/j.jneb.2017.08.006","title":"Response to “A Comment on Mann G ‘ Smart Snacks in School Legislation Does Not Change Self-Reported Snack Food and Beverage Intake of Middle School Students in Rural Appalachian Region’: Methodological Issues”","year":2018,"lang":"en","type":"letter","venue":"Journal of Nutrition Education and Behavior","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"","keywords":"Legislation; Scopus; Psychology; Snack food; Rural area; Environmental health; Political science; Medicine; Food science; MEDLINE; Law","score_opus":0.2797568639614261,"score_gpt":0.43798714022337815,"score_spread":0.15823027626195202,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2784251105","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6639859,0.000119051634,0.00002290851,0.33370066,0.00093527907,0.00120539,0.000016615524,0.0000055579576,0.000008642993],"genre_scores_gemma":[0.79732156,0.0007141298,0.009454826,0.18828851,0.002643343,0.0007369366,0.00006956695,0.000037115387,0.00073400466],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9922893,0.0026161033,0.0021588167,0.0004916305,0.002184062,0.00026008856],"domain_scores_gemma":[0.9961302,0.00081794214,0.0014684267,0.0004469241,0.00089736114,0.0002391208],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009230656,0.00030592244,0.0007604438,0.0016495024,0.00009565733,0.0002678572,0.0004954418,0.00045180143,0.0002051502],"category_scores_gemma":[0.0024729313,0.00021558118,0.00015187569,0.0004710902,0.00009247674,0.0004595813,0.000099674035,0.0010579431,0.000010950454],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023180745,0.015443206,0.46214044,0.0002024573,0.00008099806,0.00010984128,0.004360455,0.0000010659642,0.0012056133,0.000007521413,0.5046271,0.009503204],"study_design_scores_gemma":[0.002362029,0.0025280642,0.77648693,0.0016370433,0.00012576106,0.000099124256,0.0041493247,0.0000025213426,0.00041986044,0.00129823,0.2105785,0.00031261708],"about_ca_topic_score_codex":0.000090902766,"about_ca_topic_score_gemma":0.00008901379,"teacher_disagreement_score":0.3143465,"about_ca_system_score_codex":0.00044886945,"about_ca_system_score_gemma":0.00017355075,"threshold_uncertainty_score":0.87911457},"labels":[],"label_agreement":null},{"id":"W2790302041","doi":"10.1016/j.jclinepi.2018.03.001","title":"Response to Dr. Livia Puljak's letter about our published article","year":2018,"lang":"en","type":"letter","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Scopus; Publication bias; Systematic review; MEDLINE; Attrition; Reporting bias; Psychology; Medicine; Medical education; Meta-analysis; Political science; Law","score_opus":0.5651543854416219,"score_gpt":0.5667698303157579,"score_spread":0.0016154448741360072,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2790302041","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.057310626,0.00017470715,0.0053914944,0.92278945,0.013509725,0.00033610195,0.000017240429,0.000017085047,0.00045358934],"genre_scores_gemma":[0.008699056,0.00005855863,0.0054755267,0.94070995,0.036091104,0.000010096596,0.0000030341087,0.000035531943,0.008917165],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.8956357,0.07087606,0.02502044,0.0019316934,0.004821534,0.0017145263],"domain_scores_gemma":[0.7611112,0.21388386,0.015233917,0.0034138665,0.005092835,0.0012643253],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity","insufficient_payload"],"category_scores_codex":[0.4055864,0.0005424379,0.0058505526,0.00076299603,0.00016516523,0.00030688941,0.004298797,0.0024845968,0.0018825618],"category_scores_gemma":[0.7543132,0.0003216221,0.0027462922,0.00070168317,0.0004871057,0.0005234345,0.00057050056,0.007048631,0.0039053052],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017809962,0.00015840096,0.031173492,0.0000106849075,0.00015868033,0.00025893236,0.000048156737,0.000050008133,0.00006447705,0.00001528636,0.9610845,0.0051963865],"study_design_scores_gemma":[0.0007348835,0.0019396062,0.107109584,0.00015810053,0.00007220665,0.00005824642,0.000030206364,0.00005424402,0.000009533005,0.03189531,0.8576499,0.0002881406],"about_ca_topic_score_codex":0.00001501973,"about_ca_topic_score_gemma":0.0000110581905,"teacher_disagreement_score":0.34872678,"about_ca_system_score_codex":0.00015953301,"about_ca_system_score_gemma":0.00071974314,"threshold_uncertainty_score":0.9999236},"labels":[],"label_agreement":null},{"id":"W2808430903","doi":"10.1007/s40037-018-0435-8","title":"Examining the effects of gaming and guessing on script concordance test scores","year":2018,"lang":"en","type":"article","venue":"Perspectives on Medical Education","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; McGill University","funders":"","keywords":"Concordance; Likert scale; Test (biology); Statistics; Psychology; Scale (ratio); Social psychology; Mathematics; Medicine","score_opus":0.07302416389855387,"score_gpt":0.39259414271459697,"score_spread":0.3195699788160431,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2808430903","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98776454,0.0008352269,0.000702612,0.004102796,0.001124895,0.0002365793,5.3363857e-7,0.000012502269,0.005220295],"genre_scores_gemma":[0.9979042,0.00008754558,0.00026811394,0.0010212393,0.0004780351,0.000019410276,3.3015138e-7,0.000004815007,0.00021633001],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9970338,0.00029858388,0.00032431303,0.00041777198,0.0017808783,0.0001446601],"domain_scores_gemma":[0.99336505,0.005458388,0.00020647362,0.00042049002,0.0004267031,0.00012290016],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0033707845,0.00010315469,0.00016491849,0.000109862514,0.0002550722,0.00009046225,0.0003677225,0.000049708095,0.0001780414],"category_scores_gemma":[0.08151824,0.00005806861,0.00003074696,0.00032765744,0.0008475005,0.00011533799,0.000049338232,0.00018812549,0.00003286883],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012625262,0.0020765243,0.11937331,0.00009640296,0.00003268077,0.000001771694,0.08036845,0.0000033859026,0.012250045,0.015785996,0.013486005,0.75639915],"study_design_scores_gemma":[0.0007782149,0.0029860237,0.83024776,0.0023519532,0.00003196646,0.0000058189066,0.124077246,0.0010993957,0.015721383,0.019500284,0.0029235526,0.00027641645],"about_ca_topic_score_codex":0.00004376108,"about_ca_topic_score_gemma":0.000013266801,"teacher_disagreement_score":0.75612277,"about_ca_system_score_codex":0.0000762343,"about_ca_system_score_gemma":0.00026618992,"threshold_uncertainty_score":0.9262185},"labels":[],"label_agreement":null},{"id":"W2809585708","doi":"10.7939/r3tm7251q","title":"Impact of raters' levels of respiratory training and experience on the quality of spirometric interpretations in epidemiological studies","year":2000,"lang":"en","type":"article","venue":"Library and Archives Canada (Government of Canada)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"","keywords":"Epidemiology; Quality (philosophy); Medicine; Psychology; Spirometry; Applied psychology; Environmental health; Internal medicine; Epistemology; Asthma","score_opus":0.21748750175156903,"score_gpt":0.3594278734719293,"score_spread":0.14194037172036025,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2809585708","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.990181,0.00043764926,0.00001945936,0.0008074569,0.0000270566,0.00013307815,0.00009469909,8.339193e-7,0.008298721],"genre_scores_gemma":[0.999327,0.000098350734,0.00015935338,0.0002850162,0.000004409104,0.0000047002454,1.2538023e-7,0.0000020235623,0.000119033975],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99633235,0.0005938428,0.0009211688,0.00020984074,0.0018027759,0.0001400267],"domain_scores_gemma":[0.99320346,0.006112825,0.00037285555,0.00023592866,0.0000038065082,0.000071114286],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000567512,0.00010579767,0.0004539318,0.00005851042,0.00005757193,0.000006347277,0.00028981298,0.000016438576,0.00012401846],"category_scores_gemma":[0.00069817185,0.000059087888,0.00005135574,0.00029572725,0.00037973293,0.00016911286,0.00006990263,0.00008452695,1.3013937e-9],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006198133,0.00011003691,0.88892716,0.00008036297,0.00009987473,0.0000036274807,0.0051600374,0.0015219084,0.025952714,0.015179,0.00012199295,0.06222347],"study_design_scores_gemma":[0.00017664988,0.00028098977,0.96865034,0.000121136414,0.000003486497,1.9872563e-7,0.0139925545,0.00023729974,0.011409796,0.0049853376,0.0000775997,0.000064635686],"about_ca_topic_score_codex":0.0010513734,"about_ca_topic_score_gemma":0.002356283,"teacher_disagreement_score":0.07972315,"about_ca_system_score_codex":0.000008756344,"about_ca_system_score_gemma":0.00039538366,"threshold_uncertainty_score":0.24095342},"labels":[],"label_agreement":null},{"id":"W2886074557","doi":"10.1177/0956797617698528","title":"Increasing the Similarity of Lineup Fillers to the Suspect Improves the Applied Value of Lineups Without Improving Memory Performance: Commentary on Colloff, Wade, and Strange (2016)","year":2018,"lang":"en","type":"letter","venue":"Psychological Science","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"","keywords":"Suspect; Psychology; Similarity (geometry); Value (mathematics); Cognitive psychology; Psychoanalysis; Social psychology; Artificial intelligence; Criminology","score_opus":0.09294168632646475,"score_gpt":0.3678399947924385,"score_spread":0.27489830846597374,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2886074557","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6403815,0.000072165916,0.00009472038,0.35421044,0.0007390762,0.0014844866,0.00004187407,0.000015563854,0.0029601778],"genre_scores_gemma":[0.7560856,0.000056825364,0.00051260425,0.24236326,0.00082656444,0.00004879482,0.0000023611092,0.00000875189,0.00009525937],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9922687,0.0008269474,0.0010885978,0.0012440147,0.0039283237,0.0006434435],"domain_scores_gemma":[0.9932384,0.0035061412,0.00076910766,0.0019451211,0.000425118,0.000116112074],"candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.026014658,0.00037914803,0.0005705695,0.00016010537,0.001183053,0.00032155847,0.00490948,0.00026534408,0.000138513],"category_scores_gemma":[0.0027770805,0.0001364509,0.00016767239,0.0013345142,0.004745168,0.00015279518,0.0007777314,0.001404012,0.000020758762],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007717425,0.00048047368,0.005426429,0.00010141723,0.000054944652,0.0000043673876,0.002499655,0.0004085754,0.019419547,0.00009370826,0.88972384,0.081015326],"study_design_scores_gemma":[0.002822903,0.010103219,0.5021387,0.00082046114,0.00037513018,0.00005993065,0.0064826794,0.009258195,0.015114435,0.008789431,0.4418226,0.0022123603],"about_ca_topic_score_codex":0.00019877896,"about_ca_topic_score_gemma":0.000025465273,"teacher_disagreement_score":0.49671224,"about_ca_system_score_codex":0.00008683011,"about_ca_system_score_gemma":0.00009748763,"threshold_uncertainty_score":0.99796337},"labels":[],"label_agreement":null},{"id":"W2886409188","doi":"","title":"Survey of Canadian human anatomy courses in Kinesiology and Physical Education","year":2014,"lang":"en","type":"article","venue":"European Journal of Anatomy","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Kinesiology; Accreditation; Syllabus; Intraclass correlation; Medical education; Human anatomy; Gross anatomy; Physical education; Medicine; Psychology; Reliability (semiconductor); Anatomy; Mathematics education","score_opus":0.10231163467910354,"score_gpt":0.378678326262698,"score_spread":0.27636669158359445,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2886409188","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98977244,0.00013984859,0.000043145963,0.0008472344,0.0002197336,0.000055173976,0.0000028626057,0.0000013145359,0.008918226],"genre_scores_gemma":[0.99957365,0.0000204229,0.00011772102,0.000136794,0.00009950766,1.8944864e-7,0.0000012374395,0.000005779748,0.000044728127],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99627006,0.0021248097,0.0007301053,0.00016959284,0.0005546678,0.00015075301],"domain_scores_gemma":[0.9980034,0.00041849827,0.00045626512,0.0002232919,0.0007052888,0.00019324002],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01206359,0.00008798084,0.00029986195,0.00089140504,0.00006655197,0.000047111473,0.0004730483,0.000017279406,0.000053265307],"category_scores_gemma":[0.0022268966,0.00006228545,0.00005434641,0.0006576548,0.00013795334,0.00014452875,0.000035379744,0.00016939026,0.000014440364],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000029582061,0.0002576419,0.91665065,0.000009993243,0.00001705102,0.000017581167,0.00085434306,0.00009180489,0.00084919384,0.0019395886,0.00905588,0.07022666],"study_design_scores_gemma":[0.00032935728,0.00025166094,0.96849686,0.00003998877,0.0000055628566,0.000007043247,0.00030585588,0.00007541179,0.00010501668,0.0014988495,0.028819282,0.000065109765],"about_ca_topic_score_codex":0.0063942354,"about_ca_topic_score_gemma":0.023959137,"teacher_disagreement_score":0.07016155,"about_ca_system_score_codex":0.00003314244,"about_ca_system_score_gemma":0.00019252156,"threshold_uncertainty_score":0.99385107},"labels":[],"label_agreement":null},{"id":"W2908490955","doi":"10.3138/ptc.2017-49.f","title":"La version franco-canadienne du « STrengthening the Reporting of OBservational studies in Epidemiology » (STROBE) Statement : L’outil STROBE","year":2019,"lang":"fr","type":"article","venue":"Physiotherapy Canada","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":true,"ca_institutions":"Children's Hospital of Eastern Ontario; University of Ottawa; Université du Québec en Outaouais; Université de Montréal","funders":"","keywords":"Strengthening the reporting of observational studies in epidemiology; Observational study; Inter-rater reliability; CLARITY; Reliability (semiconductor); Psychology; Ambiguity; Meaning (existential); Cohen's kappa; Kappa; Medical education; Guideline; Applied psychology; Medical physics; Medicine; Computer science; Statistics; Linguistics; Mathematics; Rating scale","score_opus":0.251915228834975,"score_gpt":0.421755229837808,"score_spread":0.16984000100283297,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2908490955","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96568096,0.0066617937,0.00015773567,0.02331783,0.0030846868,0.0005351528,0.00007303554,0.000005696546,0.0004831094],"genre_scores_gemma":[0.9949725,0.001418235,0.00079610344,0.0012524053,0.00015702384,0.000022669128,0.000008834002,0.000015806118,0.0013564134],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.9917623,0.0017202131,0.0038506733,0.000666259,0.0012691624,0.00073140446],"domain_scores_gemma":[0.98779404,0.007681512,0.0032599661,0.00074859813,0.00038565637,0.00013020582],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.018261664,0.00030559537,0.000992873,0.00011087893,0.00021358972,0.000025769152,0.00068766554,0.000096379066,0.0012700008],"category_scores_gemma":[0.0047324104,0.00021066601,0.0001842244,0.00068490003,0.00035952625,0.00022690163,0.00013115,0.00038544886,0.000009160984],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00020631973,0.00027821714,0.7661554,0.00034447075,0.00042761274,0.00002943865,0.0070134406,0.101030834,0.002520949,0.006723158,0.02744656,0.08782357],"study_design_scores_gemma":[0.0023923533,0.0005205189,0.38294813,0.0017650189,0.000049191396,0.000007771732,0.045731515,0.022522327,0.0007299145,0.018717507,0.52396786,0.0006479052],"about_ca_topic_score_codex":0.63533765,"about_ca_topic_score_gemma":0.7788702,"teacher_disagreement_score":0.4965213,"about_ca_system_score_codex":0.0012877395,"about_ca_system_score_gemma":0.001593117,"threshold_uncertainty_score":0.99964297},"labels":[],"label_agreement":null},{"id":"W2912251989","doi":"10.1002/0470011815.b2a04002","title":"Agreement, Measurement of","year":2005,"lang":"en","type":"other","venue":"Encyclopedia of Biostatistics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph","funders":"","keywords":"Categorical variable; Agreement; Unanimity; Measure (data warehouse); Kappa; Interpretation (philosophy); Statistics; Mathematics; Monotonic function; Interval (graph theory); Computer science; Data mining; Combinatorics; Mathematical analysis; Geometry; Linguistics","score_opus":0.07732314664788557,"score_gpt":0.3405152790255392,"score_spread":0.2631921323776536,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2912251989","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000032131516,0.0021616481,0.014454486,0.00018102088,0.0012825914,0.00062862574,0.0010766541,0.000029642148,0.9801532],"genre_scores_gemma":[0.003768655,0.0054717856,0.051643718,0.000097198485,0.0008131275,0.000025228555,0.000050389404,0.00024180497,0.9378881],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9910139,0.00026589126,0.0018698272,0.0006066522,0.0059255115,0.00031821986],"domain_scores_gemma":[0.99527603,0.00046313624,0.0019723536,0.001307437,0.00084548705,0.00013556713],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0040182373,0.00037008477,0.0009315858,0.00056033593,0.00003188811,0.000025238462,0.001157388,0.00025354914,0.022091202],"category_scores_gemma":[0.0042563463,0.00028583978,0.00021821448,0.0004957739,0.00030724957,0.00004198372,0.00017614699,0.00019254198,0.0004988403],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00001400027,0.00027370604,0.0013789523,0.00014052333,0.00009750362,0.000002500754,0.000116712996,0.0000129027485,0.000095388314,0.0016057751,0.9035875,0.0926745],"study_design_scores_gemma":[0.00036729354,0.00016831628,0.0015372512,0.000301166,0.00009596272,4.5697394e-7,0.00012440058,0.000022737071,0.00026496113,0.0022601648,0.99457705,0.0002802588],"about_ca_topic_score_codex":0.00018286951,"about_ca_topic_score_gemma":0.0005066054,"teacher_disagreement_score":0.09239425,"about_ca_system_score_codex":0.0000852796,"about_ca_system_score_gemma":0.00027383785,"threshold_uncertainty_score":0.99995935},"labels":[],"label_agreement":null},{"id":"W2913588687","doi":"10.1002/pds.4701","title":"Conditional validation sampling for consistent risk estimation with binary outcome data subject to misclassification","year":2019,"lang":"en","type":"article","venue":"Pharmacoepidemiology and Drug Safety","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Carleton University; Institute of Population and Public Health; University of Ottawa","funders":"Natural Sciences and Engineering Research Council of Canada; Mitacs","keywords":"Statistics; Contingency table; Categorical variable; Sampling (signal processing); Sample size determination; Simple random sample; Stratified sampling; Sampling design; Estimator; Odds ratio; Econometrics; Computer science; Medicine; Mathematics; Population","score_opus":0.3722356889622388,"score_gpt":0.482635052189559,"score_spread":0.11039936322732019,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2913588687","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4720392,0.00011070755,0.5101651,0.014951812,0.00046323877,0.0013905654,0.0006010855,0.00003770279,0.00024063913],"genre_scores_gemma":[0.9607617,0.00005539811,0.036401596,0.0015339624,0.000078697005,0.0000720093,0.00085928384,0.000007985687,0.00022933002],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9961367,0.0012253305,0.0010304314,0.0008736028,0.0004796415,0.0002542844],"domain_scores_gemma":[0.9866155,0.0117282625,0.00051277917,0.00070304325,0.00030421396,0.0001362503],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.023538465,0.00017151432,0.00040829708,0.00013724228,0.00034843234,0.00004897667,0.00047564157,0.00006642509,0.00027032467],"category_scores_gemma":[0.007785028,0.00011715116,0.00006072265,0.00021389175,0.00011031488,0.00041786037,0.00014581448,0.00017205346,0.00017955544],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012596645,0.00012895699,0.88137394,0.000054601485,0.000091470196,3.654473e-7,0.00021189661,0.07008699,0.0017838714,0.0035071815,0.014903479,0.026597584],"study_design_scores_gemma":[0.002473515,0.00023452037,0.5709619,0.000043968157,0.00018558142,0.000008671124,0.00043954793,0.33437467,0.00052281824,0.026563913,0.063783415,0.00040745395],"about_ca_topic_score_codex":0.000033088916,"about_ca_topic_score_gemma":0.000014079278,"teacher_disagreement_score":0.48872256,"about_ca_system_score_codex":0.000059916863,"about_ca_system_score_gemma":0.00006321951,"threshold_uncertainty_score":0.93199676},"labels":[],"label_agreement":null},{"id":"W2965392335","doi":"10.1111/jth.14540","title":"Reply: Method agreement analysis and interobserver reliability of the ISTH proposed definitions for effective hemostasis in the management of major bleeding: Methodological issues","year":2019,"lang":"en","type":"letter","venue":"Journal of Thrombosis and Haemostasis","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Kappa; Cohen's kappa; Medicine; Statistics; Scopus; MEDLINE; Mathematics; Family medicine","score_opus":0.3948113073500375,"score_gpt":0.4581062183798813,"score_spread":0.0632949110298438,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2965392335","genre_codex":"commentary","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.17726445,0.0012803354,0.01723683,0.79876846,0.00058171607,0.004186639,0.00039753865,0.000005031438,0.0002790205],"genre_scores_gemma":[0.3765601,0.008887292,0.45842522,0.15449443,0.00071776786,0.00032765424,0.000050915452,0.00008808709,0.00044854064],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9895511,0.003998801,0.0027278808,0.0007537604,0.0026278028,0.00034061563],"domain_scores_gemma":[0.98533297,0.0090457415,0.0030908254,0.0012336564,0.001246134,0.000050654962],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.030604498,0.0004084321,0.002159015,0.00055679143,0.0002026843,0.00018803618,0.0014908493,0.00031873962,0.00012959568],"category_scores_gemma":[0.002508774,0.00018857351,0.0013769548,0.0015155878,0.00045312074,0.00021182775,0.00047788405,0.0009124508,8.516483e-7],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014274962,0.0021679595,0.10561965,0.005117038,0.018526074,0.000033584227,0.008128251,0.0006195793,0.0007792401,0.0012493806,0.793193,0.06313876],"study_design_scores_gemma":[0.006111494,0.0044910195,0.6002513,0.0032354596,0.029712152,0.000046837427,0.014553029,0.00053830515,0.013684369,0.11724916,0.20890877,0.0012180988],"about_ca_topic_score_codex":0.00019020795,"about_ca_topic_score_gemma":0.000033578825,"teacher_disagreement_score":0.644274,"about_ca_system_score_codex":0.00014394587,"about_ca_system_score_gemma":0.000054767297,"threshold_uncertainty_score":0.99819666},"labels":[],"label_agreement":null},{"id":"W2979614843","doi":"10.1016/j.jbiomech.2019.109410","title":"A machine-learning method for classifying and analyzing foot placement: Application to manual material handling","year":2019,"lang":"en","type":"article","venue":"Journal of Biomechanics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; Institut de Recherche Robert-Sauvé en Santé et en Sécurité du Travail; Centre intégré universitaire de santé et de services sociaux de la Capitale-Nationale; Centre for Interdisciplinary Research in Rehabilitation; Centre Intégré Universitaire de Santé et de Services Sociaux du Centre-Sud-de-l'Île-de-Montréal; Centre Intégré Universitaire de Santé et de Services Sociaux du Saguenay–Lac-Saint-Jean","funders":"Mitacs; Institut de Recherche Robert-Sauvé en Santé et en Sécurité du Travail","keywords":"Foot (prosody); Computer science; Task (project management); Set (abstract data type); Artificial intelligence; Displacement (psychology); Machine learning; Class (philosophy); Engineering; Psychology","score_opus":0.06836018796622308,"score_gpt":0.3844547014322206,"score_spread":0.31609451346599754,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2979614843","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.25362772,0.00011182311,0.7446864,0.000558779,0.0005874958,0.00036996967,0.000009300929,0.0000065932772,0.00004190543],"genre_scores_gemma":[0.86913306,0.00003421193,0.13024506,0.00016722183,0.00020081678,0.0000107517435,0.00000309763,0.000013567382,0.00019219708],"study_design_codex":"bench_or_experimental","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99738103,0.00019986932,0.00091422466,0.0002957443,0.0009944729,0.00021466639],"domain_scores_gemma":[0.99793833,0.0005732444,0.0007319765,0.0001989661,0.00042692025,0.00013054637],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011636821,0.00012648106,0.00035478533,0.0004280654,0.00015805737,0.00030913972,0.000362173,0.00007297566,0.00006893439],"category_scores_gemma":[0.000794075,0.00009144168,0.00012966686,0.00038529633,0.00000741266,0.00022100002,0.00012035954,0.00015562208,0.000017841285],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00049235346,0.000064298125,0.0010756373,0.00006202631,0.00007187561,0.0000011479981,0.0005781684,0.0016110473,0.65145147,0.0008354116,0.00019246759,0.3435641],"study_design_scores_gemma":[0.0026489892,0.002539166,0.00016334918,0.00023718867,0.00014225695,0.000042347703,0.002657126,0.73508257,0.09740926,0.033892013,0.12472495,0.00046081553],"about_ca_topic_score_codex":0.0000059515887,"about_ca_topic_score_gemma":0.0000047635003,"teacher_disagreement_score":0.7334715,"about_ca_system_score_codex":0.000084093386,"about_ca_system_score_gemma":0.000037428938,"threshold_uncertainty_score":0.4033113},"labels":[],"label_agreement":null},{"id":"W2983818347","doi":"10.3138/ptc-2018-0110","title":"How to Optimize Measurement Protocols: An Example of Assessing Measurement Reliability Using Generalizability Theory","year":2019,"lang":"en","type":"article","venue":"Physiotherapy Canada","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"University of Waterloo; McMaster University","funders":"","keywords":"Generalizability theory; Reliability (semiconductor); Computer science; Observational error; Terminology; Reliability engineering; Measurement uncertainty; Standard error; Data mining; Statistics; Power (physics); Mathematics","score_opus":0.29425411750762454,"score_gpt":0.3971427089229444,"score_spread":0.10288859141531986,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2983818347","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9360336,0.0001277802,0.04688909,0.0013601503,0.0007386068,0.014330816,0.00002325238,0.000045178083,0.0004515043],"genre_scores_gemma":[0.98656535,0.0000022734116,0.011482418,0.0008630486,0.00012669506,0.0007971509,0.0000015300873,0.000039462244,0.000122069025],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.98400253,0.002239758,0.0013323997,0.0015309881,0.010230448,0.00066388637],"domain_scores_gemma":[0.9919217,0.00037552483,0.00065593387,0.0031318488,0.0034972718,0.00041770752],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.02526036,0.000452353,0.0009118806,0.00017757845,0.00025430936,0.0005061935,0.001515033,0.00009276248,0.00067430956],"category_scores_gemma":[0.0014014686,0.0003401339,0.00023974427,0.0009953593,0.00011455277,0.0008211219,0.00014238838,0.00020454483,0.000006246549],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006956255,0.0013363094,0.029886434,0.00033384838,0.00016009335,0.0000018692994,0.0010314962,0.07897588,0.8466217,0.00070307846,0.002658326,0.03759534],"study_design_scores_gemma":[0.009422516,0.0034933335,0.1360393,0.0019014447,0.00015345588,0.0000057344587,0.011013761,0.057363298,0.5059815,0.08031958,0.18928686,0.0050192033],"about_ca_topic_score_codex":0.3096522,"about_ca_topic_score_gemma":0.13066196,"teacher_disagreement_score":0.3406402,"about_ca_system_score_codex":0.00259437,"about_ca_system_score_gemma":0.002913091,"threshold_uncertainty_score":0.99990505},"labels":[],"label_agreement":null},{"id":"W2990689782","doi":"10.5753/cbie.sbie.2019.199","title":"Avaliação de Juízes: Um Modelo Estatístico para Perfilação de Avaliadores","year":2019,"lang":"pt","type":"article","venue":"Anais do XXX Simpósio Brasileiro de Informática na Educação (SBIE 2019)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Canadian Bureau for International Education","keywords":"Concordance; Competence (human resources); Statistics; Psychology; Mathematics; Computer science; Social psychology; Medicine","score_opus":0.0738775683116693,"score_gpt":0.35990389654441995,"score_spread":0.28602632823275065,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2990689782","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9566788,0.003147206,0.014201158,0.0034945535,0.004672976,0.0031998062,0.0005192597,0.00035989142,0.013726342],"genre_scores_gemma":[0.9623288,0.002266592,0.0051677027,0.0066702245,0.00057663926,0.0001935167,0.00009998245,0.000219855,0.022476722],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9798973,0.0014362595,0.0044088685,0.0027750842,0.006364209,0.0051182867],"domain_scores_gemma":[0.98460674,0.0042625107,0.001682395,0.004970495,0.0016177454,0.0028600902],"candidate_categories":["metaepi_narrow","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"category_scores_codex":[0.0152313765,0.0020834194,0.0024591554,0.0014121055,0.001232696,0.0032057443,0.0059897252,0.0016135753,0.017078822],"category_scores_gemma":[0.007865026,0.0018325456,0.0015819493,0.002443928,0.0009960536,0.0027371235,0.0017113031,0.0025431328,0.018170025],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012284426,0.0028264723,0.5936267,0.0010544448,0.0007809999,0.00015063104,0.02626781,0.012510697,0.003999447,0.0029835019,0.22697492,0.12759593],"study_design_scores_gemma":[0.0104124565,0.0049040834,0.30460447,0.0038724698,0.0019169948,0.0006154871,0.053513866,0.42131388,0.009353657,0.030806584,0.14914988,0.009536167],"about_ca_topic_score_codex":0.0017775528,"about_ca_topic_score_gemma":0.000120117715,"teacher_disagreement_score":0.4088032,"about_ca_system_score_codex":0.0022267634,"about_ca_system_score_gemma":0.006275301,"threshold_uncertainty_score":0.99975806},"labels":[],"label_agreement":null},{"id":"W2996204535","doi":"10.1364/josaa.381305","title":"Predicting the Farnsworth–Munsell D15 and Holmes–Wright-A lantern outcomes with computer-based color vision tests","year":2019,"lang":"en","type":"article","venue":"Journal of the Optical Society of America A","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Defence Research and Development Canada; University of Waterloo","funders":"Canadian Institute for Military and Veteran Health Research; Defence Research and Development Canada","keywords":"Contrast (vision); Wright; Mathematics; Artificial intelligence; Computer science; Art; Art history","score_opus":0.03117396616887044,"score_gpt":0.31091650768609497,"score_spread":0.27974254151722455,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2996204535","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9756011,0.00007522548,0.0064514377,0.017121071,0.00024447884,0.00025271584,0.0000031677187,0.0000054373504,0.00024538892],"genre_scores_gemma":[0.98176193,0.000028875784,0.016309917,0.0016823413,0.0000648291,9.99476e-7,1.1830598e-7,0.000006439504,0.0001445746],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99648815,0.00022778234,0.0007409654,0.00020494254,0.0021373131,0.00020086588],"domain_scores_gemma":[0.99559003,0.0025760802,0.00090144656,0.000427974,0.0004079861,0.00009648656],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0030788365,0.0001432308,0.00045806228,0.000024051493,0.00017263339,0.00014663168,0.0008812023,0.000053490003,0.000063253916],"category_scores_gemma":[0.00032205257,0.000051610095,0.00043605684,0.0003208433,0.0005241011,0.00014765465,0.00013291827,0.000376152,0.000010711344],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000247034,0.0005184167,0.9592103,0.00003904378,0.00023711218,0.0000023992015,0.0017662279,0.008599677,0.0027659063,0.00003927567,0.0064356425,0.020138979],"study_design_scores_gemma":[0.002329175,0.0035096076,0.876703,0.0004799932,0.0001591695,0.00004144936,0.0036637785,0.09793051,0.000875202,0.00089616096,0.0131665785,0.0002453222],"about_ca_topic_score_codex":0.000018873996,"about_ca_topic_score_gemma":0.0000022751901,"teacher_disagreement_score":0.08933084,"about_ca_system_score_codex":0.00003912965,"about_ca_system_score_gemma":0.000085681495,"threshold_uncertainty_score":0.21045986},"labels":[],"label_agreement":null},{"id":"W3006084164","doi":"10.1186/s13643-020-01291-z","title":"Inter-rater reliability and validity of risk of bias instrument for non-randomized studies of exposures: a study protocol","year":2020,"lang":"en","type":"article","venue":"Systematic Reviews","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Public Health Ontario; Queen's University; George & Fay Yee Centre for Healthcare Innovation; University of Toronto; University of Ottawa; St. Michael's Hospital; University of Manitoba","funders":"","keywords":"Medicine; Protocol (science); Concordance; Reliability (semiconductor); Concurrent validity; Inter-rater reliability; Medical physics; Statistic; Applied psychology; Medical education; Clinical psychology; Psychology; Statistics; Internal consistency; Psychometrics; Rating scale; Alternative medicine; Pathology","score_opus":0.5050424449137488,"score_gpt":0.47030656839311014,"score_spread":0.0347358765206387,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3006084164","genre_codex":"protocol","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4503136,0.0004034464,0.0054909675,0.0001655666,0.00014637863,0.54338765,0.000027101845,0.000006581005,0.00005870093],"genre_scores_gemma":[0.82433283,0.00010949651,0.0030217294,0.000022547361,0.000021189373,0.17246705,3.441913e-7,0.000006800704,0.000018009285],"study_design_codex":"systematic_review","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9770829,0.012172624,0.008302964,0.00060092704,0.0016795546,0.00016100495],"domain_scores_gemma":[0.97870666,0.012666111,0.0058620675,0.0012744307,0.0014027843,0.000087975175],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.09922087,0.0002746112,0.0051789167,0.000090726455,0.00005647149,0.000031771277,0.0005945094,0.000046802805,0.000031856573],"category_scores_gemma":[0.14794777,0.00012125325,0.00072143186,0.00038983295,0.0003527789,0.00013002357,0.00022861306,0.00008291111,0.0000071984005],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0110702105,0.0023232154,0.19280812,0.68804485,0.0012218452,6.232079e-7,0.09792402,0.00005707189,0.00056635763,0.00012927546,0.0022472816,0.0036071122],"study_design_scores_gemma":[0.41370818,0.030962005,0.01423851,0.1469221,0.008852904,0.0000054210086,0.2671936,0.011448091,0.021345718,0.072725415,0.010003375,0.0025946938],"about_ca_topic_score_codex":0.000025614361,"about_ca_topic_score_gemma":0.000016769287,"teacher_disagreement_score":0.5411228,"about_ca_system_score_codex":0.000029024626,"about_ca_system_score_gemma":0.00004634481,"threshold_uncertainty_score":0.9275417},"labels":[],"label_agreement":null},{"id":"W3007618168","doi":"10.48550/arxiv.2003.00316","title":"Model-based ROC (mROC) curve: examining the effect of case-mix and model calibration on the ROC plot","year":2020,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Receiver operating characteristic; Calibration; Smoothing; Plot (graphics); Computer science; Sample (material); Statistics; Calibration curve; Context (archaeology); Sensitivity (control systems); Artificial intelligence; Mathematics; Engineering","score_opus":0.358866620347382,"score_gpt":0.2718861623654973,"score_spread":0.0869804579818847,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3007618168","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7602907,0.000034679862,0.23735736,0.00075558276,0.00011234773,0.0007062683,0.00003948723,0.000035084802,0.00066848885],"genre_scores_gemma":[0.9992659,0.000024430547,0.00010556379,0.0002790601,0.000036941452,0.000005108368,0.000006577923,0.000015619973,0.00026080626],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99633723,0.0011905819,0.00047748254,0.0010913649,0.00066287356,0.00024044204],"domain_scores_gemma":[0.9946722,0.0031643012,0.0005105396,0.0013350375,0.0001948435,0.00012307717],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004215191,0.0003484662,0.00047027078,0.0001504951,0.00041512487,0.00019468946,0.0012982996,0.00021183986,0.000042129046],"category_scores_gemma":[0.0012277195,0.0002003275,0.00022682334,0.0005323882,0.0002866256,0.00018306027,0.0007509814,0.0006692523,0.000016221617],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00021354848,0.000035573827,0.0038782482,0.000052541218,0.00004602445,0.00004667154,0.00034261242,0.99004054,0.0001558008,0.004273689,0.00057883223,0.00033594453],"study_design_scores_gemma":[0.00043020022,0.00025635966,0.00020596723,0.00008758046,0.00013562482,0.0000017367046,0.00030517948,0.9689771,0.00097632065,0.028393988,0.000026515785,0.00020340584],"about_ca_topic_score_codex":0.00006566005,"about_ca_topic_score_gemma":0.000076433316,"teacher_disagreement_score":0.2389752,"about_ca_system_score_codex":0.0001054612,"about_ca_system_score_gemma":0.00018210053,"threshold_uncertainty_score":0.8169118},"labels":[],"label_agreement":null},{"id":"W3010669063","doi":"","title":"Blinding in Rehabilitation Research Empirical Evidence on the Association Between Blinding and Treatment Effect Estimates","year":2020,"lang":"en","type":"article","venue":"STM:n Hallinnonalan avoin julkaisuarkisto (Julkari)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Blinding; Meta-analysis; Rehabilitation; Association (psychology); Sample size determination; Clinical study design; Randomized controlled trial; Medicine; Systematic review; Psychology; Clinical trial; MEDLINE; Clinical psychology; Physical therapy; Statistics; Biology; Internal medicine; Mathematics; Psychotherapist","score_opus":0.5791521620950734,"score_gpt":0.5027552196974594,"score_spread":0.07639694239761396,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3010669063","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9302711,0.00032630755,0.00024945862,0.06569891,0.00020010826,0.001722458,0.000025619256,0.000067495,0.0014385454],"genre_scores_gemma":[0.99781877,0.000094964824,0.00062461453,0.0005432255,0.00039505836,0.00019909136,0.000016509333,0.000029184102,0.00027856085],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9863613,0.0046572504,0.0015494286,0.0015392464,0.005012524,0.00088019704],"domain_scores_gemma":[0.91447234,0.08336877,0.0004769828,0.00077072304,0.00058925344,0.00032193906],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.039516527,0.00046616304,0.00085264264,0.0006242577,0.00082624267,0.00082497543,0.0010265436,0.00028395522,0.00019231062],"category_scores_gemma":[0.06539963,0.0002833048,0.00024917457,0.002628044,0.00033266743,0.0004826537,0.0002760758,0.0010921312,0.0005978394],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00033891632,0.00016456046,0.9807471,0.000042796863,0.0000646846,0.0000116459105,0.0041686404,0.0002707707,0.0006827292,0.0009157795,0.0037253373,0.008867062],"study_design_scores_gemma":[0.002271129,0.007079576,0.9386482,0.0008265196,0.000113884365,0.0000023305395,0.0023118476,0.006205062,0.0026484723,0.023576826,0.015614847,0.0007012755],"about_ca_topic_score_codex":0.0001367075,"about_ca_topic_score_gemma":0.00018182119,"teacher_disagreement_score":0.078711525,"about_ca_system_score_codex":0.0015955778,"about_ca_system_score_gemma":0.00023410519,"threshold_uncertainty_score":0.9999619},"labels":[],"label_agreement":null},{"id":"W3018802573","doi":"10.1007/s42979-020-00129-8","title":"Detection of Opinion Communities with the Help of Chance-Corrected Measures of Agreement","year":2020,"lang":"en","type":"article","venue":"SN Computer Science","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland","funders":"","keywords":"Jaccard index; Similarity (geometry); Scale (ratio); Cosine similarity; Computer science; Index (typography); Information retrieval; Artificial intelligence; Statistics; Psychology; Econometrics; Mathematics; Pattern recognition (psychology); World Wide Web; Image (mathematics)","score_opus":0.2141702298403869,"score_gpt":0.32400120330889387,"score_spread":0.10983097346850698,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3018802573","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6562468,0.00010245054,0.34112242,0.0015212706,0.00044544853,0.00029347252,0.000004531529,0.00001069453,0.0002529498],"genre_scores_gemma":[0.9973987,0.000016975724,0.0023608706,0.00016378495,0.000046791174,0.000005058973,3.0417803e-7,0.0000021744356,0.0000053446192],"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9958013,0.0002728843,0.0005573109,0.00026047815,0.0029404152,0.00016758603],"domain_scores_gemma":[0.9970904,0.00040831362,0.0005037634,0.00065079523,0.00128613,0.000060619728],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0044362284,0.00010368714,0.0002668849,0.00013666446,0.00019276449,0.000058975536,0.0017372756,0.000021415066,0.000026760254],"category_scores_gemma":[0.00023595056,0.00005546284,0.00006753669,0.00185623,0.0011752242,0.00025575608,0.00034363565,0.00010434604,0.0000042508473],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00044218116,0.00048468824,0.032437295,0.00018824605,0.000074071366,6.192566e-7,0.04810786,0.031544935,0.31616703,0.0012114246,0.0013007317,0.5680409],"study_design_scores_gemma":[0.0008281984,0.0033997365,0.098318055,0.000312522,0.000020329542,0.0000033372703,0.008081291,0.12118156,0.7634461,0.0013708317,0.0027610234,0.00027702848],"about_ca_topic_score_codex":0.00014313801,"about_ca_topic_score_gemma":0.00009028995,"teacher_disagreement_score":0.56776386,"about_ca_system_score_codex":0.000022373655,"about_ca_system_score_gemma":0.00013083017,"threshold_uncertainty_score":0.43301624},"labels":[],"label_agreement":null},{"id":"W3022460225","doi":"10.1002/0470011815.b2a04024","title":"Kappa and Its Dependence on Marginal Rates","year":2005,"lang":"en","type":"other","venue":"Encyclopedia of Biostatistics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Kappa; Cohen's kappa; Statistic; Statistics; Mathematics; Measure (data warehouse); Agreement; Econometrics; Computer science; Data mining; Geometry","score_opus":0.060338978963988564,"score_gpt":0.35817373683181414,"score_spread":0.2978347578678256,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3022460225","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00059962086,0.002736177,0.0015155989,0.00044578937,0.0009893283,0.00052711426,0.0009949973,0.00003972442,0.9921517],"genre_scores_gemma":[0.0055639227,0.010205446,0.012771881,0.00023092811,0.0006464024,0.000016040643,0.000032374817,0.00013748217,0.9703955],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9962557,0.0001735016,0.00070435373,0.0006166667,0.0019904368,0.00025936108],"domain_scores_gemma":[0.9974045,0.0011658134,0.00057925255,0.0005307745,0.00018312508,0.00013650005],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0013354387,0.0003030687,0.0004971021,0.00032266544,0.000054252603,0.00006546486,0.0006036612,0.00021364822,0.008525331],"category_scores_gemma":[0.002889994,0.00022463477,0.000061180755,0.0002493523,0.00015686294,0.000048073307,0.00011691441,0.00023552,0.0007559236],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000029308787,0.00013487993,0.001593471,0.000080800404,0.000031210315,0.000016531705,0.00010427659,0.000011752573,0.000023263092,0.006403362,0.90685123,0.0847199],"study_design_scores_gemma":[0.00026940243,0.00021742539,0.003098212,0.00021339393,0.0000359179,0.0000023365778,0.00008105256,0.00013385247,0.000098468925,0.0035204163,0.9920317,0.00029783975],"about_ca_topic_score_codex":0.000034352935,"about_ca_topic_score_gemma":0.00013667681,"teacher_disagreement_score":0.08518044,"about_ca_system_score_codex":0.000026794442,"about_ca_system_score_gemma":0.000101099155,"threshold_uncertainty_score":0.99238104},"labels":[],"label_agreement":null},{"id":"W3022717317","doi":"10.1002/0470011815.b2a04043","title":"Receiver Operating Characteristic (<scp>ROC</scp>) Curves","year":2005,"lang":"en","type":"other","venue":"Encyclopedia of Biostatistics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Receiver operating characteristic; Fraction (chemistry); Mathematics; Boundary (topology); Statistics; Mathematical analysis; Chromatography; Chemistry","score_opus":0.04874646265946651,"score_gpt":0.3331996619537859,"score_spread":0.2844531992943194,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3022717317","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0002911023,0.004925496,0.0058659604,0.0003117775,0.002399694,0.0009036938,0.002924712,0.0001074416,0.9822701],"genre_scores_gemma":[0.00031656888,0.023142323,0.022993818,0.0005999686,0.0012613591,0.00003342927,0.00026785015,0.00024452718,0.95114017],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9938176,0.00038831605,0.0016291514,0.00087908155,0.0028076558,0.00047814372],"domain_scores_gemma":[0.9942487,0.0025283066,0.0014199485,0.0011385113,0.00045724356,0.00020732582],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0024160605,0.00050586637,0.0009927746,0.00040643563,0.00010409201,0.00012158366,0.0012515731,0.00033102298,0.010397951],"category_scores_gemma":[0.017666979,0.0003983369,0.00017605619,0.0005365877,0.00028817865,0.000109991845,0.00021749342,0.00040721125,0.0015342133],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000002071621,0.00012790729,0.001495549,0.00025883154,0.000049732695,0.000011836373,0.00023050503,0.000005875593,0.00003681896,0.0003085749,0.96152335,0.035948917],"study_design_scores_gemma":[0.00022034207,0.0001130378,0.0054666772,0.000965992,0.00008779451,0.0000025020697,0.00018582134,0.00008174813,0.000033902004,0.00060067116,0.9920383,0.00020322447],"about_ca_topic_score_codex":0.00008340244,"about_ca_topic_score_gemma":0.000109998575,"teacher_disagreement_score":0.035745695,"about_ca_system_score_codex":0.00006323393,"about_ca_system_score_gemma":0.0002306119,"threshold_uncertainty_score":0.9998469},"labels":[],"label_agreement":null},{"id":"W3024254388","doi":"10.1017/cem.2020.114","title":"LO59: Reliability of patient reported exposure and outcome data in a prospective cohort study of older adults presenting to the emergency department","year":2020,"lang":"en","type":"article","venue":"Canadian Journal of Emergency Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Medicine; Emergency department; Medical record; Prospective cohort study; Confidence interval; Observational study; Kappa; Emergency medicine; Cohort study; Cohort; Cohen's kappa; Medical emergency; Pediatrics; Internal medicine; Psychiatry","score_opus":0.20366690364353798,"score_gpt":0.398192247962641,"score_spread":0.19452534431910304,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3024254388","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98957753,0.0008053552,0.000060552764,0.005616317,0.0018494015,0.0018734741,0.00002649126,0.0000017647802,0.0001891326],"genre_scores_gemma":[0.9996213,0.00007022026,0.000047576465,0.00005759659,0.00014721975,0.000015732065,0.0000026660086,0.000007368081,0.000030359248],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9923614,0.00048375927,0.004236004,0.0005488519,0.0020955608,0.00027437755],"domain_scores_gemma":[0.99483764,0.00018226753,0.0016288497,0.0011434635,0.001584962,0.00062279333],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009045641,0.00017961544,0.000739764,0.00024844747,0.000095646916,0.00000665919,0.0010942804,0.000043556312,0.0012743047],"category_scores_gemma":[0.022902988,0.000101673875,0.000080259655,0.0012845526,0.00009031198,0.00020279147,0.00016427001,0.00027389533,0.0000014881604],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000081992286,0.000196306,0.9683198,0.000040161427,0.00007068539,0.000026414535,0.02077779,0.00018800067,0.000020520914,0.0000034797679,0.009388499,0.0008863524],"study_design_scores_gemma":[0.0007443292,0.0013242511,0.98084575,0.000112335234,0.000091698006,0.0000021018693,0.014964549,0.00012193812,0.00001761164,0.00019151531,0.0014932473,0.00009065632],"about_ca_topic_score_codex":0.0074015153,"about_ca_topic_score_gemma":0.04559046,"teacher_disagreement_score":0.038188946,"about_ca_system_score_codex":0.00007623745,"about_ca_system_score_gemma":0.00028652887,"threshold_uncertainty_score":0.9996387},"labels":[],"label_agreement":null},{"id":"W3035592618","doi":"10.1186/s12874-020-01022-x","title":"Using multiple agreement methods for continuous repeated measures data: a tutorial for practitioners","year":2020,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":88,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Emory University","keywords":"Concordance; Cohen's kappa; Computer science; Concordance correlation coefficient; Data mining; Statistics; Strengths and weaknesses; Correlation coefficient; Pulmonary disease; Limits of agreement; Medical physics; Medicine; Mathematics; Machine learning; Psychology","score_opus":0.9644013985579888,"score_gpt":0.7181859517571121,"score_spread":0.2462154468008767,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3035592618","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0024845893,0.0004510766,0.9788967,0.012543528,0.0021252527,0.0031028646,0.00012332706,0.00005018777,0.00022247047],"genre_scores_gemma":[0.005849201,0.00007368642,0.9896866,0.0011057194,0.0024477977,0.00047534003,0.00008486683,0.000027621758,0.00024915877],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9347641,0.051040456,0.0023333617,0.0024123206,0.008000678,0.001449097],"domain_scores_gemma":[0.44727448,0.53967786,0.00089552125,0.0032219994,0.0068131993,0.0021169551],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.34703776,0.0002609011,0.0011267075,0.00031716868,0.000635898,0.0002511844,0.0032971671,0.00044785696,0.00080968585],"category_scores_gemma":[0.8918822,0.00018645363,0.00031351112,0.0010767526,0.0008022512,0.00034991544,0.0012073916,0.00078650087,0.00004570411],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009065401,0.00054224493,0.002173008,0.00036202426,0.00033551024,0.000016611159,0.0016001738,0.00051658956,0.057104804,0.005255338,0.12332341,0.7997049],"study_design_scores_gemma":[0.0039503546,0.0011122961,0.000088039764,0.000056513923,0.00006757223,0.000008103986,0.0027997934,0.2378906,0.004435614,0.032240424,0.71708375,0.00026696088],"about_ca_topic_score_codex":0.0002758091,"about_ca_topic_score_gemma":0.00034063737,"teacher_disagreement_score":0.79943794,"about_ca_system_score_codex":0.0001853395,"about_ca_system_score_gemma":0.0025231843,"threshold_uncertainty_score":0.88654923},"labels":[],"label_agreement":null},{"id":"W3040882905","doi":"10.1164/ajrccm-conference.2020.201.1_meetingabstracts.a2632","title":"Intra- and Inter-Rater Reliability of Pathologic Classification of Type of Progressive Massive Fibrosis Among Deceased U.S. Coal Miners","year":2020,"lang":"en","type":"article","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Reliability (semiconductor); Coal; Coal field; Fibrosis; Field (mathematics); Pathology; Computer science; Medicine; Engineering; Coal mining; Mathematics; Physics; Power (physics)","score_opus":0.1467529835156056,"score_gpt":0.35203900615208683,"score_spread":0.20528602263648124,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3040882905","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9950088,0.000059223916,0.0017866305,0.0017088035,0.000072128525,0.00041836887,0.00002243928,0.000010064356,0.000913528],"genre_scores_gemma":[0.9971667,0.000011187682,0.0026801312,0.00007277692,0.000011727974,0.000009347858,0.0000047604385,0.0000035742175,0.000039786737],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9973584,0.00032007354,0.00092000916,0.0004268376,0.00085210695,0.00012259037],"domain_scores_gemma":[0.9973851,0.00050668756,0.0006883752,0.0004140417,0.00091251184,0.00009330869],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016554273,0.00012123215,0.0003999737,0.000074421456,0.00002022704,0.000023460054,0.00035708898,0.00009047683,0.00036062745],"category_scores_gemma":[0.0035437054,0.000074755204,0.00009307692,0.00041836785,0.0006640413,0.0001707949,0.000119731885,0.00008468785,0.000005446533],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003213414,0.00017592467,0.92730635,0.00010547048,0.000016097927,0.0000012653147,0.0009091255,0.000081628816,0.06529099,0.00015852954,0.0021342437,0.0034990066],"study_design_scores_gemma":[0.0006997272,0.0013984586,0.84321916,0.00009645229,0.000049627528,4.49296e-7,0.006408482,0.0055066557,0.13884844,0.0033174679,0.0002775958,0.00017749632],"about_ca_topic_score_codex":0.0000339316,"about_ca_topic_score_gemma":0.000009424969,"teacher_disagreement_score":0.08408724,"about_ca_system_score_codex":0.000023886138,"about_ca_system_score_gemma":0.00005089105,"threshold_uncertainty_score":0.42424026},"labels":[],"label_agreement":null},{"id":"W3041465882","doi":"10.1161/strokeaha.120.030252","title":"Probabilistic Reasoning in Stroke","year":2020,"lang":"en","type":"review","venue":"Stroke","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Population Health Research Institute","funders":"","keywords":"Medicine; Stroke (engine); Probabilistic logic; Artificial intelligence","score_opus":0.24299888323263313,"score_gpt":0.4225504724540668,"score_spread":0.17955158922143366,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3041465882","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000016828944,0.98132557,0.00011016774,0.00019421797,0.00070022064,0.0013104187,0.00018756639,0.0000462252,0.016108785],"genre_scores_gemma":[0.00006444659,0.99459684,0.00067142566,0.000055221706,0.00028384372,0.00016512147,0.000016820677,0.000030072655,0.0041161967],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9922948,0.0010180408,0.0020768428,0.001300245,0.002789768,0.00052033906],"domain_scores_gemma":[0.99610186,0.0016492638,0.00068818394,0.0011772056,0.00016644051,0.00021704189],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.005644139,0.00049777015,0.0023624029,0.0004023821,0.00009688563,0.00036653306,0.0018679146,0.00029124646,0.000742085],"category_scores_gemma":[0.010424555,0.0003330415,0.0008043288,0.0010635068,0.00011400779,0.00015470362,0.00037273444,0.0009121207,0.0017909221],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000004852705,0.000054492306,0.00012635294,0.0012169102,0.00003133691,0.00003276986,0.00006763765,0.000014366581,3.2425194e-7,0.0007098802,0.002060475,0.99568063],"study_design_scores_gemma":[0.00014134526,0.000073848736,0.000028871393,0.0035076179,0.000108975204,0.0000050889576,0.000086288135,0.00006019017,3.4885423e-7,0.0013531974,0.9943263,0.0003079536],"about_ca_topic_score_codex":0.000022784208,"about_ca_topic_score_gemma":0.00007094188,"teacher_disagreement_score":0.99537265,"about_ca_system_score_codex":0.0003614845,"about_ca_system_score_gemma":0.00056499755,"threshold_uncertainty_score":0.99991214},"labels":[],"label_agreement":null},{"id":"W3044355609","doi":"10.1177/1055665620944053","title":"Response to “Nasalance-Based Preclassification of Oral–Nasal Balance Disorders Results in Higher Agreement of Expert Listeners: Methodological Issue”","year":2020,"lang":"en","type":"letter","venue":"The Cleft Palate-Craniofacial Journal","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; University of British Columbia","funders":"","keywords":"Audiology; Psychology; Balance (ability); Nasal passages; Medicine; Nose; Neuroscience; Anatomy","score_opus":0.33152537935838916,"score_gpt":0.4083021269548875,"score_spread":0.07677674759649833,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3044355609","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.030612305,0.000036097426,0.0026064713,0.96199423,0.002551694,0.0011604638,0.00017204534,0.000025755551,0.0008409438],"genre_scores_gemma":[0.6036978,0.00060176145,0.015246306,0.37297565,0.0038126523,0.0001335171,0.000073791154,0.000116575604,0.003341975],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98061234,0.009116141,0.0036668403,0.0010335098,0.004874637,0.00069652323],"domain_scores_gemma":[0.99030304,0.0050035757,0.002337093,0.0013535152,0.00075289153,0.0002498589],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02285797,0.00057197345,0.0015277484,0.00049846165,0.00023521265,0.00022294078,0.003056975,0.00060251565,0.0013035187],"category_scores_gemma":[0.011577601,0.00034809113,0.0006443077,0.0013557491,0.00052223494,0.00019255809,0.00025205332,0.0021263366,0.000118088414],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009012403,0.0003747381,0.010239986,0.000063586274,0.000121255514,0.00007975773,0.00276992,0.003504966,0.014900148,0.000012683295,0.9496138,0.009306721],"study_design_scores_gemma":[0.0026251546,0.00093051704,0.20637147,0.00043878032,0.000057687208,0.0000046872624,0.00055575377,0.00064154033,0.00024198978,0.0041298186,0.7834558,0.0005467809],"about_ca_topic_score_codex":0.00009767025,"about_ca_topic_score_gemma":0.000030176814,"teacher_disagreement_score":0.5890186,"about_ca_system_score_codex":0.0002956218,"about_ca_system_score_gemma":0.0003976785,"threshold_uncertainty_score":0.9998971},"labels":[],"label_agreement":null},{"id":"W3047914353","doi":"10.1093/cid/ciaa1131","title":"Latent Class Analysis and the Need for Clear Reporting of Methods","year":2020,"lang":"en","type":"letter","venue":"Clinical Infectious Diseases","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University; McGill University Health Centre","funders":"","keywords":"Medicine; Biostatistics; Library science; Epidemiology; Gerontology; Media studies; Pathology; Sociology","score_opus":0.38443529084602557,"score_gpt":0.5288713261281581,"score_spread":0.14443603528213256,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3047914353","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01782096,0.0012671317,0.048753325,0.92641664,0.0018752924,0.002541906,0.00020071273,0.00013702361,0.0009870367],"genre_scores_gemma":[0.15007348,0.000625129,0.002175946,0.84103733,0.0046638744,0.00032194357,0.00012881188,0.000061992345,0.0009114853],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.987672,0.0036072456,0.006172266,0.0010563125,0.0012003403,0.00029187265],"domain_scores_gemma":[0.95175135,0.03900687,0.006846599,0.0012575425,0.0009494459,0.00018819631],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.020575557,0.0002915068,0.0022980492,0.00022753762,0.0002447499,0.00028557237,0.0006721254,0.00038620003,0.00016000812],"category_scores_gemma":[0.1678195,0.00015583503,0.0028964684,0.0011376672,0.0010737951,0.000077262135,0.0003184497,0.00095050473,0.000012993343],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018883824,0.00014879565,0.27754804,0.0001228365,0.0020300401,0.0000119393435,0.000043456086,0.00013915052,0.0000016465252,0.0000983786,0.69588983,0.023777017],"study_design_scores_gemma":[0.0026817478,0.0007868209,0.1107376,0.00007903828,0.011701621,0.0000016253372,0.00012988613,0.008943337,0.000013224793,0.12104292,0.74334943,0.00053275685],"about_ca_topic_score_codex":0.00004306191,"about_ca_topic_score_gemma":0.000012301042,"teacher_disagreement_score":0.16681044,"about_ca_system_score_codex":0.000030629722,"about_ca_system_score_gemma":0.00014205265,"threshold_uncertainty_score":0.8391903},"labels":[],"label_agreement":null},{"id":"W3088642850","doi":"10.1016/j.jclinepi.2020.09.033","title":"Methodologically rigorous risk of bias tools for nonrandomized studies had low reliability and high evaluator burden","year":2020,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University; Université de Montréal; University of Ottawa; Institute of Health Economics; St. Michael's Hospital; University of Manitoba; Centre Hospitalier de l’Université de Montréal; University of Toronto; George & Fay Yee Centre for Healthcare Innovation","funders":"","keywords":"Inter-rater reliability; Reliability (semiconductor); Medicine; Psychological intervention; Psychology; Physical therapy; Statistics; Mathematics; Psychiatry; Rating scale","score_opus":0.8644908462866121,"score_gpt":0.6267974750429381,"score_spread":0.23769337124367396,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3088642850","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.851627,0.0019379483,0.09220515,0.052118048,0.0013186838,0.00072044524,0.000033755936,0.000008378008,0.000030613584],"genre_scores_gemma":[0.75040525,0.0055973292,0.23856825,0.0040881,0.0012848717,0.00001797531,8.572713e-7,0.000008997172,0.000028346114],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9162431,0.064301066,0.016478375,0.00097118696,0.0015318675,0.00047444247],"domain_scores_gemma":[0.15772684,0.823528,0.012990868,0.0007457848,0.0043988535,0.0006096778],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.45571566,0.0002570131,0.006808425,0.00008502175,0.000117167736,0.000032635777,0.0009845854,0.0004024046,0.00019643028],"category_scores_gemma":[0.9642429,0.00012934404,0.0017763312,0.00025044483,0.0011299124,0.00025886073,0.00027165943,0.00089922623,0.000012073047],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.054234277,0.0009717579,0.36623833,0.0002807779,0.0021735588,0.000016375723,0.0009995799,0.02002593,0.00043003025,0.0023038087,0.030974247,0.52135134],"study_design_scores_gemma":[0.02115833,0.011200879,0.18096627,0.00013945477,0.00078712037,0.000012324321,0.0010201712,0.008884717,0.00013386276,0.75783235,0.017572844,0.00029165862],"about_ca_topic_score_codex":0.000017205151,"about_ca_topic_score_gemma":0.000004412062,"teacher_disagreement_score":0.7592269,"about_ca_system_score_codex":0.000031946336,"about_ca_system_score_gemma":0.00023975883,"threshold_uncertainty_score":0.56045526},"labels":[],"label_agreement":null},{"id":"W3090901332","doi":"10.24095/hpcdp.34.4.09f","title":"Concordance entre les résultats d'une enquête et les données de la Régie de l’assurance maladie du Québec (RAMQ) pour le diagnostic d’asthme et pour l’utilisation des services médicaux pour asthme chez les enfants","year":2014,"lang":"fr","type":"article","venue":"Maladies chroniques et blessures au Canada","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":true,"ca_institutions":"McGill University; Université de Montréal; Hôpital Notre-Dame","funders":"","keywords":"Medicine; Concordance; Humanities; Gynecology; Political science; Philosophy","score_opus":0.029104492410963013,"score_gpt":0.28897892303195083,"score_spread":0.2598744306209878,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3090901332","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.85240483,0.019247524,0.0149342995,0.09582956,0.0007201787,0.0009161613,0.00043483108,0.00020681383,0.015305787],"genre_scores_gemma":[0.9707424,0.01136026,0.006410303,0.0030370015,0.00067766436,0.00022970913,0.000045273413,0.00012229914,0.007375059],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98608875,0.0064165834,0.0017148085,0.0016059775,0.0021749733,0.001998877],"domain_scores_gemma":[0.9775791,0.018508438,0.0011405771,0.001057993,0.0009744679,0.0007394058],"candidate_categories":["metaepi_narrow","sts","scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.010694939,0.0012130381,0.0013500872,0.00025515695,0.0015298134,0.0010863743,0.001993938,0.0006106191,0.0019021966],"category_scores_gemma":[0.0063985107,0.0011323609,0.00037653852,0.0005526314,0.0020190764,0.0010837861,0.00047255386,0.0011353983,0.000055724544],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00021385946,0.0006359274,0.84077096,0.0023959249,0.00037287254,0.00020597695,0.007628983,0.038932823,0.0023898093,0.01162617,0.03627234,0.058554363],"study_design_scores_gemma":[0.0018492423,0.00025485927,0.6076369,0.004393117,0.00020473749,0.00012848897,0.009228136,0.012919008,0.0074401596,0.009736919,0.3449416,0.001266837],"about_ca_topic_score_codex":0.9749329,"about_ca_topic_score_gemma":0.9971633,"teacher_disagreement_score":0.30866924,"about_ca_system_score_codex":0.002817787,"about_ca_system_score_gemma":0.02284893,"threshold_uncertainty_score":0.9999506},"labels":[],"label_agreement":null},{"id":"W3140874822","doi":"10.1177/1971400920937403","title":"Reply: “Computed tomography assessment of anterior ethmoidal canal dehiscence: Methodological issue on interobserver agreement”","year":2020,"lang":"en","type":"letter","venue":"The Neuroradiology Journal","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Computed tomography; Dehiscence; Medicine; Nuclear medicine; Radiology; Anatomy; Surgery","score_opus":0.31662526684177866,"score_gpt":0.4314452370462024,"score_spread":0.11481997020442375,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3140874822","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02918821,0.00029842262,0.005629912,0.9562484,0.006910995,0.00066839205,0.0000673486,0.00003819172,0.000950128],"genre_scores_gemma":[0.03558335,0.00022085957,0.0045792223,0.9515916,0.0073585487,0.000033538035,0.000017296381,0.00004289019,0.00057266036],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9762279,0.014928249,0.002934181,0.0013418746,0.003724502,0.0008433045],"domain_scores_gemma":[0.9878963,0.0068610217,0.0026801769,0.0015847071,0.00073231146,0.00024547472],"candidate_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.014882781,0.00069646526,0.0019388164,0.00047573703,0.00054246373,0.00032871438,0.004933214,0.0005706736,0.0024407671],"category_scores_gemma":[0.004134858,0.00036613183,0.0012194066,0.0009194869,0.0010537427,0.00014200177,0.00061399944,0.0076898634,0.00010486736],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017280698,0.00009185299,0.008832217,0.000028566461,0.00031165345,0.00082542794,0.00019048518,0.00036412803,0.0009919167,0.000022605926,0.98404324,0.0041250847],"study_design_scores_gemma":[0.0009099668,0.004042731,0.0612477,0.0002487269,0.00024399682,0.0010633918,0.00015150834,0.0008254862,0.00020748202,0.008045859,0.9224649,0.0005482601],"about_ca_topic_score_codex":0.00004315321,"about_ca_topic_score_gemma":0.000007554392,"teacher_disagreement_score":0.061578363,"about_ca_system_score_codex":0.0001696877,"about_ca_system_score_gemma":0.00034247292,"threshold_uncertainty_score":0.99987906},"labels":[],"label_agreement":null},{"id":"W3151387929","doi":"10.1093/aje/kwab068","title":"Invited Commentary: Toward Better Bias Analysis","year":2021,"lang":"en","type":"letter","venue":"American Journal of Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Publication bias; Econometrics; Psychology; Computer science; Statistics; MEDLINE; Mathematics; Political science; Law","score_opus":0.3509040920935143,"score_gpt":0.4276153762947408,"score_spread":0.07671128420122647,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3151387929","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017728774,0.0009841122,0.014845959,0.9648915,0.0013135695,0.00011273141,0.000053628475,0.000008507122,0.00006120992],"genre_scores_gemma":[0.0035175027,0.0003908051,0.011248723,0.98100555,0.0036229696,0.0000054590714,0.00011453066,0.000023175166,0.00007128241],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9730791,0.017580325,0.005307116,0.0009426646,0.0022393588,0.0008514378],"domain_scores_gemma":[0.9543814,0.034203697,0.008115934,0.0015156162,0.0015124689,0.00027089208],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.038421642,0.0005130687,0.0048834714,0.0018272826,0.00011524425,0.000096234944,0.0024944015,0.00045108068,0.0014201647],"category_scores_gemma":[0.03422911,0.00032996986,0.0026426925,0.0032876928,0.0012196578,0.00017668081,0.00031241417,0.0034450297,0.000050973256],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003071941,0.00004585648,0.1702338,0.0000061906353,0.002290714,0.00031078613,0.00011562382,0.00036151995,0.0000058133787,0.0000016355594,0.8084147,0.018182656],"study_design_scores_gemma":[0.00022133128,0.00067354576,0.011247794,0.0000519355,0.0013276775,0.00007847767,0.00048663016,0.00011592021,0.0000068505806,0.0063754427,0.9790916,0.00032281523],"about_ca_topic_score_codex":0.00047462655,"about_ca_topic_score_gemma":0.000039885334,"teacher_disagreement_score":0.1706769,"about_ca_system_score_codex":0.00029321495,"about_ca_system_score_gemma":0.00008701113,"threshold_uncertainty_score":0.99991524},"labels":[],"label_agreement":null},{"id":"W3160029048","doi":"10.1016/j.jclinepi.2021.04.017","title":"Customized guidance/training improved the psychometric properties of methodologically rigorous risk of bias instruments for non-randomized studies","year":2021,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University; Université de Montréal; University of Ottawa; Institute of Health Economics; St. Michael's Hospital; University of Manitoba; Centre Hospitalier de l’Université de Montréal; University of Toronto; George & Fay Yee Centre for Healthcare Innovation","funders":"","keywords":"Medicine; Randomized controlled trial; Reliability (semiconductor); Physical therapy; Statistic; Medical physics; Psychology; Statistics; Surgery","score_opus":0.8685239314559975,"score_gpt":0.6052991067119616,"score_spread":0.26322482474403586,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3160029048","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92130476,0.0063944915,0.059939843,0.008230703,0.0031535176,0.0008202739,0.000011734339,0.000004353143,0.00014029456],"genre_scores_gemma":[0.8447378,0.007174741,0.14642188,0.0011447329,0.00035587416,0.000034695528,3.376886e-7,0.000007966535,0.000121950754],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9197899,0.05810966,0.019586395,0.00063931127,0.0013792216,0.00049549463],"domain_scores_gemma":[0.31797293,0.64055556,0.030697273,0.0013556895,0.009134057,0.00028445717],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4737074,0.00024910225,0.008235257,0.00027479243,0.00014173203,0.000018038796,0.0013133329,0.00031314022,0.000057078865],"category_scores_gemma":[0.9235801,0.000099661986,0.0031655477,0.0008404949,0.0020874017,0.00014162176,0.00024299712,0.0008086676,0.0000029682787],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.15415882,0.0016346296,0.24556921,0.00025886757,0.007193858,0.0000050775516,0.0026465713,0.0026762492,0.00460898,0.0017209197,0.005271794,0.57425505],"study_design_scores_gemma":[0.2052322,0.0053784307,0.09562595,0.00096006045,0.0019423515,0.00005326868,0.01178065,0.011588357,0.004111021,0.65589124,0.0069460697,0.00049041095],"about_ca_topic_score_codex":0.000016385859,"about_ca_topic_score_gemma":0.000011107014,"teacher_disagreement_score":0.6541703,"about_ca_system_score_codex":0.000033540353,"about_ca_system_score_gemma":0.0004686317,"threshold_uncertainty_score":0.7691118},"labels":[],"label_agreement":null},{"id":"W3161848643","doi":"10.1016/j.jval.2021.02.011","title":"Bias Assessment in Outcomes Research: The Role of Relative Versus Absolute Approaches","year":2021,"lang":"en","type":"article","venue":"Value in Health","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":29,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"Qatar National Research Fund; Fonds National de la Recherche Luxembourg; Australian National University; Qatar Foundation","keywords":"Consistency (knowledge bases); Independence (probability theory); Statistics; Scale (ratio); Relative value; Intraclass correlation; Econometrics; Quality (philosophy); Psychology; Mathematics; Computer science; Artificial intelligence; Psychometrics; Geography","score_opus":0.8214349676296412,"score_gpt":0.5378024960010247,"score_spread":0.2836324716286165,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3161848643","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9404354,0.0021835808,0.00013096386,0.0346331,0.0005265034,0.0008538275,0.000012077556,0.000008981463,0.021215549],"genre_scores_gemma":[0.99749106,0.0001597937,0.0018869718,0.00010040933,0.000022057218,0.00003762318,0.0000016174664,0.00000523993,0.00029519742],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9889509,0.0057198326,0.001276462,0.00050404796,0.00309403,0.0004547051],"domain_scores_gemma":[0.98822504,0.010273624,0.00026890045,0.0008696183,0.0002789708,0.000083825325],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.046520896,0.00010190754,0.0003727464,0.00027916604,0.00015256005,0.00007057689,0.00062378775,0.000061406456,0.000091948044],"category_scores_gemma":[0.008858708,0.000060754865,0.00009629849,0.0017889059,0.00018090085,0.000157542,0.00024474206,0.0006471804,0.00003329663],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013263733,0.0009929192,0.754216,0.000064990214,0.000038345046,0.000010725178,0.015360805,0.03518652,0.00005439245,0.14959,0.00065439293,0.043698248],"study_design_scores_gemma":[0.0011829353,0.00030252017,0.6536033,0.00016323345,0.0000026584353,6.2666516e-7,0.04187143,0.019431397,0.00021525755,0.27353066,0.009577093,0.000118895114],"about_ca_topic_score_codex":0.0031390954,"about_ca_topic_score_gemma":0.0013369379,"teacher_disagreement_score":0.12394065,"about_ca_system_score_codex":0.00042834162,"about_ca_system_score_gemma":0.0010868073,"threshold_uncertainty_score":0.9994901},"labels":[],"label_agreement":null},{"id":"W3172444956","doi":"10.1109/access.2021.3084050","title":"The Matthews Correlation Coefficient (MCC) is More Informative Than Cohen’s Kappa and Brier Score in Binary Classification Assessment","year":2021,"lang":"en","type":"article","venue":"IEEE Access","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":429,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Brier score; Kappa; Cohen's kappa; Correlation; Computer science; Statistics; Artificial intelligence; Binary classification; Metric (unit); Binary number; Odds; Machine learning; Mathematics; Natural language processing; Logistic regression; Support vector machine; Arithmetic","score_opus":0.18775632908039266,"score_gpt":0.4289589341158113,"score_spread":0.24120260503541865,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3172444956","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97914374,0.00016071853,0.0086197965,0.0072176084,0.00080750056,0.0005620633,0.00001814902,0.000017264081,0.0034531697],"genre_scores_gemma":[0.9980797,0.00013871134,0.00012844181,0.00083573366,0.00003697666,0.000073678966,0.000012601965,0.0000061933374,0.00068792887],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9964032,0.00035390395,0.00083761336,0.0004393026,0.0017144784,0.000251458],"domain_scores_gemma":[0.9974095,0.0009074268,0.0003736406,0.00067638437,0.0005540413,0.00007899453],"candidate_categories":["scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.0037293513,0.00015692365,0.00021693888,0.00013144161,0.00041617776,0.0010920417,0.00068508304,0.000084851505,0.00015105965],"category_scores_gemma":[0.00049004506,0.000097781805,0.00006522963,0.0008687254,0.00019730121,0.0008947063,0.00021167588,0.0002486232,0.000055734297],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007983487,0.0003187938,0.90601397,0.000031388303,0.00003205771,0.0000099428435,0.0065279775,0.008035145,0.0009332365,0.0019573676,0.02247477,0.05358554],"study_design_scores_gemma":[0.00041220145,0.00003853438,0.9109122,0.00006662534,0.000010366498,0.0000034214522,0.0047701066,0.069857344,0.0010629058,0.0043630293,0.008355946,0.00014727928],"about_ca_topic_score_codex":0.000066481334,"about_ca_topic_score_gemma":0.00022662322,"teacher_disagreement_score":0.061822202,"about_ca_system_score_codex":0.00016970598,"about_ca_system_score_gemma":0.0001908761,"threshold_uncertainty_score":0.9999449},"labels":[],"label_agreement":null},{"id":"W3175630060","doi":"","title":"Inter-rater reliability of clinicians' ratings of preschool children using the FOCUS©: Focus on the Outcomes of Communication Under Six.","year":2013,"lang":"en","type":"article","venue":"Canadian journal of speech-language pathology and audiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"Holland Bloorview Kids Rehabilitation Hospital; Laurentian University","funders":"","keywords":"Inter-rater reliability; Focus (optics); Psychology; Reliability (semiconductor); Developmental psychology; Clinical psychology; Rating scale","score_opus":0.06294481726840102,"score_gpt":0.3340803302059863,"score_spread":0.2711355129375853,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3175630060","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9868803,0.00079252536,0.0005551832,0.0110302465,0.00015604739,0.00032257813,0.000022245005,0.0000012712899,0.00023962841],"genre_scores_gemma":[0.9984805,0.000039469367,0.0008416377,0.0005566483,0.000037056594,0.0000030707704,7.8180926e-7,0.000005255357,0.000035561163],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99628866,0.0016639901,0.0013552664,0.00019334523,0.00028232523,0.00021644385],"domain_scores_gemma":[0.99478304,0.0023267583,0.0013603255,0.0008071215,0.0006182835,0.00010444608],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0071159387,0.00013700295,0.00057988084,0.00019516038,0.0001500757,0.000023576598,0.00096678233,0.00016323428,0.00048402385],"category_scores_gemma":[0.0047486853,0.000065811866,0.00018050212,0.00018261287,0.0013581075,0.00012412491,0.00008379784,0.00045925807,0.0000053587414],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004510379,0.000054117154,0.9761297,0.0000058299224,0.00009386681,0.000009329948,0.0031136873,0.00026649662,0.005668142,0.0008063982,0.0006824395,0.013124901],"study_design_scores_gemma":[0.00047165807,0.0005901351,0.9401953,0.00007205586,0.00006595139,0.0001251082,0.0059862444,0.000060804436,0.0077814916,0.04451973,0.000034722023,0.00009678483],"about_ca_topic_score_codex":0.003626425,"about_ca_topic_score_gemma":0.004258906,"teacher_disagreement_score":0.043713335,"about_ca_system_score_codex":0.000034770386,"about_ca_system_score_gemma":0.00022013583,"threshold_uncertainty_score":0.5684963},"labels":[],"label_agreement":null},{"id":"W3190138084","doi":"10.1093/jssam/smab022","title":"A Model-Assisted Approach for Finding Coding Errors in Manual Coding of Open-Ended Questions","year":2021,"lang":"en","type":"article","venue":"Journal of Survey Statistics and Methodology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Actua; University of Waterloo; Roche (Canada)","funders":"","keywords":"Coding (social sciences); Computer science; Statistics; Natural language processing; Artificial intelligence; Mathematics","score_opus":0.7870611956524755,"score_gpt":0.5441244124235828,"score_spread":0.24293678322889267,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3190138084","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.071355864,0.00014066743,0.9273774,0.0002445166,0.0002861363,0.00016495815,0.0002880419,0.0000010531656,0.0001413317],"genre_scores_gemma":[0.33733797,0.0001228901,0.6623521,0.00005782419,0.000014757362,0.00000358836,0.0000103411485,0.000004355226,0.000096195436],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9933288,0.0041255536,0.0014844252,0.0002742765,0.00057789066,0.00020907079],"domain_scores_gemma":[0.98161906,0.015807202,0.0009172976,0.00018416054,0.0013780873,0.00009420297],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05333626,0.00010813313,0.0007419871,0.00028749663,0.00010964914,0.00011616508,0.00042355814,0.000086996115,0.000020472131],"category_scores_gemma":[0.043901175,0.00008432916,0.00006985809,0.00039386604,0.000109462846,0.00016619994,0.00017099184,0.0002289664,1.3080471e-7],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0039863065,0.0027080409,0.36854446,0.0007042751,0.00070526,0.00016460693,0.007477638,0.101188526,0.111069724,0.21183698,0.026978757,0.16463543],"study_design_scores_gemma":[0.0039705643,0.0006481734,0.5165039,0.00020619223,0.00013896497,0.0001586504,0.0046952646,0.19040418,0.0022212537,0.28018364,0.00050901127,0.0003602097],"about_ca_topic_score_codex":0.00008526182,"about_ca_topic_score_gemma":0.0004676698,"teacher_disagreement_score":0.2659821,"about_ca_system_score_codex":0.00005217147,"about_ca_system_score_gemma":0.0002605016,"threshold_uncertainty_score":0.97478956},"labels":[],"label_agreement":null},{"id":"W3210268695","doi":"10.1136/oem-2021-epi.361","title":"RF-4 Inter-rater reliability of occupational exposure assessment in a case-control study","year":2021,"lang":"en","type":"article","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Centre Hospitalier de l’Université de Montréal","funders":"","keywords":"Concordance; Checklist; Cohen's kappa; Coding (social sciences); Statistic; Inter-rater reliability; Population; Reliability (semiconductor); Statistics; Applied psychology; Medicine; Computer science; Psychology; Environmental health; Mathematics","score_opus":0.14601194912680432,"score_gpt":0.4238998104811826,"score_spread":0.27788786135437826,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3210268695","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9858213,0.000022454677,0.009622331,0.0010631118,0.00032993362,0.000629264,0.000030199482,0.000011920527,0.0024694935],"genre_scores_gemma":[0.99737674,9.4684333e-7,0.0015380741,0.00022056312,0.000029432807,0.000054339067,0.0000017347696,0.0000040694886,0.0007740726],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9946785,0.0010162125,0.0014828962,0.00068411394,0.0019260856,0.00021220182],"domain_scores_gemma":[0.9960924,0.00148518,0.00020398581,0.0010161956,0.0011284439,0.00007380145],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.011341445,0.00014921027,0.0004259869,0.00015171115,0.0000654305,0.00012359422,0.00037535,0.000062253224,0.0058350847],"category_scores_gemma":[0.0027917775,0.00009870419,0.00014624688,0.00067782897,0.00006728837,0.00026622682,0.0001607151,0.00021023145,0.000042409665],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000084703686,0.0019285335,0.9907882,0.000007302253,0.00001691117,0.00020858854,0.00063255045,0.00080971344,0.00059553806,0.00012092701,0.0005658609,0.004241162],"study_design_scores_gemma":[0.0016522774,0.00037843286,0.9837675,0.0000139221665,0.000016009752,0.000021410782,0.00903772,0.0010344147,0.00049057126,0.003056856,0.00040911412,0.0001217655],"about_ca_topic_score_codex":0.00018890119,"about_ca_topic_score_gemma":0.0024891002,"teacher_disagreement_score":0.011555476,"about_ca_system_score_codex":0.00012300804,"about_ca_system_score_gemma":0.00030132476,"threshold_uncertainty_score":0.99507374},"labels":[],"label_agreement":null},{"id":"W3217637789","doi":"10.1016/j.cpet.2021.09.005","title":"Evidence-Based Artificial Intelligence in Medical Imaging","year":2021,"lang":"en","type":"review","venue":"PET Clinics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Joseph’s Healthcare Hamilton; McMaster University; University of Toronto","funders":"","keywords":"Artificial intelligence; Medicine; Software; Medical imaging; Machine learning; Data science; Sample (material); Computer science","score_opus":0.6320804372227717,"score_gpt":0.5604265466069183,"score_spread":0.07165389061585348,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3217637789","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000006477571,0.98669505,0.008059402,0.0021014225,0.0022864724,0.00050305505,0.000012830078,0.000026118174,0.00030919962],"genre_scores_gemma":[0.000044955956,0.99758524,0.0010176888,0.00070774485,0.00043639535,0.000066558496,0.00001668573,0.00002119035,0.0001035577],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98489577,0.0027408393,0.00475129,0.0014260298,0.005694582,0.0004914933],"domain_scores_gemma":[0.9724682,0.024063518,0.0009858557,0.0016578678,0.0005437214,0.0002808329],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.04213572,0.0004175656,0.0023473932,0.0004508118,0.00009770823,0.00047386097,0.0026648352,0.0002722179,0.0054114847],"category_scores_gemma":[0.10924888,0.00029073434,0.0010861803,0.0020170407,0.0002664092,0.00021838819,0.00045239477,0.001340086,0.001477529],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00000377129,0.00017888405,0.00014967223,0.0006949584,0.000009693242,0.00027727016,0.00001952124,0.000030214442,8.715088e-9,0.00048806518,0.0009895861,0.99715835],"study_design_scores_gemma":[0.000032161548,0.000024330435,0.0000059736185,0.040165596,0.00009354075,0.000016265536,0.00011974076,0.006196398,3.6962356e-7,0.009153693,0.9438515,0.00034045786],"about_ca_topic_score_codex":0.000016844784,"about_ca_topic_score_gemma":0.00013108525,"teacher_disagreement_score":0.9968179,"about_ca_system_score_codex":0.0002199467,"about_ca_system_score_gemma":0.003233128,"threshold_uncertainty_score":0.99995446},"labels":[],"label_agreement":null},{"id":"W32320309","doi":"10.1007/978-0-8176-4542-7_9","title":"Agreement Between Two Ratings with Different Ordinal Scales","year":2007,"lang":"en","type":"book-chapter","venue":"Birkhäuser Boston eBooks","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cancer Care Ontario","funders":"","keywords":"Ordinal Scale; Agreement; Kappa; Cohen's kappa; Ordinal data; Mathematics; Statistics; Rating scale; Scale (ratio); Level of measurement; Psychology; Linguistics; Geography","score_opus":0.22055782727867868,"score_gpt":0.367782393741022,"score_spread":0.14722456646234333,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W32320309","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.036900423,0.00025115785,0.004454104,0.000547535,0.0006675757,0.0018411351,0.00011499955,0.00017261735,0.95505047],"genre_scores_gemma":[0.17154336,0.000019646108,0.0014587577,0.00073060853,0.0011554336,0.0000581973,0.000051225263,0.0001498263,0.8248329],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98745793,0.00017488917,0.0022792446,0.0019244597,0.0072350544,0.0009283923],"domain_scores_gemma":[0.99391735,0.0010625374,0.0013462196,0.0021939576,0.00092735846,0.0005525609],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0050194697,0.0011872565,0.001544712,0.0007482238,0.0005091693,0.0005678679,0.0018929497,0.00049492525,0.002169393],"category_scores_gemma":[0.00019653994,0.0007509139,0.00059263746,0.00008804058,0.00068443606,0.00014458095,0.0005620622,0.001049265,0.0012133901],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011256547,0.00036281763,0.06123018,0.0002605708,0.0014535283,0.0004354421,0.0017102396,0.000044127286,0.0006406854,0.17235295,0.03163166,0.72875214],"study_design_scores_gemma":[0.0025942905,0.0014728528,0.01125334,0.001094294,0.0004555879,0.00002473376,0.00022081874,0.000016516586,0.0021979958,0.12030882,0.85814935,0.0022114231],"about_ca_topic_score_codex":0.00007137557,"about_ca_topic_score_gemma":0.0010512808,"teacher_disagreement_score":0.82651764,"about_ca_system_score_codex":0.0004051752,"about_ca_system_score_gemma":0.00019541368,"threshold_uncertainty_score":0.9995643},"labels":[],"label_agreement":null},{"id":"W4207058677","doi":"10.1016/j.jclinepi.2022.01.017","title":"Minimal important changes in standard deviation units are highly variable and no universally applicable value can be determined","year":2022,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":20,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"Canadian Institutes of Health Research; Shionogi","keywords":"Interquartile range; Statistics; Standard deviation; Mathematics; Medicine","score_opus":0.36576805802673124,"score_gpt":0.46650879546282675,"score_spread":0.10074073743609552,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4207058677","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9650066,0.0003171707,0.0019671437,0.03071079,0.0012422871,0.0002716675,0.00006714102,0.000007082845,0.0004101049],"genre_scores_gemma":[0.984869,0.0004876907,0.0077216346,0.006184082,0.00037966127,0.000011611792,0.0000043666046,0.000010293738,0.00033162785],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98494506,0.007791981,0.0051597273,0.0004998126,0.0011997133,0.00040368288],"domain_scores_gemma":[0.95830494,0.03468653,0.0052886344,0.00041588856,0.00097624166,0.00032779202],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.11529899,0.00015816845,0.0018161618,0.00026183674,0.00017003024,0.000026336918,0.00069048064,0.00016703499,0.0003387404],"category_scores_gemma":[0.11167263,0.000113204995,0.00016656444,0.0006305911,0.00018494917,0.00013170886,0.0002846539,0.00087317836,0.000002890332],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014961186,0.00018242125,0.97014534,0.00001418399,0.00005161417,0.000053908123,0.00012745937,0.0041036387,0.00016289628,0.003380106,0.014854009,0.0054282793],"study_design_scores_gemma":[0.004162179,0.005672983,0.47048247,0.00008312697,0.0000846743,0.00008234451,0.0018426069,0.00873206,0.00003500386,0.08661543,0.42188668,0.00032042892],"about_ca_topic_score_codex":0.00010655545,"about_ca_topic_score_gemma":0.00056795625,"teacher_disagreement_score":0.4996629,"about_ca_system_score_codex":0.00016984093,"about_ca_system_score_gemma":0.00059710303,"threshold_uncertainty_score":0.9109859},"labels":[],"label_agreement":null},{"id":"W4213160529","doi":"10.1016/j.envint.2022.107136","title":"Assessing the quality of evidence in studies estimating prevalence of exposure to occupational risk factors: The QoE-SPEO approach applied in the systematic reviews from the WHO/ILO Joint Estimates of the Work-related Burden of Disease and Injury","year":2022,"lang":"en","type":"article","venue":"Environment International","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Toronto Metropolitan University; Impact","funders":"Agencia Española de Cooperación Internacional para el Desarrollo; Universiteit van Amsterdam; University of Sydney; National Institute for Occupational Safety and Health; Aarhus Universitet; Bundesministerium für Gesundheit; World Health Organization","keywords":"Systematic review; Environmental health; Exposure assessment; Spurious relationship; Quality (philosophy); Medicine; Computer science; MEDLINE","score_opus":0.43892465778140927,"score_gpt":0.4402757951463774,"score_spread":0.0013511373649681135,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4213160529","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9921976,0.002793909,0.0011089623,0.0016552453,0.00011819029,0.0019914876,0.00011153825,0.0000015906713,0.000021524573],"genre_scores_gemma":[0.9984676,0.00022367561,0.0008897464,0.00005619388,0.000017317145,0.00032407744,0.0000039044753,0.000003725966,0.000013741223],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9913999,0.003231982,0.001998296,0.00031979967,0.0029453915,0.0001046364],"domain_scores_gemma":[0.9866014,0.010558877,0.0019391627,0.0008143434,0.00006819478,0.000018018653],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.022210179,0.00015024,0.0004382297,0.000056434154,0.00021193919,0.00005457604,0.0016552049,0.000020564958,0.00011496355],"category_scores_gemma":[0.011671175,0.000052291147,0.00014114068,0.00035696654,0.00042668442,0.00015288818,0.00075171725,0.00026485618,0.0000010581965],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000089811634,0.0002581603,0.7393622,0.0006748352,0.000107321735,8.218957e-8,0.02755673,0.22976132,0.0006668136,0.00049027894,0.00013629731,0.0008961489],"study_design_scores_gemma":[0.00013882092,0.00003620122,0.9713644,0.0019202688,0.00007446506,1.3663502e-7,0.015180011,0.0065407525,0.000112806556,0.0045514572,0.000012881066,0.000067798515],"about_ca_topic_score_codex":0.00014388094,"about_ca_topic_score_gemma":0.0000052831,"teacher_disagreement_score":0.2320022,"about_ca_system_score_codex":0.00011129767,"about_ca_system_score_gemma":0.000038356895,"threshold_uncertainty_score":0.9966539},"labels":[],"label_agreement":null},{"id":"W4224306105","doi":"10.1016/j.ajodo.2021.12.007","title":"Assessing the performance of diagnostic test accuracy measures","year":2022,"lang":"en","type":"editorial","venue":"American Journal of Orthodontics and Dentofacial Orthopedics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital","funders":"","keywords":"Test (biology); Statistics; Computer science; Medicine; Mathematics; Geology","score_opus":0.06976136545771053,"score_gpt":0.3884011828602047,"score_spread":0.31863981740249414,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4224306105","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6207663,0.002362045,0.001672619,0.0006154376,0.373812,0.0003488511,0.0001392196,0.000009053649,0.00027451184],"genre_scores_gemma":[0.70918757,0.031269144,0.002007918,0.00024541753,0.25682673,0.000016305987,0.00001969503,0.00007836707,0.00034883388],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.98583204,0.0012055744,0.0028766342,0.00046762408,0.00920928,0.0004088617],"domain_scores_gemma":[0.91628116,0.0722749,0.0070002726,0.00084524613,0.0033138827,0.0002845103],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.021848882,0.0004110729,0.0013966851,0.00038956315,0.00057605724,0.0007672479,0.001774434,0.00016939899,0.00013544575],"category_scores_gemma":[0.21204595,0.00024631055,0.00052976527,0.0011062126,0.0010279051,0.0006005505,0.0005312237,0.0017625645,0.0000033393135],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009345772,0.00024790716,0.5051841,0.000053924006,0.00016277893,0.000049835184,0.000709268,0.00049984985,0.000026472564,0.000013053817,0.19349712,0.2994622],"study_design_scores_gemma":[0.00043104147,0.00185749,0.022822611,0.0002014809,0.00040674125,0.00003051205,0.0024270946,0.00011595041,0.00001458637,0.00009555662,0.9712836,0.0003133305],"about_ca_topic_score_codex":0.00007793329,"about_ca_topic_score_gemma":0.000035721827,"teacher_disagreement_score":0.7777865,"about_ca_system_score_codex":0.00011012696,"about_ca_system_score_gemma":0.0014145372,"threshold_uncertainty_score":0.9999989},"labels":[],"label_agreement":null},{"id":"W4224989672","doi":"10.3390/sym14020262","title":"An Empirical Comparative Assessment of Inter-Rater Agreement of Binary Outcomes and Multiple Raters","year":2022,"lang":"en","type":"article","venue":"Symmetry","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Princess Margaret Cancer Centre; University Health Network; York University; Public Health Ontario; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Cohen's kappa; Kappa; Inter-rater reliability; Context (archaeology); Statistic; Mathematics; Summary statistics; Population; Medicine","score_opus":0.2176509483273682,"score_gpt":0.4693801055011164,"score_spread":0.2517291571737482,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4224989672","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.996466,0.00005958419,0.00091701146,0.0007839099,0.00033936126,0.00032944386,0.00006166577,0.000007940706,0.0010351206],"genre_scores_gemma":[0.99765843,0.0000028020454,0.0018580261,0.00026458362,0.000010080122,0.00001954215,0.000008662695,0.000004361814,0.00017349572],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9959061,0.0009163117,0.0009047108,0.00043969505,0.0016508304,0.00018234072],"domain_scores_gemma":[0.99800396,0.000730164,0.00036426162,0.0006084795,0.00020157004,0.000091592745],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0042601204,0.00014891746,0.00051047874,0.00027445683,0.0001462476,0.000041380903,0.0005726083,0.00002941571,0.0008843808],"category_scores_gemma":[0.00017160308,0.00010209876,0.00012081174,0.00046406142,0.00019114262,0.00017210511,0.00043142485,0.0001728876,0.0000061711694],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00005934498,0.000742861,0.98573303,0.000014587605,0.000056552373,0.0000021908668,0.0014723716,0.00081704825,0.0073389537,0.00016899567,0.0020493136,0.0015447708],"study_design_scores_gemma":[0.00087997946,0.0012559907,0.96895194,0.000011797509,0.000020425992,0.0000011338228,0.013760214,0.008793017,0.0029259229,0.0013336529,0.0019140075,0.00015191229],"about_ca_topic_score_codex":0.000059324495,"about_ca_topic_score_gemma":0.000015595544,"teacher_disagreement_score":0.016781064,"about_ca_system_score_codex":0.00008473315,"about_ca_system_score_gemma":0.00006441027,"threshold_uncertainty_score":0.968335},"labels":[],"label_agreement":null},{"id":"W4229629218","doi":"10.1111/acem.13515","title":"In Reply:","year":2018,"lang":"en","type":"letter","venue":"Academic Emergency Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"BC Children's Hospital; University of British Columbia","funders":"","keywords":"Kappa; Medicine; Cohen's kappa; Agreement; Documentation; Psychosocial; Value (mathematics); Statistic; Statistics; Psychiatry; Mathematics; Computer science","score_opus":0.3009726596357991,"score_gpt":0.46423642493452916,"score_spread":0.16326376529873005,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4229629218","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0023549986,0.0031475297,0.00008074988,0.94805765,0.015503932,0.00053935085,0.000010760476,0.0000424026,0.030262647],"genre_scores_gemma":[0.000493654,0.004616362,0.000039666633,0.87051857,0.034847505,0.00006736782,0.00004113096,0.00004133224,0.0893344],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98606044,0.000813714,0.004495115,0.001601619,0.0062044794,0.0008246243],"domain_scores_gemma":[0.9950794,0.0009944591,0.0012327546,0.0019525491,0.00060315576,0.00013769907],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity","insufficient_payload"],"category_scores_codex":[0.01777846,0.00051815034,0.0010730365,0.0009035496,0.000106755164,0.000002616815,0.0031996677,0.0014088538,0.06126476],"category_scores_gemma":[0.018275434,0.00033190593,0.00021118658,0.0018417527,0.00046273507,0.00029010704,0.00028810714,0.0052581583,0.0036088817],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000009203023,0.00001063102,0.009830767,0.00003804601,0.000017256438,0.000042504817,0.0005207046,0.0000010331444,0.00010196815,0.000030220042,0.98908454,0.00031313114],"study_design_scores_gemma":[0.00027898868,0.00015387766,0.0010426046,0.00027982154,0.000034015327,0.0000048903275,0.0001227328,0.000024762097,0.000025336602,0.075412855,0.92230046,0.00031966952],"about_ca_topic_score_codex":0.000276555,"about_ca_topic_score_gemma":0.000027043894,"teacher_disagreement_score":0.07753905,"about_ca_system_score_codex":0.00017860555,"about_ca_system_score_gemma":0.00011453283,"threshold_uncertainty_score":0.9999133},"labels":[],"label_agreement":null},{"id":"W4229874894","doi":"10.1007/978-94-007-0753-5_2025","title":"Ordinal Alpha","year":2014,"lang":"en","type":"book-chapter","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Learning Partnership; University of British Columbia","funders":"","keywords":"Alpha (finance); Psychology; Clinical psychology","score_opus":0.23728906902065514,"score_gpt":0.3670903399179083,"score_spread":0.12980127089725316,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4229874894","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000019533443,0.00011611255,0.009687594,0.00164535,0.0008352666,0.00021703259,0.0000070414176,0.000046858415,0.9874252],"genre_scores_gemma":[0.0037862856,0.00002980053,0.0007050441,0.0010778843,0.00040785375,0.000003890923,0.0000045704937,0.000017275714,0.9939674],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.99450374,0.00005540069,0.00090554485,0.00076430576,0.0035477057,0.00022328198],"domain_scores_gemma":[0.9971364,0.000657305,0.00032036903,0.0012631475,0.00047714316,0.00014564532],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0051517664,0.00028975477,0.0005049354,0.00019667858,0.00011268006,0.0002454566,0.001157126,0.00025871937,0.061577424],"category_scores_gemma":[0.00068934745,0.00017821317,0.00034764412,0.000042204076,0.00013484564,0.00006314475,0.00020495636,0.00028997264,0.02328504],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000008712623,0.000008666728,0.000079531696,0.000004897061,0.000021329033,0.000003417175,0.000011683511,0.0000068719546,0.000009223806,0.49123645,0.40929133,0.09931788],"study_design_scores_gemma":[0.00007627119,0.000060605777,0.00005396707,0.000024606494,0.000012718122,0.0000016346816,0.000006060984,0.000029388812,0.00001360311,0.323258,0.67629653,0.00016661869],"about_ca_topic_score_codex":0.000008635322,"about_ca_topic_score_gemma":0.00003691217,"teacher_disagreement_score":0.26700518,"about_ca_system_score_codex":0.0000587264,"about_ca_system_score_gemma":0.000065375185,"threshold_uncertainty_score":0.97747546},"labels":[],"label_agreement":null},{"id":"W4230740620","doi":"10.1503/cmaj.050708","title":"Correction","year":2005,"lang":"en","type":"article","venue":"Canadian Medical Association Journal","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Computer science; Information retrieval; World Wide Web; Data science","score_opus":0.04622657456481362,"score_gpt":0.31910396413610276,"score_spread":0.27287738957128915,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4230740620","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.23748486,0.00031002582,0.0038004809,0.5392657,0.020550197,0.00022358412,0.000011043281,0.000046642264,0.1983075],"genre_scores_gemma":[0.9553874,0.000051777744,0.0002457437,0.023225952,0.0047069546,0.0000029092037,0.0000010696773,0.000005984246,0.016372217],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99299705,0.00035425706,0.0006519909,0.00016283074,0.005490431,0.00034342398],"domain_scores_gemma":[0.99681824,0.00063076743,0.00024830172,0.00013820025,0.0008694383,0.0012950198],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.015734864,0.00006982266,0.00013744514,0.00021411564,0.00037508793,0.0003414463,0.00047139073,0.0002083445,0.030907068],"category_scores_gemma":[0.024475314,0.000052081938,0.000107232154,0.00038179796,0.000022760774,0.00027973182,0.000012736289,0.00067218125,0.0037046957],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000010887096,0.000011491698,0.03393397,1.1082e-7,0.000007896043,0.000008366399,0.00008709437,0.00006398849,0.0000029673158,0.000075638,0.67007935,0.29572803],"study_design_scores_gemma":[0.00020572588,0.00001854971,0.034951057,0.000010196494,0.00000331088,0.00006088921,0.00028500156,0.0022453337,0.000016460257,0.001251166,0.9608882,0.00006409958],"about_ca_topic_score_codex":0.00059110776,"about_ca_topic_score_gemma":0.05357155,"teacher_disagreement_score":0.71790254,"about_ca_system_score_codex":0.001376668,"about_ca_system_score_gemma":0.0016395355,"threshold_uncertainty_score":0.997071},"labels":[],"label_agreement":null},{"id":"W4233521617","doi":"10.1111/j.1541-0420.2008.01004_2.x","title":"Discussions","year":2008,"lang":"en","type":"article","venue":"Biometrics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Biostatistics; Epidemiology; Library science; Public health; Citation; Medicine; Gerontology; Sociology; Computer science; Pathology","score_opus":0.5836070243354328,"score_gpt":0.43811985529992914,"score_spread":0.14548716903550363,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4233521617","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92235595,0.0005656756,0.024494788,0.00468814,0.0018047999,0.00027778404,0.00002213844,0.00009202185,0.045698714],"genre_scores_gemma":[0.9911416,0.00007718634,0.0021305697,0.00020483488,0.00007800186,0.0000043642385,0.0000012649832,0.0000034064192,0.0063587544],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.996762,0.00009652505,0.0003958584,0.00029143374,0.0022699179,0.00018431462],"domain_scores_gemma":[0.9983129,0.00067108334,0.00009032523,0.0005622668,0.00023665873,0.00012678117],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0024720258,0.000074203555,0.00013462569,0.0010916196,0.00027100657,0.000063950916,0.000600274,0.000046733676,0.0008673998],"category_scores_gemma":[0.007998812,0.00004045896,0.00010413724,0.008111205,0.000108802546,0.00013603909,0.00011331339,0.00006788965,0.0019308558],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000148325125,0.0006293344,0.32344332,0.00000397742,0.000017130065,0.00003221468,0.0005076907,0.00005109456,0.0035283247,0.0031923922,0.43676358,0.23181611],"study_design_scores_gemma":[0.00019877292,0.00007419814,0.30577105,0.000003071902,0.0000034517066,0.00000806462,0.00020247351,0.0001408326,0.0007253501,0.0076720016,0.6850672,0.00013356359],"about_ca_topic_score_codex":0.000008739492,"about_ca_topic_score_gemma":0.0000014492093,"teacher_disagreement_score":0.2483036,"about_ca_system_score_codex":0.000038198024,"about_ca_system_score_gemma":0.000034805595,"threshold_uncertainty_score":0.99884623},"labels":[],"label_agreement":null},{"id":"W4240930308","doi":"10.1093/med/9780199685219.003.0008","title":"Reliability","year":2014,"lang":"en","type":"book","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Reliability (semiconductor); Intraclass correlation; Statistics; Standard error; Variance (accounting); Reliability engineering; Mathematics; Engineering; Psychometrics; Power (physics)","score_opus":0.1625810120239491,"score_gpt":0.3687530867046284,"score_spread":0.20617207468067933,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4240930308","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00004620123,0.000072216775,0.0047788965,0.0010666145,0.0011612322,0.00039524943,0.0000107624155,0.00008368407,0.99238515],"genre_scores_gemma":[0.0008404442,0.0000101564,0.0009230532,0.0011291002,0.00047214743,0.000012320214,0.000011402156,0.000017262195,0.9965841],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9920497,0.0003411035,0.0013650722,0.0011676474,0.0047794594,0.00029699947],"domain_scores_gemma":[0.99346495,0.002260421,0.0004336719,0.0027791918,0.0008855784,0.00017616843],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.015659902,0.00032247658,0.00069388084,0.0001810338,0.00012329382,0.00027222236,0.0018230438,0.00040312065,0.02004875],"category_scores_gemma":[0.005638648,0.00019331736,0.00046019463,0.00014511129,0.00021416681,0.00011766334,0.0002925718,0.0004527492,0.020525588],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000061725764,0.000028175946,0.00011341543,0.000013351436,0.000008131462,8.149338e-7,0.000016354783,0.000035764144,0.000004234605,0.016440934,0.96314883,0.020183818],"study_design_scores_gemma":[0.000061984814,0.000040174953,0.00025146836,0.000018449768,0.000009290953,4.21004e-7,0.0000047463986,0.00005984529,0.0000076613505,0.39323762,0.6061531,0.00015524308],"about_ca_topic_score_codex":0.000005984499,"about_ca_topic_score_gemma":0.000032539596,"teacher_disagreement_score":0.3767967,"about_ca_system_score_codex":0.00025789,"about_ca_system_score_gemma":0.00036340186,"threshold_uncertainty_score":0.98084706},"labels":[],"label_agreement":null},{"id":"W4243378373","doi":"10.31234/osf.io/v6crg","title":"Validation as Evaluating Desired and Undesired Effects: Insights from Cross-Classified Mixed Effects Model","year":2021,"lang":"en","type":"preprint","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Reliability (semiconductor); Variance (accounting); Computer science; Reliability engineering; Variance components; Validity; Data mining; Statistics; Psychometrics; Mathematics; Engineering","score_opus":0.28922857418109643,"score_gpt":0.4363073635794985,"score_spread":0.14707878939840208,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4243378373","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8908841,0.00094009,0.101073995,0.00021647773,0.0019160676,0.0018462788,0.000011351745,0.00013461868,0.002977048],"genre_scores_gemma":[0.9690508,0.00006946437,0.028749593,0.00036269418,0.00015246379,0.00024399116,0.00015028434,0.000036447815,0.0011842239],"study_design_codex":"bench_or_experimental","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.987538,0.0027398942,0.0017439704,0.0025515407,0.004969537,0.0004570559],"domain_scores_gemma":[0.98724335,0.008111195,0.0008720161,0.0020992572,0.0013599917,0.00031419046],"candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.006405668,0.00069381815,0.0012021869,0.00030898457,0.00048604826,0.003703459,0.0011987798,0.00074780727,0.00022979213],"category_scores_gemma":[0.01779376,0.0005188284,0.00038707897,0.0004677318,0.00023687725,0.0005744494,0.0017577901,0.0007208706,0.00010883949],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008093573,0.0015454647,0.019248532,0.0019235,0.0012490111,0.00012839751,0.0073522865,0.2617334,0.5199486,0.006191446,0.0022024207,0.17766756],"study_design_scores_gemma":[0.0016316499,0.00016796016,0.032797746,0.000685298,0.0002038729,0.0000011024744,0.00033156574,0.42991742,0.13884425,0.39471737,0.000018546729,0.0006832229],"about_ca_topic_score_codex":0.0003121279,"about_ca_topic_score_gemma":0.000067807625,"teacher_disagreement_score":0.3885259,"about_ca_system_score_codex":0.00032161272,"about_ca_system_score_gemma":0.00057571696,"threshold_uncertainty_score":0.99972636},"labels":[],"label_agreement":null},{"id":"W4246401766","doi":"10.1002/9781118445112.stat05095","title":"Bias, Overview","year":2014,"lang":"en","type":"other","venue":"Wiley StatsRef: Statistics Reference Online","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Chronic Disease Prevention Alliance of Canada","funders":"","keywords":"Interpretation (philosophy); Computer science; Data science; Data collection; Statistics; Information retrieval; Data mining; Econometrics; Mathematics; Programming language","score_opus":0.40002221701344864,"score_gpt":0.44192226801467055,"score_spread":0.041900051001221905,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4246401766","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000700524,0.012868081,0.34379557,0.0012576855,0.0045898277,0.0022454842,0.08190979,0.0007213057,0.5525422],"genre_scores_gemma":[0.00017734217,0.012712552,0.12859975,0.001246826,0.0009024832,0.000047031073,0.0034723599,0.00049790274,0.85234374],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98977524,0.00092052954,0.0019397755,0.0016065156,0.0049626823,0.0007952511],"domain_scores_gemma":[0.9929027,0.0017236333,0.0016265232,0.0025400908,0.00076980604,0.00043727583],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0036187433,0.0008244253,0.0014690561,0.000662446,0.00015497748,0.00042013425,0.0022642831,0.0005168813,0.040911537],"category_scores_gemma":[0.0057321633,0.0005922994,0.00017993081,0.00073916826,0.00038203408,0.00009005169,0.00040530605,0.00080452935,0.0071119596],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000015406888,0.00027053943,0.00034726073,0.00013920464,0.00006466842,0.0000152457915,0.000031387175,0.000016699603,0.000007177812,0.018920513,0.8720167,0.10815518],"study_design_scores_gemma":[0.00042577888,0.00021421844,0.00044539038,0.0007132859,0.00007736386,0.0000025211154,0.00007018656,0.00054702035,0.0000027931821,0.057296004,0.9395368,0.0006686482],"about_ca_topic_score_codex":0.00047553083,"about_ca_topic_score_gemma":0.003579229,"teacher_disagreement_score":0.29980153,"about_ca_system_score_codex":0.00014438933,"about_ca_system_score_gemma":0.00039829392,"threshold_uncertainty_score":0.99965286},"labels":[],"label_agreement":null},{"id":"W4249975290","doi":"10.1503/cmaj.1041744","title":"Kappa statistic","year":2005,"lang":"en","type":"letter","venue":"Canadian Medical Association Journal","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"","keywords":"Kappa; Cohen's kappa; Statistic; Statistics; Computer science; Mathematics","score_opus":0.0616572979115928,"score_gpt":0.325775439806404,"score_spread":0.2641181418948112,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4249975290","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00014603406,0.00024022747,0.0006155509,0.97769654,0.004266673,0.00015217857,0.00016712942,0.000013419418,0.01670224],"genre_scores_gemma":[0.0004214698,0.00010143979,0.00035372088,0.9461531,0.029693414,0.000007651802,0.00003751352,0.000022960427,0.023208741],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9780245,0.001333615,0.0017326231,0.00048729402,0.017419921,0.0010020814],"domain_scores_gemma":[0.99229395,0.0026251094,0.0010684638,0.00043147148,0.001869967,0.0017110354],"candidate_categories":["metaresearch","scholarly_communication","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity","insufficient_payload"],"category_scores_codex":[0.020496123,0.00028207028,0.000582718,0.0006131491,0.0005154426,0.0010703596,0.0018316518,0.0019280278,0.08679047],"category_scores_gemma":[0.03713951,0.00021462953,0.0003208947,0.0004966985,0.00007845971,0.0002084269,0.00004130253,0.006798795,0.007144552],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[8.3929484e-7,0.000008006284,0.001921485,0.0000033960544,0.00005986286,0.0009439561,0.00006250508,0.000006714605,1.2097648e-7,0.000026142512,0.9224826,0.074484386],"study_design_scores_gemma":[0.00026927853,0.00002881087,0.0015914005,0.00006135713,0.000026994867,0.00013269205,0.00007787921,0.000109695364,3.0562884e-7,0.0052569606,0.99220455,0.0002400845],"about_ca_topic_score_codex":0.0013915554,"about_ca_topic_score_gemma":0.026445793,"teacher_disagreement_score":0.079645924,"about_ca_system_score_codex":0.0034158903,"about_ca_system_score_gemma":0.0076727346,"threshold_uncertainty_score":0.9999666},"labels":[],"label_agreement":null},{"id":"W4253148760","doi":"10.1002/0470011815.b2a00001","title":"Bias, Overview","year":2005,"lang":"en","type":"other","venue":"Encyclopedia of Biostatistics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Chronic Disease Prevention Alliance of Canada","funders":"","keywords":"Interpretation (philosophy); Computer science; Statistics; Data science; Information retrieval; Mathematics; Programming language","score_opus":0.15507713619805066,"score_gpt":0.3854803442423336,"score_spread":0.2304032080442829,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4253148760","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000011372688,0.006740042,0.003956071,0.00031017547,0.0015398238,0.0004007201,0.0012804631,0.000057502246,0.9857038],"genre_scores_gemma":[0.00005625923,0.018388133,0.03006124,0.00021738649,0.0007552387,0.0000095521855,0.000051080417,0.00015336926,0.9503077],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9950071,0.0002598762,0.0011993519,0.00057642715,0.0026724613,0.0002847931],"domain_scores_gemma":[0.996376,0.0010610129,0.0010619805,0.0011439016,0.00022384814,0.00013329675],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.002109396,0.00032300845,0.0007221274,0.00038972253,0.00003655782,0.000056238096,0.0010249272,0.00026679575,0.04293883],"category_scores_gemma":[0.004515686,0.00023706042,0.00017269695,0.00048610498,0.00020137113,0.000042182055,0.00015825694,0.00020904609,0.002174127],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000032359605,0.000085678184,0.00073723646,0.00006584207,0.000026295444,0.0000042723627,0.000057017663,0.000003810693,0.0000016687713,0.0031332516,0.840756,0.1551257],"study_design_scores_gemma":[0.00016853976,0.000061909,0.00096414046,0.00019085231,0.00004567362,8.23763e-7,0.000053786054,0.000023829934,0.000008462074,0.0045607374,0.99366343,0.00025781879],"about_ca_topic_score_codex":0.00012738147,"about_ca_topic_score_gemma":0.00032356632,"teacher_disagreement_score":0.15486789,"about_ca_system_score_codex":0.000038982864,"about_ca_system_score_gemma":0.0001636935,"threshold_uncertainty_score":0.9986028},"labels":[],"label_agreement":null},{"id":"W4253597500","doi":"10.1007/0-387-30677-3_5","title":"Measurement Fundamentals","year":2006,"lang":"en","type":"book-chapter","venue":"Health Informatics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Computer science","score_opus":0.3308280014206468,"score_gpt":0.39249727942235774,"score_spread":0.06166927800171096,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4253597500","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000021686034,0.001630043,0.006533122,0.002897,0.001345882,0.0013707271,0.000117109885,0.0000925361,0.9859919],"genre_scores_gemma":[0.0053365454,0.0011176429,0.007578964,0.018862668,0.00082389184,0.00006850376,0.00013151027,0.000106945234,0.9659733],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98357606,0.00009296683,0.0053829355,0.0003670482,0.009938858,0.00064213073],"domain_scores_gemma":[0.9939617,0.00019202098,0.0029128075,0.0014998752,0.0010669918,0.0003666316],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.018192519,0.00051207386,0.0009948673,0.00044122335,0.00041869312,0.00041606496,0.0010678492,0.00031151983,0.0018119102],"category_scores_gemma":[0.00051529874,0.00038473762,0.00035285484,0.00011567935,0.00014999039,0.00031909443,0.00024062133,0.0005908885,0.0058686966],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00001232535,0.00005062287,0.00008739278,0.00038135308,0.000040178045,0.0000015015415,0.0005253453,0.00029264943,5.0366015e-7,0.022192948,0.88057816,0.09583702],"study_design_scores_gemma":[0.0002855317,0.00019265743,0.00012609817,0.0004589392,0.00001462974,0.0000041871654,0.00016439437,0.00027081117,0.000004463713,0.050113212,0.94802314,0.00034190642],"about_ca_topic_score_codex":0.000039144237,"about_ca_topic_score_gemma":0.000088193934,"teacher_disagreement_score":0.09549511,"about_ca_system_score_codex":0.0013683547,"about_ca_system_score_gemma":0.00093192025,"threshold_uncertainty_score":0.99986047},"labels":[],"label_agreement":null},{"id":"W4283735617","doi":"10.1002/hed.27130","title":"Augmenting inter‐rater concordance of radiologic extranodal extension in <scp>HPV</scp> ‐positive oropharyngeal carcinoma: A multicenter study","year":2022,"lang":"en","type":"article","venue":"Head & Neck","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Princess Margaret Cancer Centre; University of Toronto","funders":"","keywords":"Certainty; Concordance; Kappa; Inter-rater reliability; Medicine; Internal medicine; Radiology; Nuclear medicine; Psychology; Mathematics; Rating scale","score_opus":0.10176807180291711,"score_gpt":0.3508374837631717,"score_spread":0.2490694119602546,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4283735617","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9961598,0.0004821387,0.00019826919,0.00021211155,0.00077715586,0.0012171096,0.000026945667,0.000021889218,0.00090458407],"genre_scores_gemma":[0.99861145,0.000005991059,0.00030490445,0.00034777346,0.000056110755,0.00012199903,0.0000063648513,0.000013086252,0.0005323256],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99400765,0.0015635465,0.0013035665,0.0009212901,0.0017130575,0.0004908904],"domain_scores_gemma":[0.9973251,0.0012151039,0.00039557338,0.0006788515,0.00028801156,0.00009733843],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00534215,0.0002478942,0.0006340439,0.00025837548,0.0002719223,0.00007851029,0.0007946966,0.000046200224,0.00055095553],"category_scores_gemma":[0.001165892,0.00019012972,0.00017378591,0.00066236773,0.00013215658,0.00025144595,0.00059414894,0.00043376195,0.000057434023],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002514793,0.0019573632,0.9664021,0.000020524718,0.000025869416,0.0002067755,0.009269217,0.00095661177,0.0140121635,0.000021238828,0.0016800181,0.005196602],"study_design_scores_gemma":[0.0032722494,0.0018280282,0.96489686,0.000048866747,0.00002410186,0.0000109810235,0.021619549,0.0041926033,0.0019962692,0.00039950537,0.0015927054,0.000118271106],"about_ca_topic_score_codex":0.00030393718,"about_ca_topic_score_gemma":0.00020789795,"teacher_disagreement_score":0.012350332,"about_ca_system_score_codex":0.00022784054,"about_ca_system_score_gemma":0.00007118574,"threshold_uncertainty_score":0.7753265},"labels":[],"label_agreement":null},{"id":"W4294012667","doi":"10.1097/jte.0000000000000248","title":"Vertical Versus Horizontal Assessment Methods for Scoring Physiotherapy Entrance Interviews","year":2022,"lang":"en","type":"article","venue":"Journal of Physical Therapy Education","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Intraclass correlation; Inter-rater reliability; Reliability (semiconductor); Descriptive statistics; Correlation; Statistics; Psychology; Cohort; Mathematics; Psychometrics; Rating scale","score_opus":0.23418066243392052,"score_gpt":0.5653326466877174,"score_spread":0.3311519842537969,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4294012667","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8790997,0.00095916784,0.108949035,0.0037841205,0.006264397,0.00055722037,0.0000035735895,0.0000082865345,0.00037448414],"genre_scores_gemma":[0.9781375,0.0001039997,0.019771075,0.00036855182,0.0014063604,0.000116689094,0.0000017888638,0.000013735249,0.00008033614],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99638873,0.0009987791,0.0007505515,0.00026384272,0.0014014918,0.00019658712],"domain_scores_gemma":[0.99701726,0.001651965,0.0004244936,0.00032662123,0.0004714236,0.000108214495],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0057693813,0.00014027115,0.00038872965,0.00013081302,0.00025874565,0.0001335548,0.00074733526,0.000019816387,0.00026743137],"category_scores_gemma":[0.00038301427,0.0000968222,0.000516392,0.0003902395,0.00004577062,0.00033464626,0.00004553091,0.00030716034,0.0000038619064],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009415298,0.00213149,0.00018028107,0.0000049575006,0.00007366938,1.4387865e-7,0.0013385927,0.0004106678,0.14516984,0.0033102422,0.0020621656,0.84437644],"study_design_scores_gemma":[0.005441609,0.011298266,0.007903546,0.000060193946,0.000058083144,0.000007781629,0.0062006856,0.0070831226,0.042089514,0.14572017,0.7737235,0.00041357044],"about_ca_topic_score_codex":0.0000048624142,"about_ca_topic_score_gemma":5.0368766e-7,"teacher_disagreement_score":0.84396285,"about_ca_system_score_codex":0.00040382167,"about_ca_system_score_gemma":0.0004388603,"threshold_uncertainty_score":0.39482945},"labels":[],"label_agreement":null},{"id":"W4308945921","doi":"10.2147/clep.s368303","title":"Implementing Multiple Imputation for Missing Data in Longitudinal Studies When Models are Not Feasible: An Example Using the Random Hot Deck Approach","year":2022,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Jewish General Hospital","funders":"","keywords":"Imputation (statistics); Missing data; Longitudinal data; Statistics; Data mining; Medicine; Computer science; Mathematics","score_opus":0.9359089854971832,"score_gpt":0.6142285258336672,"score_spread":0.321680459663516,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4308945921","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.51046467,0.0011192268,0.48258168,0.0038399035,0.0006489968,0.0012132821,0.00009444371,0.000022281047,0.000015499234],"genre_scores_gemma":[0.90937823,0.00008326398,0.08752143,0.0023685715,0.0003106128,0.00018188305,0.0001236368,0.000016123693,0.00001628224],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9772592,0.013970486,0.0047248667,0.0020338432,0.00097223296,0.0010393258],"domain_scores_gemma":[0.8540591,0.13946709,0.0027550552,0.0029107104,0.00057817006,0.00022988209],"candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21688648,0.00023418384,0.0014194988,0.000142703,0.0013398405,0.000076283235,0.002126575,0.00011513037,0.00005805094],"category_scores_gemma":[0.12063811,0.00015011027,0.00026646294,0.0003987786,0.00038846093,0.0005840997,0.0020792861,0.0005880645,0.0000018342624],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011500448,0.00055519555,0.7805195,0.00003978351,0.000109172564,0.0000023455307,0.0012052217,0.17665078,0.000038582846,0.00074761186,0.005130471,0.03385128],"study_design_scores_gemma":[0.0023138472,0.00017912414,0.03763835,0.000014669454,0.000052441184,0.000004976898,0.007308568,0.78091705,0.0000018656712,0.16772242,0.0036822457,0.00016447563],"about_ca_topic_score_codex":0.0005360982,"about_ca_topic_score_gemma":0.0008213921,"teacher_disagreement_score":0.7428812,"about_ca_system_score_codex":0.00015008943,"about_ca_system_score_gemma":0.00016462374,"threshold_uncertainty_score":0.9999603},"labels":[],"label_agreement":null},{"id":"W4319057825","doi":"10.1093/annweh/wxad002","title":"Automated Coding of Job Descriptions From a General Population Study: Overview of Existing Tools, Their Application and Comparison","year":2023,"lang":"en","type":"article","venue":"Annals of Work Exposures and Health","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Saint John Regional Hospital; University of Toronto; University of New Brunswick","funders":"National Cancer Institute; National Institutes of Health; European Commission; University of Warwick","keywords":"Coding (social sciences); Computer science; Population; Statistics; Medicine; Mathematics; Environmental health","score_opus":0.7392532214113784,"score_gpt":0.5341445500158267,"score_spread":0.20510867139555167,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4319057825","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9950069,0.0026132385,0.000395315,0.0014023519,0.000052709624,0.00041818444,0.000039299244,0.00003095527,0.000040998446],"genre_scores_gemma":[0.9982296,0.00098059,0.00063566095,0.00008017892,0.00002489838,0.000010666369,0.000023289957,0.0000038074716,0.000011295493],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99767745,0.00027823908,0.0010851854,0.00024629466,0.00057276984,0.00014008008],"domain_scores_gemma":[0.9983966,0.00038158227,0.0006669066,0.000261794,0.00023115957,0.00006195669],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0046123024,0.00008518875,0.0004461333,0.0001270009,0.0001521914,0.00004162795,0.00013779388,0.00003576022,0.000007713687],"category_scores_gemma":[0.0002590264,0.000063713196,0.000047631664,0.00060827105,0.000043496413,0.00010884599,0.00007120289,0.00005507213,9.0364136e-7],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000027092645,0.00012183344,0.9161073,0.000074385476,0.000021790343,5.5990387e-8,0.0024140805,0.0004472137,0.00034496342,0.0010767027,0.0019619034,0.077402666],"study_design_scores_gemma":[0.00018360328,0.00023892085,0.9864707,0.00018408876,0.0000064964647,6.928525e-8,0.002428269,0.005012978,0.00010600397,0.005071063,0.00024739007,0.000050394083],"about_ca_topic_score_codex":0.0022593935,"about_ca_topic_score_gemma":0.00043598356,"teacher_disagreement_score":0.07735227,"about_ca_system_score_codex":0.000008376076,"about_ca_system_score_gemma":0.000026828282,"threshold_uncertainty_score":0.3415542},"labels":[],"label_agreement":null},{"id":"W4322767903","doi":"10.21203/rs.3.rs-2206790/v1","title":"WITHDRAWN: Inter-rater Reliability of Risk of Bias Tools for Non-Randomized Studies","year":2023,"lang":"en","type":"preprint","venue":"Research Square","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"","keywords":"Inter-rater reliability; Reliability (semiconductor); Reliability engineering; Psychology; Computer science; Engineering; Physics; Developmental psychology; Rating scale","score_opus":0.6070368058631517,"score_gpt":0.5470055814174135,"score_spread":0.06003122444573816,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4322767903","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9751287,0.0009949583,0.0073161772,0.0027933128,0.0015584825,0.010047292,0.0014828434,0.000054860993,0.0006233823],"genre_scores_gemma":[0.991769,0.0018820058,0.002895992,0.000010405787,0.00021218334,0.0015718268,0.00004327362,0.00003627105,0.0015790345],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97886527,0.0077009257,0.0035563426,0.0016473094,0.00747364,0.00075652584],"domain_scores_gemma":[0.8670857,0.11043437,0.0017251628,0.0039505325,0.016603177,0.00020108002],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.16428824,0.00041309473,0.0028622812,0.0010244104,0.00026281923,0.0003527721,0.0024738137,0.0003847067,0.00018839039],"category_scores_gemma":[0.37980068,0.0002458507,0.0015454617,0.001220536,0.0026694057,0.00025802673,0.0035118733,0.0012804209,0.00011003953],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.21054454,0.0054478175,0.3055396,0.042618778,0.0067964853,0.00003114618,0.08939952,0.03506808,0.0017494145,0.0035254322,0.20078641,0.09849276],"study_design_scores_gemma":[0.03774108,0.0013392762,0.024228895,0.0064604403,0.00028882877,3.612582e-7,0.035496395,0.0065697934,0.010329831,0.87210995,0.0046812454,0.0007539305],"about_ca_topic_score_codex":0.0006060267,"about_ca_topic_score_gemma":0.00027724227,"teacher_disagreement_score":0.8685845,"about_ca_system_score_codex":0.00025448584,"about_ca_system_score_gemma":0.0006315475,"threshold_uncertainty_score":0.9999994},"labels":[],"label_agreement":null},{"id":"W4362636314","doi":"10.1177/1536867x231161978","title":"Extended biasplot command to assess bias, precision, and agreement in method comparison studies","year":2023,"lang":"en","type":"article","venue":"The Stata Journal Promoting communications on statistics and Stata","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Statistics; Computer science; Limits of agreement; Accuracy and precision; Data mining; Mathematics; Nuclear medicine; Medicine","score_opus":0.6821565065636919,"score_gpt":0.5640156980598392,"score_spread":0.11814080850385267,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4362636314","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.51187795,0.005431054,0.41568282,0.059616495,0.00085175625,0.0032535666,0.001890371,0.00014123875,0.00125474],"genre_scores_gemma":[0.8281842,0.0056084697,0.16539657,0.00038341313,0.00003846412,0.000057826303,0.00003810112,0.000024407605,0.0002685657],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99291074,0.0030314156,0.0015748508,0.00044810044,0.0016294083,0.0004054655],"domain_scores_gemma":[0.9803253,0.016219396,0.00057685457,0.0019276943,0.0007368457,0.000213958],"candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.032106325,0.00023268905,0.00052968663,0.00047458755,0.0015873901,0.0008534907,0.0019261874,0.000036797694,0.000020037396],"category_scores_gemma":[0.011596926,0.00014233957,0.000038514085,0.001092348,0.0003012477,0.00019159852,0.001696773,0.0006847375,0.00004360394],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00022188728,0.0006836731,0.017668886,0.000091096896,0.00024997236,0.000027782842,0.03665916,0.0020092872,0.00024466336,0.04566475,0.07490741,0.8215714],"study_design_scores_gemma":[0.00250304,0.0018716543,0.19335875,0.0011916632,0.00016437222,0.00007568775,0.11731057,0.2688134,0.0001709505,0.35358772,0.06012063,0.0008315455],"about_ca_topic_score_codex":0.000059315676,"about_ca_topic_score_gemma":0.00039189836,"teacher_disagreement_score":0.82073987,"about_ca_system_score_codex":0.000110758694,"about_ca_system_score_gemma":0.00007938226,"threshold_uncertainty_score":0.9997124},"labels":[],"label_agreement":null},{"id":"W4377140177","doi":"10.3390/ani13101664","title":"Inter-Rater Reliability of Scoring Systems for Abomasal Lesions in Quebec Veal Calves","year":2023,"lang":"en","type":"article","venue":"Animals","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université de Montréal","funders":"Ministère de l'Agriculture, des Pêcheries et de l'Alimentation","keywords":"Inter-rater reliability; Lesion; Medicine; Kappa; Concordance; Internal medicine; Surgery; Statistics; Mathematics; Rating scale","score_opus":0.22359830233783815,"score_gpt":0.4084871304609386,"score_spread":0.18488882812310045,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4377140177","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99661744,0.0001242319,0.0006357853,0.0008471525,0.0005711346,0.00060882483,0.000036834914,0.000037738053,0.00052087737],"genre_scores_gemma":[0.99798566,0.000018830426,0.00018136558,0.000027681512,0.00007213776,0.00009469442,0.0000027670842,0.000007821074,0.0016090437],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.996994,0.00026743687,0.0010711642,0.0004902627,0.00086404535,0.00031307098],"domain_scores_gemma":[0.9971345,0.0016845585,0.00019087865,0.000579143,0.00034321338,0.00006771107],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008259094,0.0001230917,0.000384058,0.0002502553,0.00007984639,0.00012336655,0.00044229205,0.000077412245,0.0000633948],"category_scores_gemma":[0.0037325076,0.00008649365,0.00015301083,0.00073417195,0.00010075264,0.00021843452,0.0001494426,0.00009140803,0.0001427279],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005145479,0.00049863604,0.8598137,0.00047346725,0.00004273608,0.000018229792,0.0048805545,0.0071387975,0.0569091,0.008540067,0.042157546,0.019012658],"study_design_scores_gemma":[0.00095730636,0.00046796488,0.92308205,0.0003983471,0.000017147144,9.959589e-7,0.006380728,0.007446737,0.006629019,0.021322036,0.032971274,0.00032641756],"about_ca_topic_score_codex":0.0058468482,"about_ca_topic_score_gemma":0.0030856985,"teacher_disagreement_score":0.06326836,"about_ca_system_score_codex":0.00008332876,"about_ca_system_score_gemma":0.00008549116,"threshold_uncertainty_score":0.88387245},"labels":[],"label_agreement":null},{"id":"W4380758257","doi":"10.1016/j.jclinepi.2023.06.009","title":"A scoping review establishes need for consensus guidance on reporting health equity in observational studies","year":2023,"lang":"en","type":"review","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":13,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children; Canadian Agency for Drugs and Technologies in Health; St. Joseph’s Healthcare Hamilton; London Health Sciences Centre; Ottawa Hospital; Carleton University; Queen's University; International Development Research Centre; Impact; Canadian Association of Emergency Physicians; Public Health Ontario; Dalhousie University; Royal College of Physicians and Surgeons of Canada; University of Toronto; Centre for Global Health Research; Cochrane; Ottawa Public Health; University of Ottawa; Canadian Council on Animal Care; Bruyère; McMaster University","funders":"Canadian Institutes of Health Research; World Health Organization","keywords":"Observational study; Equity (law); MEDLINE; Medicine; Actuarial science; Business; Political science; Pathology","score_opus":0.9823343326318191,"score_gpt":0.7855852011404475,"score_spread":0.19674913149137163,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4380758257","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000022569046,0.957156,0.0004614724,0.03631184,0.0036898837,0.0022630575,0.000019621983,0.000011219748,0.000064309505],"genre_scores_gemma":[0.0000013219395,0.9798651,0.009043956,0.009764679,0.0010804262,0.0001044808,0.0000063321845,0.000020419937,0.00011330411],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.8314874,0.045331642,0.11713217,0.0017848423,0.0030190984,0.0012448048],"domain_scores_gemma":[0.08131704,0.77061415,0.14266999,0.0013813946,0.003517926,0.0004994889],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6498553,0.00043375193,0.022148954,0.00041235698,0.000166675,0.000037126185,0.001384899,0.00046182142,0.000024763285],"category_scores_gemma":[0.96221954,0.00024609207,0.0036373085,0.0010737112,0.0003754799,0.00011304158,0.00039170854,0.0017695636,0.00003332535],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00005799348,0.0001306095,0.0035521481,0.08411906,0.00022507191,0.000027073122,0.000011899263,0.00016322808,4.4207376e-9,0.00096053345,0.110773236,0.79997915],"study_design_scores_gemma":[0.0003344443,0.0007391296,0.0010634398,0.6029842,0.00014986702,0.00003303307,0.00004070553,0.000032081018,5.7618887e-9,0.0476238,0.34682253,0.00017677006],"about_ca_topic_score_codex":0.000010810178,"about_ca_topic_score_gemma":0.000093789786,"teacher_disagreement_score":0.79980236,"about_ca_system_score_codex":0.0004092932,"about_ca_system_score_gemma":0.0049481285,"threshold_uncertainty_score":0.9999991},"labels":[],"label_agreement":null},{"id":"W4385694112","doi":"10.1159/000531054","title":"Clinical Validation of Novel Digital Measures: Statistical Methods for Reliability Evaluation","year":2023,"lang":"en","type":"review","venue":"Digital Biomarkers","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Bayer (Canada)","funders":"","keywords":"Reliability (semiconductor); Categorical variable; Computer science; Reliability engineering; Measure (data warehouse); Data mining; Machine learning; Engineering","score_opus":0.6501770228660378,"score_gpt":0.6044108355425473,"score_spread":0.04576618732349058,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385694112","genre_codex":"methods","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000733605,0.4048096,0.57292366,0.000102537444,0.003019441,0.008316054,0.0069746226,0.00013955192,0.0036411523],"genre_scores_gemma":[0.000834067,0.9095527,0.07903821,0.000040144154,0.000645147,0.0016975319,0.00708194,0.00025056335,0.0008597254],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98501986,0.0018293554,0.006162271,0.0019321496,0.0045310324,0.00052533305],"domain_scores_gemma":[0.9589745,0.034633633,0.0020616413,0.0016545614,0.0023968753,0.0002788373],"candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.061675776,0.0005668849,0.0026200605,0.00047458243,0.00012339037,0.0013057103,0.0013798323,0.0005596306,0.0000828314],"category_scores_gemma":[0.18035023,0.00039001286,0.001954429,0.0015352364,0.0005897849,0.0010931061,0.00031964664,0.0002822809,0.00020263092],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006865538,0.00043987407,0.00014903022,0.0018152879,0.00036522205,2.7962034e-7,0.000015891152,0.000010941596,5.6337035e-7,0.00015116307,0.003194521,0.9937886],"study_design_scores_gemma":[0.0007828067,0.00031582682,0.000099408186,0.0020410698,0.0008267511,0.000002285778,0.0001436173,0.0014492335,0.0000041722205,0.0202075,0.97360635,0.00052098994],"about_ca_topic_score_codex":0.0000039271886,"about_ca_topic_score_gemma":9.766052e-7,"teacher_disagreement_score":0.9932676,"about_ca_system_score_codex":0.00032573947,"about_ca_system_score_gemma":0.0010900886,"threshold_uncertainty_score":0.99985516},"labels":[],"label_agreement":null},{"id":"W4386504557","doi":"10.1111/apa.16957","title":"High inter‐rater reliability between physicians and nurses utilising modified Downes' scores in preterm respiratory distress","year":2023,"lang":"en","type":"article","venue":"Acta Paediatrica","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Vancouver Coastal Health; Royal Columbian Hospital; St. Michael's Hospital","funders":"Bill and Melinda Gates Foundation","keywords":"Concordance; Medicine; Cohen's kappa; Respiratory distress; Inter-rater reliability; Kappa; Statistic; Intensive care; Pediatrics; Concordance correlation coefficient; Respiratory therapist; Emergency medicine; Intensive care medicine; Internal medicine; Surgery; Statistics; Rating scale","score_opus":0.09805814331907288,"score_gpt":0.34845895603568405,"score_spread":0.2504008127166112,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386504557","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9963195,0.00010972379,0.000068813584,0.0017095653,0.00067077397,0.000514858,0.000106664826,0.00010318527,0.00039691498],"genre_scores_gemma":[0.9991021,0.000055240824,0.00008510628,0.00017335772,0.0003447583,0.000044479308,0.000017838669,0.000015866302,0.00016122004],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99533325,0.0006392959,0.0010574264,0.00100658,0.0014279601,0.0005354908],"domain_scores_gemma":[0.9959419,0.0024046055,0.00028627683,0.00104186,0.00018737403,0.00013801016],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004005374,0.00026794797,0.0005236695,0.0004227632,0.00020034034,0.00033729104,0.00082818477,0.00014788013,0.00005169557],"category_scores_gemma":[0.0030775783,0.00019704357,0.0001344348,0.0015604428,0.0002771995,0.00054005237,0.00037947952,0.00032017494,0.000099809404],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007848095,0.00011312998,0.9200017,0.000044787965,0.000013244093,0.000011522519,0.0013001447,0.00021829062,0.00058169157,0.00004513296,0.008341266,0.069250606],"study_design_scores_gemma":[0.00063162553,0.0001439365,0.9794143,0.00006823426,0.00002926979,2.201662e-7,0.00059063593,0.00045241756,0.00072227645,0.012159199,0.005469857,0.00031805816],"about_ca_topic_score_codex":0.00016485747,"about_ca_topic_score_gemma":0.000070218724,"teacher_disagreement_score":0.06893255,"about_ca_system_score_codex":0.00010287248,"about_ca_system_score_gemma":0.00005306215,"threshold_uncertainty_score":0.8035203},"labels":[],"label_agreement":null},{"id":"W4389166029","doi":"10.1016/j.drugalcdep.2023.111043","title":"Evaluating the agreement between different substance use recall periods in multiple HIV cohorts","year":2023,"lang":"en","type":"article","venue":"Drug and Alcohol Dependence","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University; University of British Columbia; British Columbia Centre on Substance Use","funders":"National Institute on Drug Abuse; National Institute of Mental Health","keywords":"Cohen's kappa; Kappa; Recall; Medicine; Agreement; Concordance; Statistics; Recall bias; Psychology; Clinical psychology; Demography; Mathematics; Internal medicine; Pathology","score_opus":0.4071326534392604,"score_gpt":0.42793529633616845,"score_spread":0.02080264289690803,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389166029","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99569756,0.00024459703,0.00023926755,0.0025532725,0.00029449977,0.00070997415,0.000027737187,0.000047438563,0.00018563248],"genre_scores_gemma":[0.99659747,0.00018802461,0.00020658562,0.00016981948,0.00007983666,0.000094797986,0.000010428772,0.000009968948,0.0026430592],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9939117,0.00080865336,0.0009191565,0.00089811423,0.0029314053,0.000530959],"domain_scores_gemma":[0.99461895,0.0039450987,0.00022774741,0.00087162317,0.00019122836,0.00014537963],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011039331,0.00025070508,0.0003998309,0.00017529356,0.00046910028,0.00052901835,0.0008707076,0.0000673017,0.00013808889],"category_scores_gemma":[0.0025510862,0.00014622504,0.00010639945,0.0008715412,0.00018810535,0.0003943444,0.0002945234,0.00034358923,0.00025098052],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00002149848,0.0000414129,0.9644806,0.000008435078,0.000012503857,0.000011871968,0.0019507755,0.0009927533,0.00091164885,0.000060540482,0.0009257276,0.03058227],"study_design_scores_gemma":[0.0006110669,0.000046537494,0.97687066,0.00011000046,0.00001636073,0.0000015419631,0.0014862623,0.015163909,0.0006471568,0.0036754066,0.0011462698,0.00022481498],"about_ca_topic_score_codex":0.00021900452,"about_ca_topic_score_gemma":0.0013484757,"teacher_disagreement_score":0.030357454,"about_ca_system_score_codex":0.00008286444,"about_ca_system_score_gemma":0.000039646315,"threshold_uncertainty_score":0.5962884},"labels":[],"label_agreement":null},{"id":"W4389410166","doi":"10.1186/s13643-023-02389-w","title":"Inter-rater reliability of risk of bias tools for non-randomized studies","year":2023,"lang":"en","type":"article","venue":"Systematic Reviews","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Centre Hospitalier de l’Université de Montréal; Université de Montréal","funders":"","keywords":"Intraclass correlation; Checklist; Medicine; Reliability (semiconductor); Inter-rater reliability; Scale (ratio); Statistics; Rating scale; Mathematics; Clinical psychology; Psychometrics; Psychology; Cartography","score_opus":0.4923679694819516,"score_gpt":0.46932421469341784,"score_spread":0.023043754788533766,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389410166","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.84486914,0.015836142,0.04316949,0.0013876979,0.0044028633,0.08816609,0.00021235368,0.000096379285,0.001859862],"genre_scores_gemma":[0.9816961,0.0059341677,0.005202759,0.00006358467,0.00009149606,0.0054663876,0.000006399168,0.000018931003,0.0015201197],"study_design_codex":"systematic_review","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9827873,0.0075868964,0.007282352,0.0005126029,0.0015938592,0.00023694888],"domain_scores_gemma":[0.9294542,0.061447863,0.004836722,0.0022515072,0.0019431602,0.00006653159],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.18077348,0.00024421216,0.004876456,0.00021225984,0.00007443594,0.000069569985,0.0008213623,0.00006297815,0.000068797235],"category_scores_gemma":[0.41076532,0.00010536239,0.0014397618,0.0009159545,0.0003204946,0.0002357486,0.00016235093,0.00007785238,0.00037106196],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0048861587,0.0006622394,0.014323663,0.8152596,0.0017029912,0.0000018077583,0.04209247,0.00053733744,0.00095069373,0.0021651809,0.09376968,0.023648193],"study_design_scores_gemma":[0.10677017,0.0016109527,0.004395115,0.20585844,0.005358189,0.0000058479254,0.075367875,0.020138796,0.0079955105,0.50689405,0.063360624,0.002244443],"about_ca_topic_score_codex":0.000014165252,"about_ca_topic_score_gemma":0.000012795629,"teacher_disagreement_score":0.60940117,"about_ca_system_score_codex":0.00004317505,"about_ca_system_score_gemma":0.000036100177,"threshold_uncertainty_score":0.8435661},"labels":[],"label_agreement":null},{"id":"W4389615068","doi":"10.1148/radiol.230492","title":"Measuring Interrater Reliability","year":2023,"lang":"en","type":"editorial","venue":"Radiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre Hospitalier de l’Université de Montréal","funders":"","keywords":"Medicine; Inter-rater reliability; Neuroradiology; Radiology; Scholarship; Medical physics; Nuclear medicine; Psychiatry; Psychology; Rating scale; Neurology","score_opus":0.21623695076910274,"score_gpt":0.3877990363725101,"score_spread":0.17156208560340738,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389615068","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0020925505,0.00016868855,0.00043607596,0.0010642055,0.9923473,0.00033201813,0.000096185366,0.00017939125,0.0032835598],"genre_scores_gemma":[0.003797459,0.00017127897,0.00021812362,0.000088210996,0.98084545,0.00010662236,0.000112974296,0.000059849513,0.014600002],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9901486,0.0018243586,0.0017470551,0.0017671048,0.0037827354,0.0007301161],"domain_scores_gemma":[0.97955763,0.016008103,0.0005871495,0.0023919859,0.0012599106,0.00019520124],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.022543304,0.00043873483,0.0011803512,0.00039271207,0.00019126761,0.00022653751,0.0025921448,0.0014015689,0.001012177],"category_scores_gemma":[0.097608596,0.00030253516,0.0004576108,0.0005908238,0.00036038808,0.00014867815,0.0005751962,0.0014455646,0.0075609973],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000055529035,0.00003692952,0.00052172155,0.000028973469,0.000043992604,0.00001657558,0.00014626831,0.00014536722,0.000050633535,0.000026350124,0.9964296,0.0024980549],"study_design_scores_gemma":[0.00025110383,0.00015314914,0.0005039124,0.000052525593,0.000026487083,0.0000016129137,0.00007285012,0.0000852401,0.000029082446,0.019645048,0.9788671,0.00031185767],"about_ca_topic_score_codex":0.00008531289,"about_ca_topic_score_gemma":0.0001236169,"teacher_disagreement_score":0.07506529,"about_ca_system_score_codex":0.0003943268,"about_ca_system_score_gemma":0.0003669044,"threshold_uncertainty_score":0.99994266},"labels":[],"label_agreement":null},{"id":"W4390586053","doi":"10.1016/j.jpi.2023.100358","title":"Use of n-grams and K-means clustering to classify data from free text bone marrow reports","year":2024,"lang":"en","type":"article","venue":"Journal of Pathology Informatics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"","keywords":"Computer science; Cluster analysis; Artificial intelligence; Security token; Centroid; Bone marrow; Text mining; Natural language processing; Data mining; Pattern recognition (psychology); Pathology; Medicine","score_opus":0.2526589685442207,"score_gpt":0.37809989781726455,"score_spread":0.12544092927304384,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4390586053","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.88522094,0.00049024465,0.10890969,0.0028930064,0.0017842278,0.0001690168,0.00014161986,0.000011470643,0.0003798033],"genre_scores_gemma":[0.7890837,0.00026834224,0.20886663,0.0012399639,0.00025013127,0.0000019627785,0.000014443956,0.0000139384165,0.00026084133],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9955433,0.00013372951,0.0026609818,0.00016300897,0.0013206608,0.00017831163],"domain_scores_gemma":[0.9962146,0.0010335604,0.0009892207,0.0012466052,0.00037031097,0.00014570008],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007822753,0.00012747574,0.0004587161,0.00034016577,0.00005529346,0.0003288798,0.00079411647,0.00011505293,0.00010603327],"category_scores_gemma":[0.006537768,0.00008268482,0.00009004189,0.00031264775,0.00012861755,0.001405493,0.00080871047,0.00033478922,0.000028429276],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003229299,0.00038161827,0.07952043,0.00046007795,0.00042113752,0.005276441,0.030758515,0.013218207,0.014229385,0.000341044,0.5623993,0.2926709],"study_design_scores_gemma":[0.001436397,0.0020994407,0.064472094,0.0021549077,0.00052083354,0.010712201,0.016557807,0.23478377,0.001225465,0.05797716,0.60719705,0.00086285005],"about_ca_topic_score_codex":0.000011075072,"about_ca_topic_score_gemma":0.000052268042,"teacher_disagreement_score":0.29180807,"about_ca_system_score_codex":0.00003620294,"about_ca_system_score_gemma":0.0001017886,"threshold_uncertainty_score":0.78267914},"labels":[],"label_agreement":null},{"id":"W4391036726","doi":"10.4236/ojepi.2024.141005","title":"Cautionary Remarks When Testing Agreement between Two Raters for Continuous Scale Measurements: A Tutorial in Clinical Epidemiology with Implementation Using R","year":2024,"lang":"en","type":"article","venue":"Open Journal of Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Statistic; Scale (ratio); Sample size determination; Statistics; Sample (material); Linear regression; Degrees of freedom (physics and chemistry); Data mining; Computer science; Medicine; Econometrics; Mathematics","score_opus":0.7023443814366066,"score_gpt":0.5922363059400028,"score_spread":0.11010807549660384,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391036726","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6810825,0.0014924443,0.29189518,0.01645016,0.0056168954,0.0027096374,0.000061766696,0.000016385353,0.0006750639],"genre_scores_gemma":[0.6420902,0.000048821963,0.35300872,0.0018859778,0.0027118998,0.000075551965,0.000029976001,0.000033274955,0.000115596136],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9750859,0.01305601,0.00888342,0.0010911316,0.00096583273,0.0009177031],"domain_scores_gemma":[0.942783,0.05165753,0.0032383476,0.00056107406,0.0014293195,0.00033071957],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.22963852,0.0003135127,0.0023224703,0.00044745946,0.00021615106,0.0001494124,0.0014394049,0.0002078938,0.0002983991],"category_scores_gemma":[0.05831591,0.00020132046,0.0003976872,0.0005247041,0.00033545916,0.0008341631,0.0003043549,0.0006648405,0.000016530721],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00065359636,0.0001013662,0.91180694,0.000025090509,0.0002403405,0.000020108962,0.00020419226,0.0031431417,0.00031970892,0.0004580141,0.011513314,0.0715142],"study_design_scores_gemma":[0.014159615,0.008068411,0.4736662,0.00198333,0.0008067163,0.00033510086,0.0030341828,0.013386149,0.0001845435,0.4174974,0.06587006,0.001008305],"about_ca_topic_score_codex":0.0006490139,"about_ca_topic_score_gemma":0.0004239247,"teacher_disagreement_score":0.43814072,"about_ca_system_score_codex":0.0005322691,"about_ca_system_score_gemma":0.00087593705,"threshold_uncertainty_score":0.9496163},"labels":[],"label_agreement":null},{"id":"W4391432388","doi":"10.1097/ede.0000000000001726","title":"Application of a Web-based Tool for Quantitative Bias Analysis: The Example of Misclassification Due to Self-reported Body Mass Index","year":2024,"lang":"en","type":"article","venue":"Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto","funders":"National Cancer Institute; National Institutes of Health","keywords":"Body mass index; Index (typography); Statistics; Computer science; Mathematics; Medicine; World Wide Web; Internal medicine","score_opus":0.3506473340767288,"score_gpt":0.45331573356540744,"score_spread":0.10266839948867862,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391432388","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19466019,0.00019019024,0.7961149,0.007809286,0.00013860557,0.00090971275,0.000038872335,0.000030254012,0.00010797228],"genre_scores_gemma":[0.9795833,0.0000060739876,0.019519813,0.0003605593,0.00003114354,0.00040185964,0.000029314973,0.0000055674054,0.00006237411],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.995324,0.001485989,0.001839964,0.0006301221,0.0005181153,0.00020180116],"domain_scores_gemma":[0.9740092,0.023567915,0.00077380217,0.00093411596,0.00066735223,0.000047588783],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03042231,0.00012093226,0.0006411141,0.00040526866,0.000071084556,0.000018406077,0.0005096328,0.000106817126,0.00006466523],"category_scores_gemma":[0.021754581,0.000069941976,0.00033672998,0.0018781472,0.00013432436,0.00006247313,0.00003160204,0.00009667031,0.000030649808],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004519093,0.0004554446,0.6308511,0.000208479,0.0026929413,0.0000013601909,0.0011661007,0.08613123,0.05585988,0.18155468,0.020376101,0.020250797],"study_design_scores_gemma":[0.00021373092,0.00024078593,0.2600663,0.000019027175,0.00048361652,4.6395692e-7,0.0002471052,0.6327364,0.0015480678,0.06832453,0.03601257,0.00010743203],"about_ca_topic_score_codex":0.00034020114,"about_ca_topic_score_gemma":0.00012145082,"teacher_disagreement_score":0.7849231,"about_ca_system_score_codex":0.000049605864,"about_ca_system_score_gemma":0.0001581581,"threshold_uncertainty_score":0.9983843},"labels":[],"label_agreement":null},{"id":"W4392815531","doi":"10.1016/j.hrthm.2024.03.018","title":"Reply to the Editor— Clinical potential of conduction system pacing versus biventricular pacing in heart failure: A trial sequential analysis and methodological comment","year":2024,"lang":"en","type":"letter","venue":"Heart Rhythm","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Population Health Research Institute","funders":"","keywords":"Medicine; Cardiac pacing; Heart failure; Cardiology; Internal medicine","score_opus":0.2859121744866028,"score_gpt":0.44055001089441,"score_spread":0.1546378364078072,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392815531","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06977507,0.00036326077,0.00780874,0.88824755,0.031499676,0.0021622255,0.00006646549,0.000058651505,0.00001839265],"genre_scores_gemma":[0.5510204,0.000074241376,0.018481554,0.30024445,0.12885976,0.0006162508,0.00017640014,0.00011108619,0.00041587648],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9831243,0.00815515,0.002977537,0.0015812869,0.0036031196,0.00055858603],"domain_scores_gemma":[0.9924328,0.005170679,0.0005776202,0.001267061,0.00040133324,0.0001505421],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.035026617,0.0004133326,0.001874798,0.00085749035,0.00022679269,0.00046518305,0.00039888307,0.00071099895,0.00013993535],"category_scores_gemma":[0.0046510077,0.00024842448,0.0011731888,0.0022184348,0.00023487283,0.0001370552,0.00045298497,0.0019677216,0.00009978872],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018850396,0.00005986961,0.00031916253,0.000094608695,0.0004973728,0.00013710417,0.00011605669,0.0010311935,0.0000491552,0.000015137281,0.9954906,0.00030467595],"study_design_scores_gemma":[0.004267802,0.0007827293,0.00022708489,0.00024121118,0.0010574163,0.00009721952,0.000916555,0.00093078456,0.00014633799,0.00017264341,0.990846,0.0003142249],"about_ca_topic_score_codex":0.00058387336,"about_ca_topic_score_gemma":0.000066296496,"teacher_disagreement_score":0.58800304,"about_ca_system_score_codex":0.00024285876,"about_ca_system_score_gemma":0.00015552562,"threshold_uncertainty_score":0.9999968},"labels":[],"label_agreement":null},{"id":"W4392858836","doi":"10.1145/3626253.3635599","title":"A Generalized Framework for Describing Question Randomization","year":2024,"lang":"en","type":"article","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Generalizability theory; Randomization; Computer science; Randomized controlled trial; Randomized experiment; Machine learning; Artificial intelligence; Mathematics; Statistics; Medicine","score_opus":0.2836176893562416,"score_gpt":0.4298999124077332,"score_spread":0.14628222305149158,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392858836","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.028463094,0.00040348442,0.9641924,0.0037952564,0.0012249978,0.0005283825,0.0000029901516,0.000115132396,0.0012742579],"genre_scores_gemma":[0.80090505,0.000040774663,0.19439349,0.0005654101,0.0003450037,0.00013319746,0.000007100671,0.000010208851,0.003599792],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99820876,0.00019152288,0.0004480298,0.0003537449,0.000658634,0.0001393231],"domain_scores_gemma":[0.99752945,0.0019047289,0.000043052143,0.00025004358,0.0002275974,0.00004510993],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007404106,0.00007912973,0.0001488403,0.000121573146,0.00011502088,0.0006357662,0.00019872111,0.00006597907,0.0007823583],"category_scores_gemma":[0.005949764,0.000048326496,0.000143957,0.00040976517,0.000022005375,0.00025937194,0.000023312508,0.00006385868,0.00018656257],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016877883,0.00002848727,0.00036678836,0.000016315427,0.000017851335,6.349979e-7,0.00026312604,0.0018103563,0.0014800127,0.8756535,0.018942982,0.10125116],"study_design_scores_gemma":[0.00067497446,0.00003697574,0.00006583461,0.00009679321,0.000016201886,6.6102234e-7,0.00010224121,0.08883119,0.0019285551,0.87205,0.036106057,0.00009053762],"about_ca_topic_score_codex":0.000012369717,"about_ca_topic_score_gemma":0.000011360264,"teacher_disagreement_score":0.7724419,"about_ca_system_score_codex":0.000047577447,"about_ca_system_score_gemma":0.000037665563,"threshold_uncertainty_score":0.85662746},"labels":[],"label_agreement":null},{"id":"W4393385387","doi":"10.31219/osf.io/wzqxg","title":"Bayesian estimation in multiple comparisons","year":2024,"lang":"en","type":"preprint","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Bayesian probability; Estimation; Computer science; Bayes estimator; Artificial intelligence; Econometrics; Statistics; Machine learning; Mathematics; Economics","score_opus":0.2554361265663144,"score_gpt":0.41617030139968475,"score_spread":0.16073417483337032,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4393385387","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.21934928,0.0009624514,0.62406605,0.022101337,0.006880249,0.0025818285,0.00008081187,0.0003652168,0.123612754],"genre_scores_gemma":[0.9767734,0.000008635821,0.020726997,0.00013643294,0.00004933626,0.00006765537,0.000015964195,0.000009236308,0.0022123444],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9953755,0.00027751163,0.0011779645,0.00092060625,0.0020140922,0.00023431756],"domain_scores_gemma":[0.9976648,0.0008886389,0.00017186879,0.0010316349,0.00015827402,0.00008478944],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.006041273,0.00022400105,0.00043694745,0.00046432606,0.000056139514,0.0006684497,0.00093388563,0.00022284397,0.0012818706],"category_scores_gemma":[0.0024878823,0.00015471081,0.00019383627,0.0004867962,0.00006474676,0.00006880838,0.0013655411,0.00079752674,0.0020955505],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000036622638,0.000532301,0.115605764,0.0002333052,0.000053390453,0.000022330954,0.0013840036,0.52210075,0.0001116817,0.006092341,0.26127955,0.09254793],"study_design_scores_gemma":[0.00010980218,0.000012325424,0.010856953,0.00014542024,0.0000089704345,3.5464024e-7,0.00026281638,0.6707475,0.00011444719,0.3123533,0.0052230395,0.00016500353],"about_ca_topic_score_codex":0.00044998905,"about_ca_topic_score_gemma":0.002007557,"teacher_disagreement_score":0.7574241,"about_ca_system_score_codex":0.00017486159,"about_ca_system_score_gemma":0.000160144,"threshold_uncertainty_score":0.9996311},"labels":[],"label_agreement":null},{"id":"W4396912353","doi":"10.1016/j.neuchi.2024.101566","title":"Reliability of study endpoint adjudication in a pragmatic trial on brain arteriovenous malformations","year":2024,"lang":"en","type":"article","venue":"Neurochirurgie","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta Hospital; Centre Hospitalier de l’Université de Montréal; Health Sciences Centre","funders":"","keywords":"Medicine; Clinical endpoint; Adjudication; Inter-rater reliability; Adverse effect; Clinical trial; Reliability (semiconductor); Pediatrics; Surgery; Internal medicine; Psychology; Rating scale; Developmental psychology","score_opus":0.09841165656943207,"score_gpt":0.3866456987383119,"score_spread":0.2882340421688798,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4396912353","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98887664,0.0000321323,0.00055138447,0.004329209,0.00085140776,0.002027114,0.000018563293,0.000054257296,0.0032593159],"genre_scores_gemma":[0.99923563,0.0000058557257,0.00009598917,0.00028057763,0.00006025013,0.00014941671,0.00000385309,0.000008459981,0.00015998134],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9948919,0.0009081785,0.0015012291,0.0005736546,0.0019133308,0.00021170113],"domain_scores_gemma":[0.9950802,0.0035163828,0.00022019591,0.0009792125,0.00013978883,0.000064225],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.00823251,0.0001548465,0.00034822527,0.00045109744,0.00007410065,0.00017489493,0.0005219942,0.000051582232,0.00032527855],"category_scores_gemma":[0.009701512,0.00011055141,0.00015132943,0.0010875511,0.00006606372,0.00034113784,0.00008397814,0.00027757464,0.00029726914],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.07265651,0.06939993,0.02309579,0.0015403617,0.00026579844,0.00037892803,0.0707227,0.04301908,0.021705931,0.032340046,0.09963685,0.56523806],"study_design_scores_gemma":[0.03681269,0.021373305,0.6537602,0.00061084435,0.00011368318,0.00001093078,0.0061416277,0.030036503,0.001042128,0.18184581,0.06730283,0.0009494289],"about_ca_topic_score_codex":0.000028222472,"about_ca_topic_score_gemma":0.000057699934,"teacher_disagreement_score":0.6306644,"about_ca_system_score_codex":0.00009276145,"about_ca_system_score_gemma":0.00014034277,"threshold_uncertainty_score":0.9986402},"labels":[],"label_agreement":null},{"id":"W4398301064","doi":"10.7910/dvn/fmk6sq/bhbbmy","title":"agreement_results1.rtf","year":2020,"lang":"en","type":"dataset","venue":"Harvard Dataverse","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph","funders":"","keywords":"Methylene blue; Lymph; Staining; Degree (music); Methylene; Pathology; Medicine; Chemistry; Medicinal chemistry; Organic chemistry; Physics","score_opus":0.12691836379604451,"score_gpt":0.3431787049445279,"score_spread":0.2162603411484834,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4398301064","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000014785193,0.000002020934,0.00012611141,0.00023190625,0.001888043,0.00065678754,0.9955525,0.000058350808,0.0014695147],"genre_scores_gemma":[0.00001547404,0.00036270107,0.00021269763,0.0023599893,0.00070159836,0.000037701204,0.9948275,0.000016676684,0.0014656788],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9894438,0.00060042617,0.001696286,0.0017694051,0.0058815405,0.00060858787],"domain_scores_gemma":[0.99314517,0.00081148825,0.0008472427,0.0043271654,0.00038658147,0.00048237195],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0041376557,0.00063580886,0.0009900184,0.00034954585,0.00027059147,0.00091253273,0.004893061,0.00037586258,0.12796213],"category_scores_gemma":[0.00848715,0.00044952327,0.00046931882,0.0007505934,0.00021784492,0.0005363407,0.0017183031,0.0008235783,0.76305413],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008503475,0.00017561876,0.000033955163,0.000051615458,0.00008535568,0.00010580864,0.000023583649,0.000018906168,0.000026906437,0.000027965765,0.9978875,0.0014777435],"study_design_scores_gemma":[0.00058676174,0.00015044915,0.00008043122,0.00006635354,0.00012177617,0.0000037995558,0.00014615891,0.000034560326,0.000043443968,0.0007222542,0.9975346,0.0005094616],"about_ca_topic_score_codex":0.00015579928,"about_ca_topic_score_gemma":0.00021807144,"teacher_disagreement_score":0.635092,"about_ca_system_score_codex":0.00015513471,"about_ca_system_score_gemma":0.00020306445,"threshold_uncertainty_score":0.99986476},"labels":[],"label_agreement":null},{"id":"W4398457448","doi":"10.7910/dvn/hgihgc/5afaas","title":"study 1 analyses.R","year":2020,"lang":"en","type":"dataset","venue":"Harvard Dataverse","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Biology","score_opus":0.3457737450820914,"score_gpt":0.4375629867893381,"score_spread":0.09178924170724667,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4398457448","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000084246305,8.498526e-7,0.000066357105,0.00004119283,0.0011868782,0.00085772714,0.99735874,0.000040413324,0.00036360873],"genre_scores_gemma":[0.0001781554,0.00004902263,0.00012481477,0.0009442011,0.00040570888,0.000041846502,0.9978503,0.000012724278,0.00039324147],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9900882,0.0010007207,0.0014365151,0.0015894602,0.0054898285,0.00039523837],"domain_scores_gemma":[0.9934803,0.0006935783,0.00063833507,0.0044657504,0.00037690773,0.0003450903],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0052097533,0.00047178802,0.00093213335,0.0003815509,0.00023366149,0.000805093,0.004285842,0.00020359192,0.082048126],"category_scores_gemma":[0.00843936,0.00033528404,0.00034867498,0.0011064495,0.00011764204,0.000404729,0.0015783655,0.0006719866,0.7128122],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000046906553,0.00058359787,0.00015052765,0.000015006976,0.00016071514,0.00025228944,0.00007079348,0.000036439887,0.000008934524,0.0000022446677,0.9982451,0.0004274154],"study_design_scores_gemma":[0.0005092964,0.00028219333,0.0003573345,0.000018924296,0.00029390314,0.000002434269,0.0014856014,0.000015790883,0.0000069771536,0.00020586397,0.996457,0.000364691],"about_ca_topic_score_codex":0.00030347513,"about_ca_topic_score_gemma":0.00036754768,"teacher_disagreement_score":0.63076407,"about_ca_system_score_codex":0.00010572691,"about_ca_system_score_gemma":0.00021443557,"threshold_uncertainty_score":0.999913},"labels":[],"label_agreement":null},{"id":"W4398768486","doi":"10.1590/1980-265x-tce-2023-0171en","title":"CREATION AND CONTENT VALIDITY OF 5R-MEDSAFE: ASSESSING ADHERENCE TO THE SAFE DRUG ADMINISTRATION “FIVE RIGHTS”","year":2024,"lang":"en","type":"article","venue":"Texto & Contexto - Enfermagem","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Administration (probate law); Drug administration; Content validity; Drug; Content analysis; Content (measure theory); Political science; Medicine; Psychology; Nursing; Pharmacology; Sociology; Law; Social science; Health care; Mathematics","score_opus":0.770527616821497,"score_gpt":0.5405024020312108,"score_spread":0.23002521479028615,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4398768486","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9530723,0.00069741893,0.021842783,0.00761275,0.0010950139,0.0012034256,0.000055734694,0.00008391948,0.014336667],"genre_scores_gemma":[0.9924738,0.00002435478,0.0005705811,0.0004347237,0.00015684706,0.000058119917,0.000011709669,0.000010477172,0.0062594074],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9952313,0.00072350615,0.0011938179,0.0007585947,0.0017962427,0.0002965594],"domain_scores_gemma":[0.99618405,0.0019596783,0.00026995782,0.00076527294,0.0006497532,0.00017130846],"candidate_categories":["scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.007361514,0.00025770825,0.00040355566,0.00018125745,0.0003551516,0.0011129997,0.00065980246,0.000079523656,0.0009743237],"category_scores_gemma":[0.0016185564,0.00015162936,0.0001630069,0.0006060741,0.0002515207,0.0008561007,0.00018462115,0.0002573482,0.00037445946],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003103513,0.00044697625,0.009778741,0.0001765801,0.00016285706,0.00005604901,0.015312757,0.00024746402,0.023489246,0.042185653,0.11459538,0.7932379],"study_design_scores_gemma":[0.0016840557,0.0018200268,0.19940346,0.0014067217,0.00037170414,0.000048348797,0.016579533,0.014756198,0.04721631,0.063077234,0.6522689,0.0013675146],"about_ca_topic_score_codex":0.00040970644,"about_ca_topic_score_gemma":0.0008949664,"teacher_disagreement_score":0.7918704,"about_ca_system_score_codex":0.000104888204,"about_ca_system_score_gemma":0.00018751007,"threshold_uncertainty_score":0.9999389},"labels":[],"label_agreement":null},{"id":"W4398837123","doi":"10.7910/dvn/udgofr/uwf7cn","title":"Replication_code_PSLE_Riambau_Ostwald.do","year":2020,"lang":"en","type":"dataset","venue":"Harvard Dataverse","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Replication (statistics); Face (sociological concept); Code (set theory); Placebo; Computer science; Sociology; Biology; Programming language; Virology; Medicine; Social science; Alternative medicine","score_opus":0.13666328025013277,"score_gpt":0.35713924839825406,"score_spread":0.2204759681481213,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4398837123","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000024882174,0.000002899533,0.00045660362,0.00030321247,0.0015002736,0.00057285256,0.9959575,0.00006772812,0.0011364772],"genre_scores_gemma":[0.000013461352,0.00033532467,0.0005599987,0.0027017577,0.00062613,0.000062520434,0.9949212,0.000017450768,0.0007621636],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99133617,0.0005770481,0.0015263803,0.0018320178,0.004312997,0.0004153729],"domain_scores_gemma":[0.99129844,0.000929855,0.0008753046,0.005910487,0.00057033316,0.00041558102],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.004926148,0.00045597975,0.0007375905,0.00029710712,0.0002452357,0.0009309674,0.004312195,0.00034007287,0.11492541],"category_scores_gemma":[0.014824636,0.0003572971,0.00034891444,0.0008752126,0.0001776214,0.0004414017,0.0012850418,0.0006996709,0.7929514],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000045400797,0.00010306615,0.000021719316,0.00003136289,0.00004507177,0.000028805247,0.00001973996,0.000013578688,0.000017410188,0.00009931625,0.9957894,0.0037850935],"study_design_scores_gemma":[0.00027365802,0.000066888395,0.000046670517,0.000044009525,0.00007899152,0.000005169291,0.00008178609,0.00002876662,0.000020958923,0.0018536492,0.997101,0.00039842303],"about_ca_topic_score_codex":0.000073102055,"about_ca_topic_score_gemma":0.0000385017,"teacher_disagreement_score":0.678026,"about_ca_system_score_codex":0.00012724794,"about_ca_system_score_gemma":0.00028098596,"threshold_uncertainty_score":0.9998879},"labels":[],"label_agreement":null},{"id":"W4398951211","doi":"10.7910/dvn/s6b7se/z6le0q","title":"Replication_code_PSLE_Riambau_Ostwald.do","year":2020,"lang":"en","type":"dataset","venue":"Harvard Dataverse","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Replication (statistics); Code (set theory); Face (sociological concept); Placebo; Computer science; Psychology; Programming language; Medicine; Sociology; Virology; Alternative medicine; Social science; Pathology","score_opus":0.13666328025013277,"score_gpt":0.35713924839825406,"score_spread":0.2204759681481213,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4398951211","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000024882174,0.000002899533,0.00045660362,0.00030321247,0.0015002736,0.00057285256,0.9959575,0.00006772812,0.0011364772],"genre_scores_gemma":[0.000013461352,0.00033532467,0.0005599987,0.0027017577,0.00062613,0.000062520434,0.9949212,0.000017450768,0.0007621636],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99133617,0.0005770481,0.0015263803,0.0018320178,0.004312997,0.0004153729],"domain_scores_gemma":[0.99129844,0.000929855,0.0008753046,0.005910487,0.00057033316,0.00041558102],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.004926148,0.00045597975,0.0007375905,0.00029710712,0.0002452357,0.0009309674,0.004312195,0.00034007287,0.11492541],"category_scores_gemma":[0.014824636,0.0003572971,0.00034891444,0.0008752126,0.0001776214,0.0004414017,0.0012850418,0.0006996709,0.7929514],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000045400797,0.00010306615,0.000021719316,0.00003136289,0.00004507177,0.000028805247,0.00001973996,0.000013578688,0.000017410188,0.00009931625,0.9957894,0.0037850935],"study_design_scores_gemma":[0.00027365802,0.000066888395,0.000046670517,0.000044009525,0.00007899152,0.000005169291,0.00008178609,0.00002876662,0.000020958923,0.0018536492,0.997101,0.00039842303],"about_ca_topic_score_codex":0.000073102055,"about_ca_topic_score_gemma":0.0000385017,"teacher_disagreement_score":0.678026,"about_ca_system_score_codex":0.00012724794,"about_ca_system_score_gemma":0.00028098596,"threshold_uncertainty_score":0.9998879},"labels":[],"label_agreement":null},{"id":"W4399602664","doi":"10.2967/jnumed.123.267306","title":"Interreader and Intrareader Reproducibility of<sup>18</sup>F-Flotufolastat Image Interpretation in Patients with Newly Diagnosed or Recurrent Prostate Cancer: Data from Two Phase 3 Prospective Multicenter Studies","year":2024,"lang":"en","type":"article","venue":"Journal of Nuclear Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Medicine; Concordance; Nuclear medicine; Prostate cancer; Reproducibility; Prostate; Cancer; Radiology; Internal medicine; Mathematics","score_opus":0.14959149427134702,"score_gpt":0.43813232220661236,"score_spread":0.28854082793526536,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399602664","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.989865,0.0031895833,0.00016796653,0.0051901634,0.00066997064,0.0007645798,0.00011960907,0.000010382839,0.00002271574],"genre_scores_gemma":[0.99744594,0.0006941265,0.0014107815,0.00017000404,0.00022682,0.000005996167,0.0000109834855,0.000016856098,0.000018488234],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99531925,0.00039485487,0.0015832565,0.0009488422,0.0015620968,0.00019167396],"domain_scores_gemma":[0.9960345,0.0011735209,0.00062237534,0.00094477396,0.0011046508,0.00012019125],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006222767,0.00021504101,0.00071183767,0.00030082595,0.0000452618,0.00009940093,0.0005246072,0.000033119417,0.0004896284],"category_scores_gemma":[0.010833056,0.0001047455,0.000050155835,0.0004776912,0.00047514727,0.001115758,0.00023943621,0.00043104816,0.0000058602795],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.024658658,0.0020559547,0.40062436,0.00037411312,0.0009427639,0.00016916954,0.13751139,0.0003728871,0.0007112166,0.0000075040716,0.03559443,0.39697754],"study_design_scores_gemma":[0.048079,0.014959973,0.731822,0.020959884,0.0009774913,0.000017682087,0.10588177,0.061935645,0.00024499666,0.0032673718,0.011177469,0.00067668204],"about_ca_topic_score_codex":0.00013605764,"about_ca_topic_score_gemma":0.00017976659,"teacher_disagreement_score":0.39630088,"about_ca_system_score_codex":0.00021175951,"about_ca_system_score_gemma":0.00009395953,"threshold_uncertainty_score":0.9974991},"labels":[],"label_agreement":null},{"id":"W4402268166","doi":"10.1016/j.ajpe.2024.101095","title":"Generalizing AACP Poster Quality: Rater Experience Matters","year":2024,"lang":"en","type":"article","venue":"American Journal of Pharmaceutical Education","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Quality (philosophy); Psychology; Epistemology; Philosophy","score_opus":0.2456494393843736,"score_gpt":0.5436051578080804,"score_spread":0.2979557184237068,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402268166","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.93887025,0.00086510216,0.0047982074,0.05116973,0.0035155136,0.000102134916,0.000001943663,0.0000142590725,0.0006628658],"genre_scores_gemma":[0.9851016,0.00007731787,0.0016388894,0.012233021,0.00052169693,0.000006543327,6.730993e-7,0.000009469539,0.00041082565],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99614185,0.00062422873,0.0011081954,0.00029419633,0.0015962604,0.00023526314],"domain_scores_gemma":[0.99800396,0.0006475072,0.00033317498,0.00025709256,0.0004925858,0.00026567242],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0045899283,0.00013121669,0.00028022387,0.00027781233,0.00007605336,0.00049534143,0.0005241613,0.000022148188,0.0013572979],"category_scores_gemma":[0.00065509806,0.000086178414,0.00019398723,0.00071177445,0.00029743195,0.0006314563,0.000048471113,0.00029117553,0.0002435254],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000116871284,0.00029083897,0.009968118,0.000026497943,0.00006280277,0.000012343974,0.0066880295,0.00008374572,0.04947206,0.0009588068,0.034169532,0.8981504],"study_design_scores_gemma":[0.00041880764,0.00053579913,0.018864956,0.00032699876,0.00012140201,0.0002791172,0.026649838,0.0009931106,0.031340726,0.008834283,0.9111212,0.00051378674],"about_ca_topic_score_codex":0.00002754261,"about_ca_topic_score_gemma":7.911847e-7,"teacher_disagreement_score":0.8976366,"about_ca_system_score_codex":0.0001496922,"about_ca_system_score_gemma":0.000363623,"threshold_uncertainty_score":0.9995556},"labels":[],"label_agreement":null},{"id":"W4402813432","doi":"10.1016/j.arthro.2024.09.030","title":"<i>Editorial Commentary</i> : Machine Learning and Artificial Intelligence Are Valuable Tools yet Dependent on the Data Input","year":2024,"lang":"en","type":"editorial","venue":"Arthroscopy The Journal of Arthroscopic and Related Surgery","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Banff Centre; University of Calgary","funders":"","keywords":"Artificial intelligence; Computer science; Machine learning","score_opus":0.12254543520251504,"score_gpt":0.36209277246887484,"score_spread":0.2395473372663598,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402813432","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010636508,0.011868297,0.00015460416,0.014013839,0.96250015,0.0004464343,0.0002231285,0.00002418985,0.00013286641],"genre_scores_gemma":[0.030211516,0.05974818,0.000083582876,0.0009184424,0.9078915,0.000010707862,0.000110172994,0.00008898485,0.00093689235],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9890019,0.0023176344,0.002574275,0.00077527616,0.0048305723,0.0005003158],"domain_scores_gemma":[0.96453834,0.031466246,0.0019434398,0.0012134163,0.00063065276,0.00020793822],"candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.032347705,0.0005418979,0.0011347849,0.00028904388,0.0011109298,0.0013035394,0.0015515327,0.0006305647,0.00037045893],"category_scores_gemma":[0.01747166,0.00025871093,0.0002435521,0.00051808334,0.0007652965,0.0005903489,0.0010326636,0.0043123676,0.00011014764],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00030353264,0.00010766954,0.00043168492,0.000040341332,0.00031906686,0.000028883966,0.0004934278,0.00018366973,0.00005705172,0.000049241185,0.99177986,0.006205544],"study_design_scores_gemma":[0.00019061397,0.00050332595,0.000018031491,0.0016329203,0.00039648253,0.000021789878,0.0015924949,0.0005242502,0.0001936997,0.018612208,0.97598875,0.00032545358],"about_ca_topic_score_codex":0.000054604927,"about_ca_topic_score_gemma":0.00009576877,"teacher_disagreement_score":0.054608613,"about_ca_system_score_codex":0.000092716495,"about_ca_system_score_gemma":0.00043002446,"threshold_uncertainty_score":0.9999865},"labels":[],"label_agreement":null},{"id":"W4404100073","doi":"10.1183/13993003.congress-2024.pa2604","title":"A response bias exists away from Borg Category-Ratio 0-10 scale values without associated verbal descriptors for symptom intensity ratings during cardiopulmonary exercise testing","year":2024,"lang":"en","type":"article","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University Health Centre; University of British Columbia; McGill University","funders":"","keywords":"Intensity (physics); Psychology; Scale (ratio); Audiology; Statistics; Medicine; Mathematics; Physics","score_opus":0.16932530804045484,"score_gpt":0.34809937684267045,"score_spread":0.1787740688022156,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404100073","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9916208,0.00040395977,0.003183128,0.0006035891,0.0012359368,0.0009366096,0.00008583779,0.00043818358,0.0014919607],"genre_scores_gemma":[0.98531324,0.000004569558,0.0037556624,0.00016259351,0.00019873386,0.000081764614,0.000030038498,0.00003813187,0.01041527],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9935349,0.0006515572,0.0013763544,0.00133796,0.0024583868,0.00064084184],"domain_scores_gemma":[0.9921063,0.005207427,0.0002885759,0.0008072466,0.0013255978,0.00026485082],"candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.016912702,0.00040969608,0.00075604755,0.00034410303,0.0006826044,0.0011537545,0.0007188505,0.00022902849,0.00026211544],"category_scores_gemma":[0.017103778,0.000304028,0.00040020826,0.001199466,0.00021587791,0.000719981,0.0002974773,0.0003239782,0.00016980481],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008164972,0.00067524466,0.5863395,0.00017435025,0.00049781066,0.00019151128,0.012988853,0.0010211935,0.27697313,0.00007505393,0.090247706,0.02265072],"study_design_scores_gemma":[0.002733528,0.0009553063,0.6841968,0.0027181867,0.00056042115,0.000028615128,0.017147433,0.09592096,0.15861751,0.02663986,0.008249121,0.0022322237],"about_ca_topic_score_codex":0.0007867951,"about_ca_topic_score_gemma":0.00012723189,"teacher_disagreement_score":0.118355624,"about_ca_system_score_codex":0.00056039565,"about_ca_system_score_gemma":0.00031858386,"threshold_uncertainty_score":0.9999412},"labels":[],"label_agreement":null},{"id":"W4405750238","doi":"10.1016/j.jval.2024.10.2378","title":"MSR144 Group-Based Multivariate Trajectory Methods to Assess Patient-Reported Outcomes in Acute Myeloid Leukaemia: An Evaluation of Published Methods and Results","year":2024,"lang":"en","type":"article","venue":"Value in Health","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Geomechanica (Canada); Luminex (Canada)","funders":"","keywords":"Myeloid leukaemia; Multivariate statistics; Multivariate analysis; Medicine; Trajectory; Oncology; Internal medicine; Computer science; Machine learning; Physics","score_opus":0.40039424581550326,"score_gpt":0.5382868660662405,"score_spread":0.13789262025073729,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4405750238","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9311159,0.0005705833,0.0602182,0.004245875,0.0014455274,0.0021125285,0.00004919458,0.000056091216,0.00018610015],"genre_scores_gemma":[0.6406746,0.000017074544,0.35874942,0.00035884048,0.000020666113,0.00012895091,0.000017548495,0.00001570087,0.000017190321],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.96906453,0.020564085,0.0044551785,0.0016500527,0.0036723262,0.0005938215],"domain_scores_gemma":[0.9903333,0.006997318,0.000692706,0.001077518,0.00060439063,0.0002947286],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.174031,0.0002670331,0.00085143495,0.0013353303,0.00008929769,0.0002940428,0.00048277684,0.0001746331,0.00004923984],"category_scores_gemma":[0.026468204,0.00019674022,0.00012108919,0.0023585467,0.00007432526,0.00071426627,0.00011073494,0.0004479808,0.0000032808648],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00048854674,0.000860226,0.05057903,0.00024080124,0.00008708733,0.000014295339,0.008986162,0.0958443,0.011957301,0.0005039935,0.00032741102,0.83011085],"study_design_scores_gemma":[0.0027608052,0.001025158,0.46344322,0.0004342184,0.000063204425,0.0000026927885,0.0012632882,0.50558835,0.0037705745,0.020064846,0.0012416594,0.00034201393],"about_ca_topic_score_codex":0.005862232,"about_ca_topic_score_gemma":0.00061860256,"teacher_disagreement_score":0.82976884,"about_ca_system_score_codex":0.00073413947,"about_ca_system_score_gemma":0.0011711693,"threshold_uncertainty_score":0.98173225},"labels":[],"label_agreement":null},{"id":"W4408051986","doi":"10.1016/j.bjpt.2025.101193","title":"Considerations when designing, analyzing, and reporting reliability studies","year":2025,"lang":"en","type":"review","venue":"Brazilian Journal of Physical Therapy","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Reliability (semiconductor); Reliability engineering; Computer science; Engineering; Physics; Thermodynamics","score_opus":0.3977318700523943,"score_gpt":0.5038276533868976,"score_spread":0.10609578333450331,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408051986","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00054886204,0.9969512,0.00041116506,0.000641087,0.00046674552,0.0006020279,0.000008625792,0.000013330478,0.00035695444],"genre_scores_gemma":[0.0007275427,0.9956597,0.0025801663,0.00012144198,0.00050951564,0.000017610864,9.0583114e-7,0.000015970025,0.00036714788],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9897808,0.0020939445,0.005372491,0.000681052,0.0017756402,0.0002960728],"domain_scores_gemma":[0.97707206,0.011205695,0.008688362,0.00090733287,0.0019207478,0.00020578329],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.017396418,0.00048168533,0.0040406403,0.00044471075,0.00035821542,0.0004162945,0.0006755179,0.0001476743,0.00008881078],"category_scores_gemma":[0.043220304,0.0002653702,0.001632951,0.00063249015,0.00041174033,0.0003824211,0.00014506979,0.00079286617,0.000012200341],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000014939555,0.00021472253,0.0001856284,0.0010395971,0.0005393818,0.00002971816,0.0008430255,0.00002247924,0.000009784755,0.00038075887,0.009175063,0.9875449],"study_design_scores_gemma":[0.00039518057,0.0004884191,0.00007154773,0.010342082,0.000510934,0.00006126914,0.00030591074,0.000008146957,0.000043528682,0.28655446,0.700856,0.0003625242],"about_ca_topic_score_codex":0.0000019176034,"about_ca_topic_score_gemma":0.0000018397471,"teacher_disagreement_score":0.9871824,"about_ca_system_score_codex":0.00014293307,"about_ca_system_score_gemma":0.0008049056,"threshold_uncertainty_score":0.99997985},"labels":[],"label_agreement":null},{"id":"W4409833882","doi":"10.5271/sjweh.4232","title":"Response to Drs. Igor Burstyn and George Luta’s letter: Advice on better utilization of validation data to adjust odds ratios for differential exposure misclassification (recall bias)","year":2025,"lang":"en","type":"letter","venue":"Scandinavian Journal of Work Environment & Health","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"George (robot); Odds ratio; Odds; Medicine; Recall bias; Recall; Psychology; Logistic regression; Computer science; Internal medicine; Artificial intelligence; Cognitive psychology","score_opus":0.26826089703379796,"score_gpt":0.3820638494250824,"score_spread":0.11380295239128446,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409833882","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09151442,0.0002409426,0.05591193,0.848129,0.0012748307,0.0022873643,0.0006123232,0.000011202332,0.000018004817],"genre_scores_gemma":[0.33545506,0.0012204287,0.01738301,0.6314052,0.0056167473,0.00023752634,0.0019093236,0.00015911061,0.006613603],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99017584,0.002126,0.0028340404,0.0011247441,0.0031856322,0.0005537419],"domain_scores_gemma":[0.9934544,0.0017634447,0.0024406977,0.0017338438,0.00025278848,0.0003548205],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.010428263,0.00048200513,0.0010628153,0.0009890504,0.00031648093,0.00025002612,0.0015697051,0.00037685802,0.0003043954],"category_scores_gemma":[0.0016842713,0.00038110622,0.00022180549,0.00056072406,0.00010272236,0.00031644892,0.00027778567,0.0008227949,0.00003064349],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024483465,0.00024238457,0.0069821686,0.00018938606,0.00007891164,0.0000053893154,0.0005646634,0.0004834736,0.0007182743,0.000006312312,0.9384184,0.04986232],"study_design_scores_gemma":[0.0014899822,0.0025645976,0.07814965,0.0018935994,0.00015688561,0.0000048037828,0.00016992772,0.00011579849,0.0004363668,0.00020268277,0.914396,0.00041974866],"about_ca_topic_score_codex":0.000016178996,"about_ca_topic_score_gemma":0.000009605043,"teacher_disagreement_score":0.24394064,"about_ca_system_score_codex":0.00067126134,"about_ca_system_score_gemma":0.00025994226,"threshold_uncertainty_score":0.9998641},"labels":[],"label_agreement":null},{"id":"W4411404656","doi":"10.1017/9781009403092.013","title":"Interpreting and Reporting Your Results","year":2025,"lang":"en","type":"book-chapter","venue":"Cambridge University Press eBooks","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Computer science","score_opus":0.1805581086359696,"score_gpt":0.32449987844030603,"score_spread":0.14394176980433643,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411404656","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000720781,0.00013439397,0.0007152911,0.00020053315,0.00044717058,0.0004129555,0.00016206295,0.00007664067,0.99713016],"genre_scores_gemma":[0.0054939957,0.00005247286,0.0003086313,0.00011941402,0.00007497557,4.2532054e-7,0.000011184826,0.000012658779,0.9939262],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.99567544,0.00013828215,0.0015504938,0.001143145,0.0012139597,0.00027867337],"domain_scores_gemma":[0.99494296,0.00077702216,0.002414499,0.0011090161,0.00059467007,0.00016180746],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.004521613,0.0003486314,0.0006710936,0.00033759294,0.00034182763,0.00025245594,0.0008910664,0.0003187411,0.00000712799],"category_scores_gemma":[0.003973422,0.00032501246,0.00028426226,0.00002272371,0.0002658274,0.00012521849,0.0011789169,0.00056789897,0.000012601911],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00042640793,0.000017227521,0.00014685898,0.00011636829,0.00018495641,0.00045237847,0.00029216014,0.000012314997,0.00010019602,0.67298496,0.27533978,0.049926404],"study_design_scores_gemma":[0.00041492924,0.000045693658,0.00008332428,0.00062124495,0.00010196488,0.000008954999,0.00028528352,0.00020843414,0.00014365264,0.00023367144,0.9975215,0.00033131248],"about_ca_topic_score_codex":0.00013585719,"about_ca_topic_score_gemma":0.0000050817735,"teacher_disagreement_score":0.72218174,"about_ca_system_score_codex":0.00017747589,"about_ca_system_score_gemma":0.00016205657,"threshold_uncertainty_score":0.9999202},"labels":[],"label_agreement":null},{"id":"W4413408707","doi":"10.1093/annweh/wxaf046","title":"Comments on systematic review and meta-analysis of the epidemiology of man-made vitreous fibers and respiratory health outcomes","year":2025,"lang":"en","type":"article","venue":"Annals of Work Exposures and Health","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Stantec (Canada)","funders":"","keywords":"Epidemiology; Meta-analysis; Medicine; Environmental health; MEDLINE; Pathology; Biology","score_opus":0.6186994924523128,"score_gpt":0.5419520592477997,"score_spread":0.07674743320451316,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413408707","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.08987821,0.57368636,0.00006379973,0.33404946,0.00007504299,0.0020322348,0.00008652595,0.000005497012,0.00012287172],"genre_scores_gemma":[0.84003735,0.024673276,0.00020891815,0.13474198,0.0000033237582,0.00003512883,0.0000016417471,0.0000045928755,0.00029382703],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.993501,0.0028056726,0.0025001036,0.00035987643,0.0006084963,0.00022482932],"domain_scores_gemma":[0.9934889,0.0035205793,0.0018725174,0.0007979984,0.00021083545,0.00010913987],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.024561381,0.00015900444,0.0034954336,0.0002474908,0.00016126015,0.00000936233,0.0003210489,0.00005008913,0.000024023337],"category_scores_gemma":[0.0023726409,0.00007727375,0.00048210807,0.0007965993,0.00021918332,0.00003558267,0.00012768108,0.00011353249,3.083109e-7],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000690208,0.00034222947,0.8127989,0.095037036,0.026921233,2.7811976e-7,0.001034296,0.00011870695,0.0000019640672,0.0064422246,0.055259284,0.0019747927],"study_design_scores_gemma":[0.00032100524,0.0009271165,0.95885473,0.006183012,0.022832466,3.7435416e-7,0.0006837682,0.000015843796,0.00002938332,0.009158759,0.0008399429,0.00015356959],"about_ca_topic_score_codex":0.00028893366,"about_ca_topic_score_gemma":0.00018063442,"teacher_disagreement_score":0.7501591,"about_ca_system_score_codex":0.000012806453,"about_ca_system_score_gemma":0.00008066867,"threshold_uncertainty_score":0.85125333},"labels":[],"label_agreement":null},{"id":"W4414740488","doi":"10.1093/clinchem/hvaf086.543","title":"B-149 Establishing Pediatric Reference Intervals in a US Population using Refine R","year":2025,"lang":"en","type":"article","venue":"Clinical Chemistry","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Guideline; Calipers; Reference values; Outlier; Population; Confidence interval; Retrospective cohort study","score_opus":0.3851078962279259,"score_gpt":0.5191318721576088,"score_spread":0.1340239759296829,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414740488","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99183565,0.00014321263,0.0003419971,0.00053792546,0.00039263285,0.000102374324,0.0000043084606,0.000023304776,0.006618565],"genre_scores_gemma":[0.9973172,0.00004117106,0.0008525868,0.00027960405,0.00020639521,0.000005216625,0.0000075526295,0.000004021591,0.0012862494],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9961369,0.00025904173,0.0018722134,0.00066820654,0.0008304274,0.00023320211],"domain_scores_gemma":[0.99650866,0.0021255661,0.00033960078,0.0007018192,0.00023635503,0.00008802729],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009600917,0.00013002465,0.00038381742,0.00008733905,0.000073236675,0.0001980297,0.00074698706,0.00023067345,0.0002722636],"category_scores_gemma":[0.027325436,0.00010439969,0.00014216127,0.00097555446,0.000058412345,0.00024006977,0.000281605,0.00050884596,0.000024581306],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003213801,0.00014709662,0.991403,0.000040783616,0.0000036175588,0.0000029865737,0.000010150893,0.00020902159,0.00045456388,0.000028425884,0.0013803069,0.0062879194],"study_design_scores_gemma":[0.00093080674,0.000030392692,0.9584174,0.00023026903,0.000037845763,0.000001111344,0.00013179019,0.004488443,0.0011289078,0.02996057,0.0043782224,0.00026422588],"about_ca_topic_score_codex":0.00009317077,"about_ca_topic_score_gemma":0.000031910124,"teacher_disagreement_score":0.032985575,"about_ca_system_score_codex":0.00012833929,"about_ca_system_score_gemma":0.00011480118,"threshold_uncertainty_score":0.9808678},"labels":[],"label_agreement":null},{"id":"W4415900298","doi":"10.1136/jitc-2025-sitc2025.0097","title":"97 Statistical and systematic uncertainties affect biomarker reliability: a case study in multiplex immunofluorescence","year":2025,"lang":"","type":"article","venue":"Regular and Young Investigator Award Abstracts","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Multiplex; Biomarker; Immunofluorescence; Affect (linguistics)","score_opus":0.06942844363114892,"score_gpt":0.3442540651931056,"score_spread":0.27482562156195667,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415900298","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9843964,0.0070784097,0.00034291326,0.0011919491,0.001311748,0.005415051,0.000054602704,0.000053289798,0.00015562924],"genre_scores_gemma":[0.9977437,0.0002944605,0.00089880486,0.00019706845,0.00003320166,0.00023160393,0.00000439511,0.000027490736,0.00056926394],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9874678,0.003505121,0.0036111902,0.0022147158,0.0021935566,0.0010075689],"domain_scores_gemma":[0.9914002,0.0046511437,0.00067539385,0.001818284,0.0005577257,0.000897242],"candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.020333517,0.0008706836,0.001916002,0.00077891845,0.0007958395,0.0016059288,0.0007206884,0.00041940168,0.00004091349],"category_scores_gemma":[0.027733281,0.00070481247,0.00019588161,0.001467513,0.0023320506,0.00070298545,0.0006225563,0.00078635116,0.0000352731],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006240028,0.0037592382,0.9116843,0.037582494,0.00082368695,0.008122471,0.018464219,0.0007429695,0.0031827865,0.0012830896,0.0017409552,0.011989803],"study_design_scores_gemma":[0.006469405,0.0012937787,0.82402754,0.02068035,0.0012351953,0.001231843,0.09796921,0.029127391,0.000804911,0.015032262,0.00030537863,0.0018227344],"about_ca_topic_score_codex":0.005592987,"about_ca_topic_score_gemma":0.0016685603,"teacher_disagreement_score":0.087656744,"about_ca_system_score_codex":0.00039093182,"about_ca_system_score_gemma":0.00096360093,"threshold_uncertainty_score":0.9995403},"labels":[],"label_agreement":null},{"id":"W4416459542","doi":"10.1038/s44400-025-00033-z","title":"Author Correction: High inter-rater reliability in consensus diagnoses and overall assessment in the Asian Cohort for Alzheimer’s Disease Study","year":2025,"lang":"en","type":"article","venue":"npj Dementia","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canada Research Chairs; University of British Columbia; University of British Columbia Hospital","funders":"","keywords":"Medical diagnosis; Reliability (semiconductor); Disease; Cohort; Cohort study","score_opus":0.09375029666554337,"score_gpt":0.4100491476965076,"score_spread":0.3162988510309642,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4416459542","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.974276,0.00029170845,0.00077616365,0.017407253,0.0019665696,0.0039767986,0.000024069017,0.000020546677,0.0012609141],"genre_scores_gemma":[0.99758387,0.000009477376,0.00035751218,0.0006962245,0.00003658209,0.00089122134,0.000008050606,0.000005952349,0.00041113727],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99543434,0.0012267093,0.0010892097,0.0008662237,0.0010597468,0.00032379173],"domain_scores_gemma":[0.99666137,0.0019753915,0.00016803647,0.0008809632,0.00023119333,0.00008305085],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011571448,0.00021958622,0.0003676934,0.0002430739,0.0002188167,0.00038855826,0.00056122587,0.00006122604,0.0001840192],"category_scores_gemma":[0.0020633088,0.00014156297,0.00009775817,0.00051948807,0.00014111512,0.00015153555,0.00019818488,0.00026169428,0.000008908764],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015100317,0.001254717,0.9622034,0.000010794983,0.00007833379,0.0000097397415,0.00053467974,0.00016574566,0.000006261147,0.0006223727,0.026213143,0.008749816],"study_design_scores_gemma":[0.001255905,0.00021988082,0.9628841,0.000048575537,0.00019325143,3.2318414e-7,0.0032096067,0.001011539,0.000017790107,0.020682722,0.010332274,0.000143985],"about_ca_topic_score_codex":0.00047991824,"about_ca_topic_score_gemma":0.002238275,"teacher_disagreement_score":0.023307865,"about_ca_system_score_codex":0.00011259259,"about_ca_system_score_gemma":0.00016428033,"threshold_uncertainty_score":0.577277},"labels":[],"label_agreement":null},{"id":"W6045406","doi":"10.3138/cjpe.24.004","title":"Interrater Reliability in Content Analysis of Healthcare Service Quality Using Montreal’s Conceptual Framework","year":2009,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":true,"ca_institutions":"Université de Montréal","funders":"Universitetet i Bergen; Thomas Jefferson University","keywords":"Inter-rater reliability; Generalizability theory; Kappa; Reliability (semiconductor); Psychology; Cohen's kappa; Statistics; Content analysis; Service (business); Service quality; Health care; Applied psychology; Mathematics; Business; Sociology; Power (physics); Rating scale; Marketing; Political science","score_opus":0.6590834155962562,"score_gpt":0.5276751899838836,"score_spread":0.13140822561237253,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6045406","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99032587,0.00043444068,0.0019924792,0.006014579,0.00031982837,0.0008104693,0.000019702702,0.00000326706,0.000079339945],"genre_scores_gemma":[0.9958049,0.0000036910476,0.0035376367,0.0005863696,0.000044615634,0.000007678327,0.000007440222,0.0000031389004,0.000004528088],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9927831,0.0020657636,0.0023151864,0.00030909132,0.0022229352,0.00030392388],"domain_scores_gemma":[0.99167997,0.00042167038,0.0012591962,0.00051019934,0.0057282215,0.00040077462],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027292889,0.00013759766,0.0006513182,0.00091777404,0.00008740794,0.00013167862,0.000530112,0.00014258098,0.00038701922],"category_scores_gemma":[0.008222813,0.00010495899,0.00026248355,0.0028616334,0.00013020425,0.00035298106,0.000011368311,0.0003265235,0.000002550136],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001624962,0.0003029451,0.60624385,0.000017731072,0.00018842574,0.000005916031,0.013161607,0.055061467,0.00025220288,0.00064432085,0.000100017554,0.323859],"study_design_scores_gemma":[0.0005272082,0.0004706623,0.9306127,0.00014855628,0.00031212138,0.0000017084544,0.008429754,0.036908094,0.00007278523,0.022185057,0.00020549935,0.00012584343],"about_ca_topic_score_codex":0.03617253,"about_ca_topic_score_gemma":0.24625814,"teacher_disagreement_score":0.32436883,"about_ca_system_score_codex":0.00087317865,"about_ca_system_score_gemma":0.0018240806,"threshold_uncertainty_score":0.98440695},"labels":[],"label_agreement":null},{"id":"W6885476515","doi":"10.1371/journal.pmed.1003703.s003","title":"Quality assessment tool used in review for observational studies (adapted from Newcastle–Ottawa Scale).","year":2021,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Observational study; Quality assessment; Quality (philosophy); Risk assessment; Quality assurance","score_opus":0.8039214516836924,"score_gpt":0.5491709356804731,"score_spread":0.2547505160032193,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6885476515","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09536294,0.07073015,0.0008474689,0.06891678,0.0017257865,0.007492521,0.74812204,0.00020853829,0.0065938015],"genre_scores_gemma":[0.3451928,0.0029445314,0.091121204,0.057980005,0.0021147982,0.012610595,0.4761141,0.00014620523,0.011775782],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9951661,0.00059722294,0.0013026097,0.00068429543,0.0020091848,0.00024053916],"domain_scores_gemma":[0.9914044,0.005479679,0.0003680241,0.00076038606,0.0019200064,0.00006746776],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0033134012,0.00017355665,0.00060458935,0.00004681339,0.00012397581,0.00014729064,0.000508217,0.00006921602,0.10061794],"category_scores_gemma":[0.041954365,0.00013347386,0.00023206168,0.0006178439,0.000011790672,0.0003082826,0.0002096208,0.00014841468,0.00049565965],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000009743901,0.00018846657,0.01989024,0.0005227632,0.00004676861,0.000006120789,0.00012694388,0.00007883022,0.00026520263,0.00019054706,0.9642708,0.014403592],"study_design_scores_gemma":[0.00067514257,0.00003305699,0.4441422,0.005681912,0.000014129119,3.8479786e-7,0.0004929169,0.0003403542,0.00030206205,0.011218886,0.5368452,0.00025375126],"about_ca_topic_score_codex":0.000014221561,"about_ca_topic_score_gemma":0.0007017571,"teacher_disagreement_score":0.42742556,"about_ca_system_score_codex":0.00017874931,"about_ca_system_score_gemma":0.00035146705,"threshold_uncertainty_score":0.96611565},"labels":[],"label_agreement":null},{"id":"W6885830477","doi":"10.1371/journal.pone.0233732.s001","title":"Agreement test for Newcastle–Ottawa scale scores evaluated by reviewers.","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Scale (ratio); Measure (data warehouse); Test (biology); Level of measurement; Reliability (semiconductor)","score_opus":0.3904297146738943,"score_gpt":0.41125903525499835,"score_spread":0.02082932058110404,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6885830477","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0038256943,0.011963063,0.0011345933,0.081016965,0.00089163624,0.011476772,0.86538786,0.00043832528,0.023865074],"genre_scores_gemma":[0.68442786,0.00023486787,0.007540367,0.07237231,0.0037968443,0.0074109896,0.2024261,0.0002531777,0.021537503],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9955154,0.00016854171,0.0008833101,0.0007906252,0.0022628477,0.00037929282],"domain_scores_gemma":[0.99619365,0.0014538687,0.00033821573,0.000641193,0.0010593578,0.0003137125],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0016048861,0.00023125438,0.00039216207,0.00004844717,0.00020729305,0.00031273713,0.0011229643,0.000083763814,0.2617677],"category_scores_gemma":[0.045619812,0.00016676608,0.00025582916,0.0006206058,0.000017909348,0.00024133471,0.00020407964,0.00013811204,0.012933121],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000013358166,0.00007421797,0.00084482867,0.00011136553,0.000009283917,9.04386e-7,0.00011423529,0.00006107258,0.00064361363,0.0000020527998,0.9782669,0.019858189],"study_design_scores_gemma":[0.0005434988,0.00035737324,0.0012639156,0.0005259683,0.000017030816,4.9943634e-7,0.00010573238,0.004175024,0.0022524938,0.00035883527,0.9901652,0.00023442543],"about_ca_topic_score_codex":0.0000044538106,"about_ca_topic_score_gemma":0.000015321797,"teacher_disagreement_score":0.68060213,"about_ca_system_score_codex":0.000062380714,"about_ca_system_score_gemma":0.000100301746,"threshold_uncertainty_score":0.9878354},"labels":[],"label_agreement":null},{"id":"W6885846436","doi":"10.1371/journal.pone.0262885.s003","title":"Newcastle-Ottawa quality assessment form for cohort studies.","year":2022,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Cohort; Quality assessment; Quality (philosophy); Risk assessment; Cohort study; Population","score_opus":0.6777878290963861,"score_gpt":0.5391566899403626,"score_spread":0.13863113915602354,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6885846436","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029364366,0.0020394288,0.0005732325,0.012672595,0.003362603,0.009453161,0.88331133,0.00035170978,0.05887156],"genre_scores_gemma":[0.917643,0.0000108414415,0.0030042753,0.004980645,0.0006031442,0.014978663,0.043750767,0.0000524345,0.014976237],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9947128,0.00036155598,0.00088667904,0.00064872537,0.0030428295,0.00034740847],"domain_scores_gemma":[0.99522567,0.002584852,0.0004037082,0.00085296674,0.00082627224,0.00010653849],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0065512983,0.00016362236,0.00037733416,0.00009322131,0.0007933159,0.000159889,0.0010319956,0.000035150646,0.2519884],"category_scores_gemma":[0.012000899,0.00012568625,0.00027062223,0.00045387723,0.0000134078555,0.00020620368,0.0006825251,0.00020917908,0.000569481],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000141710525,0.000115727526,0.0045265933,0.000043460153,0.00006595397,0.0000017242643,0.00020409295,0.0003278672,0.000022395472,0.00033176097,0.98299396,0.011352278],"study_design_scores_gemma":[0.00042980118,0.00019760433,0.030640956,0.000044513603,0.000012727236,0.0000014525095,0.0021594788,0.0005126204,0.00006328803,0.013078014,0.9526623,0.00019725796],"about_ca_topic_score_codex":0.000009541164,"about_ca_topic_score_gemma":0.00003879167,"teacher_disagreement_score":0.8882786,"about_ca_system_score_codex":0.00042608744,"about_ca_system_score_gemma":0.0001851561,"threshold_uncertainty_score":0.99632144},"labels":[],"label_agreement":null},{"id":"W6885849940","doi":"10.1371/journal.pone.0299125.s009","title":"Author’s judgements about study quality using the adapted Ottawa-Newcastle risk of bias assessment tool.","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Quality (philosophy); Risk assessment; Quality assessment; Response bias; Data collection; Risk perception","score_opus":0.6038026136552229,"score_gpt":0.5165250097033885,"score_spread":0.08727760395183437,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6885849940","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8320861,0.0012482705,0.0008433603,0.0012179693,0.002644293,0.004171211,0.14846186,0.00022466239,0.0091022635],"genre_scores_gemma":[0.9972012,0.0000018884937,0.00045225216,0.00007751813,0.00015236188,0.00012126637,0.00073748385,0.000015503118,0.0012405005],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99207175,0.0017850715,0.0013730999,0.0006988526,0.003763379,0.0003078436],"domain_scores_gemma":[0.99459255,0.002808447,0.0005107743,0.0012802068,0.0007264308,0.000081618135],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.010806237,0.00022777983,0.00038179295,0.00015816705,0.00032440867,0.00064069725,0.0011422231,0.00007542975,0.10456625],"category_scores_gemma":[0.01340514,0.0001283826,0.00027697394,0.0010644087,0.000026401874,0.00032118277,0.0004408298,0.00035805555,0.001014778],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000029638042,0.0010169972,0.06568913,0.00020528765,0.00041155558,0.000020681073,0.0027509334,0.0037012936,0.0002946745,0.00026479573,0.88595974,0.039655246],"study_design_scores_gemma":[0.0005841135,0.00034618107,0.5492768,0.0012094519,0.0001389522,0.0000016322336,0.0064059985,0.026744485,0.00034524983,0.0027761078,0.41173953,0.00043143443],"about_ca_topic_score_codex":0.00018614465,"about_ca_topic_score_gemma":0.00016485422,"teacher_disagreement_score":0.4835877,"about_ca_system_score_codex":0.00014447437,"about_ca_system_score_gemma":0.0002500163,"threshold_uncertainty_score":0.9997631},"labels":[],"label_agreement":null},{"id":"W6885928307","doi":"10.1371/journal.pone.0304473.s004","title":"Methodological quality score of the included studies based on the Newcastle—Ottawa Scale (NOS) tool.","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Scale (ratio); Quality (philosophy); Quality Score; Data quality; Measure (data warehouse)","score_opus":0.8708561416807558,"score_gpt":0.550798302728561,"score_spread":0.32005783895219475,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6885928307","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.53661793,0.008178464,0.0005364759,0.16395158,0.005622192,0.0076058,0.2139218,0.0006035831,0.062962145],"genre_scores_gemma":[0.99588484,0.0000034206607,0.0003514132,0.0022457177,0.0001449923,0.0001772834,0.00023543769,0.000008624905,0.0009482461],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9935898,0.0025511628,0.00075131626,0.00050281186,0.0023928946,0.00021201093],"domain_scores_gemma":[0.9749516,0.022987423,0.00023078395,0.0012220036,0.0005636841,0.00004446804],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0105686765,0.00016793863,0.00035016195,0.00005842268,0.00027814292,0.00019838338,0.0013650403,0.000084834755,0.062502705],"category_scores_gemma":[0.09293068,0.00006304261,0.00036332844,0.00083673233,0.000093886294,0.00009780128,0.0004930156,0.00031307177,0.0010368633],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000031601005,0.00009988516,0.0015446515,0.00016098449,0.000039415932,0.0000028201218,0.00057313405,0.00073162036,0.00041558227,0.00089108443,0.9831888,0.012320391],"study_design_scores_gemma":[0.0004948297,0.00041476588,0.20565802,0.006215807,0.000055398123,0.0000023486268,0.0033199235,0.006714898,0.022213874,0.071993075,0.6823454,0.000571661],"about_ca_topic_score_codex":0.000007692618,"about_ca_topic_score_gemma":0.00004191694,"teacher_disagreement_score":0.4592669,"about_ca_system_score_codex":0.000067855646,"about_ca_system_score_gemma":0.00013127943,"threshold_uncertainty_score":0.99974096},"labels":[],"label_agreement":null},{"id":"W6885947014","doi":"10.1371/journal.pone.0300564.s002","title":"Correlation matrix: Test of multicollinearity.","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Medical diagnosis; Retrospective cohort study; Neurocognitive; Bacteriuria; Antibiotics; Incidence (geometry); Urinary system; Presentation (obstetrics); Clinical diagnosis","score_opus":0.2097736899081875,"score_gpt":0.4081235560669805,"score_spread":0.19834986615879296,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6885947014","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.036489334,0.010823738,0.0027823464,0.006836968,0.0039633387,0.003984961,0.8419294,0.00092223135,0.0922677],"genre_scores_gemma":[0.9923203,0.0000022894255,0.00038799513,0.00004523124,0.000112020956,0.000034941437,0.0033336862,0.00000700385,0.0037565506],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9981319,0.000049672373,0.00041738193,0.0002502629,0.0010482415,0.00010255725],"domain_scores_gemma":[0.9970688,0.0022047856,0.00007667202,0.00030521437,0.00030160332,0.00004290142],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00069633767,0.00006950235,0.00011869544,0.00010738806,0.000042984142,0.00017585966,0.00031032096,0.000056966677,0.21117651],"category_scores_gemma":[0.01622393,0.000048576858,0.00009799323,0.0005336738,0.000007742407,0.00018929406,0.00007204716,0.000103253784,0.0119158905],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00000483702,0.00009313449,0.0026341286,0.00014905023,0.000009092949,0.000007448313,0.00022451108,0.0021825156,0.0006654963,0.00014500771,0.9613965,0.032488264],"study_design_scores_gemma":[0.0001585747,0.00009806806,0.020078652,0.0014271049,0.000010127873,0.0000026654661,0.000115260256,0.224489,0.0022620778,0.004081401,0.74712384,0.00015321518],"about_ca_topic_score_codex":0.0000039087104,"about_ca_topic_score_gemma":0.0000050573644,"teacher_disagreement_score":0.95583093,"about_ca_system_score_codex":0.000026469474,"about_ca_system_score_gemma":0.00006218442,"threshold_uncertainty_score":0.9920628},"labels":[],"label_agreement":null},{"id":"W6885993894","doi":"10.1371/journal.pone.0305355.s004","title":"Quality assessment of included studies by Newcastle-Ottawa scale.","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Quality assessment; Quality (philosophy); Risk assessment; Quality assurance; Work (physics)","score_opus":0.5664514499662818,"score_gpt":0.5443827685831959,"score_spread":0.02206868138308593,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6885993894","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.123133086,0.035021193,0.0004567474,0.014293736,0.0032192273,0.0028962682,0.7078447,0.0006423722,0.11249269],"genre_scores_gemma":[0.991322,0.00002586054,0.0005424174,0.00028359814,0.00012763625,0.00014808917,0.003924601,0.000013456435,0.0036123674],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9953762,0.0003776384,0.0009879164,0.000546777,0.0024923664,0.0002190493],"domain_scores_gemma":[0.9960862,0.002274986,0.0002067064,0.0006376455,0.000706884,0.00008761072],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0033360226,0.00016085568,0.000400179,0.00011115088,0.00010825768,0.00021816706,0.0006898961,0.00006989482,0.107882924],"category_scores_gemma":[0.007662968,0.000106510204,0.00020395003,0.00064072176,0.00003362916,0.0002748809,0.00037187597,0.00017799756,0.001487441],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000026147625,0.00007237376,0.0012872773,0.0002200072,0.00004773061,0.000002431946,0.0003009065,0.000032122585,0.0009760229,0.000103859864,0.98195404,0.015000596],"study_design_scores_gemma":[0.0002219604,0.00012883238,0.023132276,0.0015297802,0.000013801645,0.0000012550823,0.0014209094,0.00088302465,0.0033546167,0.008309518,0.9607418,0.00026219315],"about_ca_topic_score_codex":0.000012314875,"about_ca_topic_score_gemma":0.000043299297,"teacher_disagreement_score":0.8681889,"about_ca_system_score_codex":0.00011624148,"about_ca_system_score_gemma":0.000136773,"threshold_uncertainty_score":0.99929},"labels":[],"label_agreement":null},{"id":"W6887749040","doi":"10.17605/osf.io/qrf2w","title":"Assessing the validity and reliability of a baseball pitch discrimination online task","year":2023,"lang":"en","type":"article","venue":"OSF Preprints (OSF Preprints)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Reliability (semiconductor); Task (project management); Action (physics); Validity; Task analysis","score_opus":0.1459806906473429,"score_gpt":0.3918010649520061,"score_spread":0.24582037430466322,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6887749040","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9424802,8.01323e-7,0.0038312045,0.0034594662,0.00031481896,0.0007284104,0.00002865798,0.00007268623,0.04908377],"genre_scores_gemma":[0.9778039,0.00005451649,0.000981056,0.00010215017,0.000042300442,0.000057431967,0.000015594951,0.000011916452,0.020931175],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9919041,0.0024382954,0.0012379177,0.0019122532,0.002154964,0.00035249052],"domain_scores_gemma":[0.9894068,0.0051728445,0.0005295012,0.004096683,0.00065115833,0.00014299715],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.053052116,0.00022129851,0.0004097574,0.00017677962,0.00035048954,0.0003908933,0.0015404429,0.00013064657,0.05939811],"category_scores_gemma":[0.04450454,0.00015103543,0.00020255877,0.0008819092,0.00044099562,0.0006665411,0.0017950656,0.00037015244,0.06939658],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006730224,0.001098457,0.8549702,0.00019114344,0.000046286215,0.000007220523,0.004505678,0.006010968,0.018257348,0.0009034901,0.029426858,0.08451505],"study_design_scores_gemma":[0.00030849574,0.0000025947816,0.9129551,0.000068903806,0.00004740844,0.0000027973788,0.0021080952,0.009761328,0.0036217845,0.054777414,0.016154695,0.00019134964],"about_ca_topic_score_codex":0.00019501509,"about_ca_topic_score_gemma":0.00007571572,"teacher_disagreement_score":0.084323704,"about_ca_system_score_codex":0.00012592493,"about_ca_system_score_gemma":0.00013487214,"threshold_uncertainty_score":0.9750821},"labels":[],"label_agreement":null},{"id":"W6901989955","doi":"10.6084/m9.figshare.12233456","title":"NOS","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Scale (ratio); Field (mathematics); Process (computing)","score_opus":0.5584769089976009,"score_gpt":0.40391553303996286,"score_spread":0.15456137595763808,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6901989955","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011571144,0.00086269685,0.00023050002,0.09447224,0.00067485886,0.0019049539,0.3900353,0.0006240234,0.49962428],"genre_scores_gemma":[0.9841181,6.7430597e-7,0.00026530505,0.009994936,0.00039132722,0.000060365903,0.0042890906,0.000008153838,0.0008720193],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9982058,0.00006000025,0.00024510923,0.00027568324,0.0010942183,0.00011917998],"domain_scores_gemma":[0.9990261,0.00032145125,0.000067589215,0.000267979,0.0001858871,0.0001309778],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00025531813,0.00006129949,0.000100324476,0.000018960736,0.000058541762,0.0001460236,0.0006046065,0.000030054343,0.68612653],"category_scores_gemma":[0.017098676,0.000041870346,0.000076153294,0.0003180307,0.0000033491615,0.00012891178,0.00013032988,0.000079056146,0.08712512],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000029298178,0.000008563244,0.00023921505,0.000006163445,0.0000017019398,0.0000027304022,0.00012049738,0.000046363257,0.000084901745,0.000018571931,0.99030215,0.009166219],"study_design_scores_gemma":[0.00008520784,0.000037376238,0.004114649,0.000046399015,8.5778163e-7,2.6057074e-7,0.00007786632,0.0006778191,0.0007813898,0.0009414135,0.9931691,0.00006764705],"about_ca_topic_score_codex":5.569714e-7,"about_ca_topic_score_gemma":0.0000010910662,"teacher_disagreement_score":0.972547,"about_ca_system_score_codex":0.000009050245,"about_ca_system_score_gemma":0.000031689004,"threshold_uncertainty_score":0.9911807},"labels":[],"label_agreement":null},{"id":"W6902034213","doi":"10.6084/m9.figshare.26579069.v1","title":"Additional file 1 of Improving social justice in observational studies: protocol for the development of a global and Indigenous STROBE-equity reporting guideline","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Agency for Drugs and Technologies in Health; Ottawa Hospital; Carleton University; Hospital for Sick Children; Western University; Royal College of Physicians and Surgeons of Canada; University of Toronto; Dalhousie University; SickKids Foundation; Bruyère; Canadian Council on Animal Care; McMaster University; Queen's University; University of Ottawa","funders":"","keywords":"Guideline; Indigenous; Observational study; Protocol (science); Coding (social sciences); Social justice; Context (archaeology)","score_opus":0.6703220811909395,"score_gpt":0.5374888847959403,"score_spread":0.13283319639499924,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6902034213","genre_codex":"dataset","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00009209626,0.000028016382,0.00014918012,0.00016024757,0.00003195179,0.017350292,0.9819728,0.000009413498,0.00020598953],"genre_scores_gemma":[0.0029800474,1.5183788e-7,0.07547168,0.0002599653,0.0005742423,0.69990456,0.22053501,0.000015605527,0.00025876166],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99664134,0.000044593962,0.001861563,0.00026467835,0.0010442046,0.00014364294],"domain_scores_gemma":[0.9917653,0.0058737905,0.0011143117,0.00013034827,0.0010938261,0.000022431263],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00246596,0.00008663749,0.00019894275,0.000032870386,0.00020589026,0.000064978056,0.00025416454,0.000042201053,0.48201004],"category_scores_gemma":[0.07787108,0.00005433508,0.000070426235,0.00035073832,0.00002241362,0.00012564502,0.00027571892,0.00007070857,0.000012264642],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000016535763,0.000035750607,0.000038002152,0.0006191383,0.000022821523,8.3759676e-7,0.000678497,0.00023955823,0.0000062317467,0.000014591064,0.9658148,0.03251321],"study_design_scores_gemma":[0.00028978928,0.0000775589,0.022499476,0.0030822672,0.000014823608,0.0000021609821,0.0034116453,0.008984269,0.000049353228,0.002187175,0.9592621,0.00013937982],"about_ca_topic_score_codex":0.000002426363,"about_ca_topic_score_gemma":0.00010456917,"teacher_disagreement_score":0.7614378,"about_ca_system_score_codex":0.0001460701,"about_ca_system_score_gemma":0.0010008693,"threshold_uncertainty_score":0.9298964},"labels":[],"label_agreement":null},{"id":"W6902157991","doi":"10.6084/m9.figshare.13290662","title":"Additional file 1 of Developing a coding taxonomy to analyze dental regulatory complaints","year":2020,"lang":"en","type":"article","venue":"Open MIND","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; University of Toronto","funders":"","keywords":"Coding (social sciences); Reliability (semiconductor); Taxonomy (biology); Table (database); Data collection","score_opus":0.32342565138858714,"score_gpt":0.3722898595968911,"score_spread":0.04886420820830395,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6902157991","genre_codex":"dataset","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18847546,0.00002155228,0.011180551,0.008655641,0.0004619071,0.0053989193,0.6169406,0.000010434375,0.16885494],"genre_scores_gemma":[0.37733233,6.8797067e-7,0.6073156,0.0024320143,0.00026941454,0.0008367914,0.007830445,0.00001734132,0.003965351],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9977735,0.00010439173,0.0005408713,0.00041145415,0.00102516,0.00014459786],"domain_scores_gemma":[0.9977864,0.0014065666,0.00020179433,0.00024130671,0.00021110059,0.00015283703],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.000899889,0.00009062296,0.00023695131,0.00006949145,0.0001233523,0.00019808931,0.0010194788,0.00002934109,0.9158635],"category_scores_gemma":[0.0034961216,0.00007537738,0.00006906353,0.0005008732,0.00005859361,0.00023316065,0.0004865757,0.0000642658,0.0044454588],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000020185595,0.000021860476,0.00055593654,0.0000020416703,0.000012936627,0.0000024595731,0.00014146203,0.00006595369,0.00018382836,0.000021136566,0.9624308,0.036541402],"study_design_scores_gemma":[0.00012554337,0.000042966476,0.008106675,0.00011275527,0.0000033875992,0.0000012469093,0.00096749066,0.0002760077,0.0012000942,0.00010540807,0.98895025,0.000108150816],"about_ca_topic_score_codex":0.0000050435265,"about_ca_topic_score_gemma":0.00004389756,"teacher_disagreement_score":0.911418,"about_ca_system_score_codex":0.000060519098,"about_ca_system_score_gemma":0.00017780525,"threshold_uncertainty_score":0.9963297},"labels":[],"label_agreement":null},{"id":"W6902295241","doi":"10.6084/m9.figshare.26579069","title":"Additional file 1 of Improving social justice in observational studies: protocol for the development of a global and Indigenous STROBE-equity reporting guideline","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Agency for Drugs and Technologies in Health; Ottawa Hospital; Carleton University; Hospital for Sick Children; Western University; Royal College of Physicians and Surgeons of Canada; University of Toronto; Dalhousie University; SickKids Foundation; Bruyère; Canadian Council on Animal Care; McMaster University; Queen's University; University of Ottawa","funders":"","keywords":"Guideline; Indigenous; Observational study; Protocol (science); Coding (social sciences); Social justice; Context (archaeology)","score_opus":0.6703220811909395,"score_gpt":0.5374888847959403,"score_spread":0.13283319639499924,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6902295241","genre_codex":"dataset","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00009209626,0.000028016382,0.00014918012,0.00016024757,0.00003195179,0.017350292,0.9819728,0.000009413498,0.00020598953],"genre_scores_gemma":[0.0029800474,1.5183788e-7,0.07547168,0.0002599653,0.0005742423,0.69990456,0.22053501,0.000015605527,0.00025876166],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99664134,0.000044593962,0.001861563,0.00026467835,0.0010442046,0.00014364294],"domain_scores_gemma":[0.9917653,0.0058737905,0.0011143117,0.00013034827,0.0010938261,0.000022431263],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00246596,0.00008663749,0.00019894275,0.000032870386,0.00020589026,0.000064978056,0.00025416454,0.000042201053,0.48201004],"category_scores_gemma":[0.07787108,0.00005433508,0.000070426235,0.00035073832,0.00002241362,0.00012564502,0.00027571892,0.00007070857,0.000012264642],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000016535763,0.000035750607,0.000038002152,0.0006191383,0.000022821523,8.3759676e-7,0.000678497,0.00023955823,0.0000062317467,0.000014591064,0.9658148,0.03251321],"study_design_scores_gemma":[0.00028978928,0.0000775589,0.022499476,0.0030822672,0.000014823608,0.0000021609821,0.0034116453,0.008984269,0.000049353228,0.002187175,0.9592621,0.00013937982],"about_ca_topic_score_codex":0.000002426363,"about_ca_topic_score_gemma":0.00010456917,"teacher_disagreement_score":0.7614378,"about_ca_system_score_codex":0.0001460701,"about_ca_system_score_gemma":0.0010008693,"threshold_uncertainty_score":0.9298964},"labels":[],"label_agreement":null},{"id":"W6904071898","doi":"10.1371/journal.pntd.0012689.s002","title":"Quality assessment of epidemiological studies from the year 2000, based on the modified Newcastle-Ottawa Quality Assessment Scale","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Quality assessment; Epidemiology; Quality (philosophy); Risk assessment; Scale (ratio); Exposure assessment","score_opus":0.7233629315348483,"score_gpt":0.5464145026599271,"score_spread":0.17694842887492124,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6904071898","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.25088796,0.004946745,0.0046186093,0.16421387,0.0031303728,0.0064140907,0.47121343,0.00052521477,0.09404973],"genre_scores_gemma":[0.9944366,0.000011690533,0.0006982928,0.0022065293,0.00021260182,0.0003055172,0.0017328232,0.000010705831,0.00038523183],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9891348,0.004888348,0.0015499354,0.00086091674,0.0032248495,0.00034118522],"domain_scores_gemma":[0.9517428,0.04536821,0.000476672,0.0017184968,0.0005950138,0.000098774275],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01902766,0.0002813243,0.00066900963,0.0000538893,0.00031992057,0.00029233992,0.001461535,0.00012969748,0.07248909],"category_scores_gemma":[0.027996408,0.000116343515,0.0004781539,0.0006029506,0.00012186778,0.00013009975,0.00035961956,0.00057542016,0.00073985645],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003620116,0.000342256,0.015713997,0.00010418745,0.00015337659,0.0000039793936,0.0004610545,0.0053934087,0.00016172674,0.0055508385,0.9636062,0.008472772],"study_design_scores_gemma":[0.0003926728,0.00022708936,0.840667,0.0013350975,0.000032831133,2.1112069e-7,0.0033019518,0.032786828,0.00018663779,0.03384592,0.086884364,0.00033936303],"about_ca_topic_score_codex":0.0001300982,"about_ca_topic_score_gemma":0.00011193948,"teacher_disagreement_score":0.87672186,"about_ca_system_score_codex":0.00026142382,"about_ca_system_score_gemma":0.00037398603,"threshold_uncertainty_score":0.9801912},"labels":[],"label_agreement":null},{"id":"W6904357030","doi":"10.1371/journal.pntd.0009890.s003","title":"QA assessment of STH studies based on modified Newcastle-Ottawa Quality Assessment Scale.","year":2021,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Quality assessment; Quality (philosophy); Quality assurance; Risk assessment; Evaluation methods","score_opus":0.6275608607273189,"score_gpt":0.538625661623456,"score_spread":0.08893519910386283,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6904357030","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.21415535,0.00206553,0.004355299,0.024660721,0.003499126,0.0052165203,0.38070613,0.00048374044,0.36485758],"genre_scores_gemma":[0.9899416,0.0000065840786,0.0033613252,0.0009813905,0.00009445558,0.00019616699,0.0042831358,0.000014736074,0.001120599],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9917494,0.0012136181,0.0013553193,0.00088319246,0.0044449787,0.0003535003],"domain_scores_gemma":[0.9921246,0.0033256426,0.00062044186,0.0015230394,0.002248179,0.00015806232],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0039603817,0.0002739731,0.00070758996,0.00014759338,0.00020721696,0.00020159708,0.0007735198,0.00010616783,0.08421524],"category_scores_gemma":[0.010241829,0.00020056542,0.00034748513,0.000771318,0.000038762646,0.00019119162,0.00031336866,0.00029541212,0.00034257682],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006120623,0.00266809,0.025590213,0.00074380124,0.00023380964,0.00005518156,0.00043096123,0.043860693,0.0020108607,0.0014120775,0.8917677,0.031165443],"study_design_scores_gemma":[0.003143623,0.0008908559,0.7941804,0.003843771,0.00006820368,0.000002503031,0.004881391,0.05076939,0.01074155,0.008178221,0.122215755,0.0010843732],"about_ca_topic_score_codex":0.000010390934,"about_ca_topic_score_gemma":0.00008319355,"teacher_disagreement_score":0.7757863,"about_ca_system_score_codex":0.00027170428,"about_ca_system_score_gemma":0.0006688619,"threshold_uncertainty_score":0.99809533},"labels":[],"label_agreement":null},{"id":"W6904401493","doi":"10.1371/journal.pone.0276701.s004","title":"Newcastle-Ottawa quality assessment scale for cross-sectional studies.","year":2022,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Scale (ratio); Quality (philosophy); Quality assessment; Measure (data warehouse); Data quality","score_opus":0.706144065715121,"score_gpt":0.5664891265198829,"score_spread":0.13965493919523808,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6904401493","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.14459492,0.00091589906,0.00038090078,0.0035373396,0.0036943734,0.0034438244,0.8223052,0.0002285571,0.02089899],"genre_scores_gemma":[0.95843554,0.0000017563781,0.0016277084,0.0017058104,0.00056227343,0.004521335,0.020947902,0.000024563033,0.012173126],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9946362,0.0004072499,0.00088793464,0.0006857917,0.0030785776,0.00030426972],"domain_scores_gemma":[0.9955282,0.0023410302,0.00036133817,0.0006339691,0.0010375091,0.000097929296],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.005658834,0.00014579445,0.00027650196,0.000091972506,0.0011125444,0.00030931082,0.0009068103,0.00004143439,0.31322455],"category_scores_gemma":[0.008553163,0.0001175691,0.00028850161,0.00042194678,0.00002366429,0.00021540433,0.00061355846,0.00022629385,0.0006870659],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000028590499,0.00017363156,0.07006221,0.000045427707,0.000028858596,0.0000011814668,0.00014584528,0.0025427802,0.000064627035,0.00013044047,0.92412883,0.0026475755],"study_design_scores_gemma":[0.00043425243,0.00013395018,0.3380051,0.000019751249,0.000002638373,0.0000019842337,0.0006511794,0.00040395203,0.00006694843,0.007218649,0.65291107,0.00015049882],"about_ca_topic_score_codex":0.0000075330213,"about_ca_topic_score_gemma":0.00003178519,"teacher_disagreement_score":0.81384057,"about_ca_system_score_codex":0.00040710813,"about_ca_system_score_gemma":0.00019196371,"threshold_uncertainty_score":0.99979824},"labels":[],"label_agreement":null},{"id":"W6904516503","doi":"10.1371/journal.pone.0240631.s007","title":"Results from OLS and QR models for the number of physician consultations for children under 1 year old in the 1&lt;sup&gt;st&lt;/sup&gt; Cycle of PMAQ (missing values imputed), Brazil.","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"MEDLINE; Health care; Quarter (Canadian coin); Health services","score_opus":0.1482042693624554,"score_gpt":0.36095136140218476,"score_spread":0.21274709203972936,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6904516503","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3016828,0.0006560713,0.0032862797,0.02070345,0.00009668068,0.0052771177,0.6660831,0.000053085972,0.0021614148],"genre_scores_gemma":[0.99227047,0.000007962236,0.0016164706,0.0012405294,0.00013931819,0.00014429004,0.0045228787,0.000014155145,0.000043919255],"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9976152,0.00017451285,0.00070405705,0.0004479594,0.0008498173,0.0002084532],"domain_scores_gemma":[0.9930729,0.005599136,0.00034935062,0.0004839615,0.00043595253,0.00005869204],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0009389014,0.00015392138,0.00029865166,0.00003746468,0.00020305692,0.00016650774,0.00074338523,0.000077365155,0.001070976],"category_scores_gemma":[0.0054094875,0.00009103267,0.00015944222,0.0004066997,0.000058758746,0.00027067636,0.00012229166,0.000110262205,0.000032938515],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005480171,0.00044652028,0.00084958633,0.00017017458,0.00025943032,7.566397e-7,0.023155337,0.09719457,0.0016291199,0.0030588084,0.8442736,0.028414031],"study_design_scores_gemma":[0.007682731,0.00036888703,0.08106698,0.0018781007,0.00018765978,0.0000016297987,0.007645037,0.5351503,0.004183704,0.34476528,0.016346222,0.0007234436],"about_ca_topic_score_codex":0.000032889857,"about_ca_topic_score_gemma":0.000023254266,"teacher_disagreement_score":0.8279274,"about_ca_system_score_codex":0.000016211297,"about_ca_system_score_gemma":0.00007779775,"threshold_uncertainty_score":0.99984217},"labels":[],"label_agreement":null},{"id":"W6904555035","doi":"10.1371/journal.pone.0292021.s003","title":"Newcastle-Ottawa Scale quality assessment of the studies.","year":2023,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Scale (ratio); Quality (philosophy); Quality assessment; Measure (data warehouse); Data quality","score_opus":0.7077092964839,"score_gpt":0.5473400207967551,"score_spread":0.16036927568714487,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6904555035","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5542492,0.0011385687,0.000031825395,0.027082622,0.003970204,0.003875947,0.29119605,0.00048397388,0.11797158],"genre_scores_gemma":[0.99355197,0.0000064716805,0.0001224852,0.00039886904,0.000104624094,0.00013665363,0.000980596,0.0000084277535,0.0046899095],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9957891,0.00044524024,0.0007100377,0.00034760425,0.0024974365,0.00021054786],"domain_scores_gemma":[0.99622154,0.0016812152,0.0003384506,0.0010025547,0.0007023512,0.000053903288],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0035798077,0.000105139385,0.0002649137,0.000066824425,0.00018160818,0.00006795578,0.001057215,0.00004638313,0.055273734],"category_scores_gemma":[0.014538677,0.00005594021,0.00021296882,0.0012409458,0.000033183987,0.00011400782,0.00057318347,0.00013101411,0.002296046],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000014790216,0.000039747734,0.015673349,0.00005338784,0.00001810703,6.3165294e-7,0.00025809614,0.00034634338,0.00020548698,0.000067130786,0.9774169,0.0059193564],"study_design_scores_gemma":[0.00020474191,0.000034408255,0.7160323,0.0004183686,0.000005800208,3.489394e-7,0.0017436713,0.00046992316,0.0010485217,0.0054462813,0.2744762,0.00011938155],"about_ca_topic_score_codex":0.000009139803,"about_ca_topic_score_gemma":0.00008262871,"teacher_disagreement_score":0.7029407,"about_ca_system_score_codex":0.000062966596,"about_ca_system_score_gemma":0.00010990242,"threshold_uncertainty_score":0.9984808},"labels":[],"label_agreement":null},{"id":"W6904677782","doi":"10.1371/journal.pone.0296026.s003","title":"Quality assessment for cross-sectional studies (Modified Newcastle-Ottawa Quality Assessment Scale criteria)*.","year":2023,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Quality assessment; Quality (philosophy); Scale (ratio); Risk assessment; Impact assessment","score_opus":0.7619917823556953,"score_gpt":0.6080136156474778,"score_spread":0.1539781667082175,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6904677782","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.41996953,0.00033651406,0.0026433885,0.0063576344,0.005672262,0.0053887684,0.5342396,0.0010485261,0.024343785],"genre_scores_gemma":[0.9568737,0.000010034304,0.0034737505,0.00078778056,0.0008528454,0.002864502,0.027641902,0.000041439824,0.0074540367],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99068034,0.0010179576,0.0021424145,0.0013388036,0.0041284207,0.0006920525],"domain_scores_gemma":[0.9886842,0.0064702416,0.00071858824,0.0012532037,0.0026431917,0.00023058402],"candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.014527918,0.00036239842,0.0007338506,0.00023321869,0.00089158036,0.0010562143,0.0011736925,0.00018928024,0.050303385],"category_scores_gemma":[0.018349305,0.00028258853,0.0005548906,0.0009421317,0.00007528729,0.00057983014,0.00060021685,0.0002997887,0.0018657981],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000070088645,0.0003375481,0.09901354,0.00043125803,0.00013981927,0.0000037549223,0.00032614754,0.0013246711,0.000926145,0.00083258434,0.89077866,0.0058157966],"study_design_scores_gemma":[0.0008813592,0.00013305765,0.8736428,0.00015062536,0.0000070416704,9.695468e-7,0.0008593238,0.001695454,0.0002164111,0.021145161,0.100896075,0.0003716893],"about_ca_topic_score_codex":0.000030256771,"about_ca_topic_score_gemma":0.00012661263,"teacher_disagreement_score":0.7898826,"about_ca_system_score_codex":0.00047485402,"about_ca_system_score_gemma":0.0003532899,"threshold_uncertainty_score":0.9999808},"labels":[],"label_agreement":null},{"id":"W6920517678","doi":"10.6084/m9.figshare.12072534.v1","title":"Additional file 4 of Sedentary behavior patterns and adiposity in children: a study based on compositional data analysis","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Children's Hospital of Eastern Ontario","funders":"","keywords":"Table (database); Compositional data; Physical activity; Sedentary behavior; Sedentary lifestyle; Data file","score_opus":0.21652186579229066,"score_gpt":0.3554182507638054,"score_spread":0.13889638497151474,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6920517678","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.035370346,0.0000035516064,0.0000030031601,0.00014623922,0.0000040297023,0.00036263926,0.96403396,0.00000718003,0.00006902498],"genre_scores_gemma":[0.42110962,2.8460859e-8,0.00012208793,0.00033507033,0.000019757468,0.00013378407,0.5782725,0.0000018195972,0.000005291174],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9973815,0.00019597397,0.00038762408,0.0005466415,0.0013909882,0.000097235046],"domain_scores_gemma":[0.9974363,0.0016901215,0.00015040506,0.00051974505,0.0001138551,0.00008958279],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00022347386,0.00009990648,0.00022792666,0.00015622367,0.00006335331,0.00006551029,0.0006843503,0.000029985475,0.98675007],"category_scores_gemma":[0.0022579813,0.0000824278,0.000073031806,0.0006404683,0.000010071615,0.00015204356,0.00029682694,0.000117842035,0.00031873782],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00001935814,0.0005593097,0.20479894,0.0000048008037,0.0000464984,0.0000054957227,0.000029968616,0.00027207294,0.0000010923917,5.903264e-8,0.7937841,0.0004782917],"study_design_scores_gemma":[0.00029209498,0.00018824816,0.9856691,0.0001363259,0.000053163792,3.636579e-7,0.000074686766,0.008511269,0.000005890246,0.0000041556314,0.004979851,0.00008490222],"about_ca_topic_score_codex":0.000029752437,"about_ca_topic_score_gemma":0.0001571515,"teacher_disagreement_score":0.9864313,"about_ca_system_score_codex":0.000018387089,"about_ca_system_score_gemma":0.000045930836,"threshold_uncertainty_score":0.40968361},"labels":[],"label_agreement":null},{"id":"W6920518269","doi":"10.6084/m9.figshare.12233456.v1","title":"NOS","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Scale (ratio); Field (mathematics); Process (computing)","score_opus":0.5584769089976009,"score_gpt":0.40391553303996286,"score_spread":0.15456137595763808,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6920518269","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011571144,0.00086269685,0.00023050002,0.09447224,0.00067485886,0.0019049539,0.3900353,0.0006240234,0.49962428],"genre_scores_gemma":[0.9841181,6.7430597e-7,0.00026530505,0.009994936,0.00039132722,0.000060365903,0.0042890906,0.000008153838,0.0008720193],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9982058,0.00006000025,0.00024510923,0.00027568324,0.0010942183,0.00011917998],"domain_scores_gemma":[0.9990261,0.00032145125,0.000067589215,0.000267979,0.0001858871,0.0001309778],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00025531813,0.00006129949,0.000100324476,0.000018960736,0.000058541762,0.0001460236,0.0006046065,0.000030054343,0.68612653],"category_scores_gemma":[0.017098676,0.000041870346,0.000076153294,0.0003180307,0.0000033491615,0.00012891178,0.00013032988,0.000079056146,0.08712512],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000029298178,0.000008563244,0.00023921505,0.000006163445,0.0000017019398,0.0000027304022,0.00012049738,0.000046363257,0.000084901745,0.000018571931,0.99030215,0.009166219],"study_design_scores_gemma":[0.00008520784,0.000037376238,0.004114649,0.000046399015,8.5778163e-7,2.6057074e-7,0.00007786632,0.0006778191,0.0007813898,0.0009414135,0.9931691,0.00006764705],"about_ca_topic_score_codex":5.569714e-7,"about_ca_topic_score_gemma":0.0000010910662,"teacher_disagreement_score":0.972547,"about_ca_system_score_codex":0.000009050245,"about_ca_system_score_gemma":0.000031689004,"threshold_uncertainty_score":0.9911807},"labels":[],"label_agreement":null},{"id":"W6920646253","doi":"10.6084/m9.figshare.12476702.v1","title":"Additional file 1 of Using multiple agreement methods for continuous repeated measures data: a tutorial for practitioners","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Measure (data warehouse); Data file; Data collection; Term (time)","score_opus":0.6110904157845687,"score_gpt":0.4802538677778531,"score_spread":0.13083654800671557,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6920646253","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000046099476,0.00003099837,0.009500506,0.000330173,0.00018427489,0.0013028845,0.98837024,0.000026268732,0.00025005656],"genre_scores_gemma":[0.00025189997,2.428838e-7,0.19273658,0.00026789226,0.00063038705,0.0012723894,0.8046659,0.000011508184,0.00016316683],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99697274,0.0002695734,0.0007846205,0.00067457295,0.001069267,0.00022921393],"domain_scores_gemma":[0.976388,0.02017554,0.0007664432,0.0006909906,0.0018369481,0.0001421],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0013488733,0.00015339255,0.00036152906,0.00006211853,0.00018865541,0.00013671965,0.00083606184,0.00008822398,0.9149437],"category_scores_gemma":[0.283754,0.00012494957,0.00020213854,0.00030912863,0.000019294743,0.00042027226,0.0002315963,0.000087115404,0.00019293901],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017147289,0.00006486032,0.0000024123863,0.000059263875,0.00004783434,3.8678473e-7,0.000075157186,0.00020985604,0.0005260927,0.000003221664,0.98350763,0.015331811],"study_design_scores_gemma":[0.00062935957,0.00014773168,0.000034789457,0.00034664114,0.000021318509,6.9135376e-7,0.00024406986,0.042291574,0.0007697892,0.00035574634,0.9550139,0.00014442774],"about_ca_topic_score_codex":0.000005518309,"about_ca_topic_score_gemma":0.000009090574,"teacher_disagreement_score":0.91475075,"about_ca_system_score_codex":0.000052779586,"about_ca_system_score_gemma":0.0002837599,"threshold_uncertainty_score":0.72227925},"labels":[],"label_agreement":null},{"id":"W6920799232","doi":"10.6084/m9.figshare.13719529","title":"Additional file 1 of The Matthews correlation coefficient (MCC) is more reliable than balanced accuracy, bookmaker informedness, and markedness in two-class confusion matrix evaluation","year":2021,"lang":"en","type":"article","venue":"Open MIND","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University Health Network","funders":"","keywords":"Randomness; Confusion; Matrix (chemical analysis); Markedness; Confusion matrix; Correlation coefficient","score_opus":0.09709000465052663,"score_gpt":0.39545205506978853,"score_spread":0.2983620504192619,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6920799232","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.87967753,0.00009522809,0.0001328298,0.003942598,0.0009066234,0.0033511154,0.068359904,0.000003472788,0.043530673],"genre_scores_gemma":[0.9052484,0.000043102373,0.009117014,0.001332079,0.000121596946,0.0009767066,0.018276108,0.000034359025,0.06485064],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9960695,0.00031098706,0.0008091967,0.000469989,0.0021506203,0.00018969121],"domain_scores_gemma":[0.9958243,0.0021313583,0.00052755914,0.00061574864,0.0008456361,0.000055435976],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0034415675,0.00014102402,0.0002689781,0.00008189505,0.00019193515,0.0003181511,0.00057766197,0.0000752354,0.7069971],"category_scores_gemma":[0.006270701,0.00008592209,0.000073533934,0.0006897772,0.00013689158,0.0004954182,0.0005072083,0.00015243923,0.0003360993],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001225433,0.00020736147,0.008851609,0.000012204145,0.0000116928695,0.0000018060946,0.0015191921,0.0041233194,0.00034633052,0.000023345388,0.93179834,0.05298226],"study_design_scores_gemma":[0.0015660292,0.000046627698,0.18148805,0.00069479574,0.000027682096,0.000009282884,0.0042648646,0.1137054,0.0015541221,0.0015207495,0.69489163,0.00023076918],"about_ca_topic_score_codex":0.00006190207,"about_ca_topic_score_gemma":0.00055616105,"teacher_disagreement_score":0.706661,"about_ca_system_score_codex":0.000113259084,"about_ca_system_score_gemma":0.0006176269,"threshold_uncertainty_score":0.7507068},"labels":[],"label_agreement":null},{"id":"W6920920990","doi":"10.6084/m9.figshare.13719529.v1","title":"Additional file 1 of The Matthews correlation coefficient (MCC) is more reliable than balanced accuracy, bookmaker informedness, and markedness in two-class confusion matrix evaluation","year":2021,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University Health Network","funders":"","keywords":"Randomness; Confusion; Matrix (chemical analysis); Markedness; Confusion matrix; Correlation coefficient","score_opus":0.08969401580553116,"score_gpt":0.3639828464937436,"score_spread":0.2742888306882124,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6920920990","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010382667,0.0000772776,0.0000043145114,0.0007551431,0.00018981959,0.0008035829,0.9828204,0.00001289782,0.004953918],"genre_scores_gemma":[0.17821991,0.000010947435,0.0004916047,0.0015658168,0.000114901,0.0017157373,0.8063076,0.000026415266,0.01154702],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99601984,0.0002357715,0.00070837303,0.00041416375,0.0024183593,0.0002035013],"domain_scores_gemma":[0.9945693,0.002999922,0.00052144536,0.0005779379,0.0012740191,0.000057381596],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0009779085,0.00015207921,0.00023558547,0.00009251269,0.00017358779,0.00015042679,0.00036392504,0.00009069016,0.9654056],"category_scores_gemma":[0.024612775,0.00009472932,0.00010300125,0.0007439293,0.00003795327,0.00033763895,0.00030554557,0.00017997577,0.0007457796],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000026869728,0.000063036394,0.0008500229,0.00003674929,0.000004536703,9.895222e-7,0.00032643328,0.0022793375,0.0000364419,0.0000073188803,0.99254,0.0038282692],"study_design_scores_gemma":[0.00072290964,0.000024479117,0.16883472,0.00255127,0.000010009937,0.000006250422,0.0010178104,0.11974056,0.0003802157,0.0008844217,0.7056516,0.00017579603],"about_ca_topic_score_codex":0.0000151113045,"about_ca_topic_score_gemma":0.00016954425,"teacher_disagreement_score":0.96465975,"about_ca_system_score_codex":0.0001201732,"about_ca_system_score_gemma":0.00049991906,"threshold_uncertainty_score":0.9836033},"labels":[],"label_agreement":null},{"id":"W6920949177","doi":"10.6084/m9.figshare.14913279.v1","title":"Additional file 6 of Validity of observational evidence on putative risk and protective factors: appraisal of 3744 meta-analyses on 57 topics","year":2021,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Observational study; Kappa; Measure (data warehouse); Sample (material); Risk assessment","score_opus":0.8339611891746483,"score_gpt":0.5000389208144426,"score_spread":0.3339222683602058,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6920949177","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0053933607,0.0000446996,0.000019050784,0.000103749284,0.000019412799,0.00021201567,0.99371403,0.000003714594,0.0004899673],"genre_scores_gemma":[0.515487,0.000005301076,0.004260791,0.0001470854,0.00014063131,0.0009364029,0.4781378,0.000013552307,0.00087141834],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99643,0.00062310416,0.00059106044,0.00036617162,0.0019025719,0.000087138134],"domain_scores_gemma":[0.9617224,0.03526523,0.0008122368,0.0003369636,0.0018193598,0.000043830863],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005311145,0.00012452563,0.00046216408,0.00007713989,0.0000625999,0.000026500671,0.0002502706,0.00005959385,0.92306125],"category_scores_gemma":[0.19485171,0.00008090948,0.0003190223,0.0004093111,0.000045004566,0.00019990004,0.000100360325,0.00013925078,0.00006960235],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000033635213,0.00021028983,0.00051622343,0.00006773269,0.00033134964,0.0000010391387,0.00023311809,0.0006408944,0.00013005208,0.000041229036,0.9968547,0.0009396902],"study_design_scores_gemma":[0.0002833965,0.0011526375,0.75759715,0.002597252,0.00041013412,9.1187775e-7,0.001408659,0.00048714728,0.094768904,0.013661228,0.12731741,0.00031517757],"about_ca_topic_score_codex":0.000010313452,"about_ca_topic_score_gemma":0.000015410711,"teacher_disagreement_score":0.92299163,"about_ca_system_score_codex":0.000022870363,"about_ca_system_score_gemma":0.0001985065,"threshold_uncertainty_score":0.8119304},"labels":[],"label_agreement":null},{"id":"W6923051379","doi":"10.1371/journal.pntd.0010702.s003","title":"Risk of Bias and quality assessments.","year":2022,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Table (database); Quality (philosophy); Data quality; Data collection; Risk assessment","score_opus":0.5954873802289661,"score_gpt":0.4769345406090012,"score_spread":0.11855283961996488,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6923051379","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.53935164,0.00054461265,0.000020507929,0.00040103393,0.00022316417,0.0005496136,0.44029006,0.000037499172,0.018581891],"genre_scores_gemma":[0.99698913,0.0000027795536,0.00014323501,0.0001323594,0.000016874235,0.00008040058,0.0019489067,0.0000032682249,0.0006830282],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9964647,0.00086259504,0.00045549672,0.00027941738,0.0018298517,0.00010793632],"domain_scores_gemma":[0.9974369,0.0015072523,0.00037789097,0.00044309208,0.00018470225,0.000050204973],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0040115826,0.000065138265,0.00016884568,0.00006470906,0.00020923834,0.00005865926,0.0004448092,0.000018007519,0.33612764],"category_scores_gemma":[0.011180734,0.000048842714,0.00007646288,0.00032314225,0.000009972722,0.000099190554,0.00047492614,0.00014353286,0.00021390941],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000043359716,0.00027021623,0.18544623,0.000065240005,0.000042210668,0.0000024146623,0.0004714522,0.0007012033,0.00022749012,0.0002777625,0.75429225,0.058160175],"study_design_scores_gemma":[0.00031888718,0.00017709761,0.49345,0.00004720167,0.000007334551,8.3318287e-7,0.0012039959,0.0003837041,0.00051720487,0.0062444434,0.49751568,0.00013363172],"about_ca_topic_score_codex":0.000034353205,"about_ca_topic_score_gemma":0.000012227295,"teacher_disagreement_score":0.45763752,"about_ca_system_score_codex":0.000029579985,"about_ca_system_score_gemma":0.00004629944,"threshold_uncertainty_score":0.9971485},"labels":[],"label_agreement":null},{"id":"W6923380391","doi":"10.1371/journal.pone.0299125.s008","title":"Author’s judgements about study quality using the adapted Ottawa-Newcastle risk of bias assessment tool.","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Quality (philosophy); Risk assessment; Quality assessment; Response bias; Data collection; Risk perception","score_opus":0.6038026136552229,"score_gpt":0.5165250097033885,"score_spread":0.08727760395183437,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6923380391","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8320861,0.0012482705,0.0008433603,0.0012179693,0.002644293,0.004171211,0.14846186,0.00022466239,0.0091022635],"genre_scores_gemma":[0.9972012,0.0000018884937,0.00045225216,0.00007751813,0.00015236188,0.00012126637,0.00073748385,0.000015503118,0.0012405005],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99207175,0.0017850715,0.0013730999,0.0006988526,0.003763379,0.0003078436],"domain_scores_gemma":[0.99459255,0.002808447,0.0005107743,0.0012802068,0.0007264308,0.000081618135],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.010806237,0.00022777983,0.00038179295,0.00015816705,0.00032440867,0.00064069725,0.0011422231,0.00007542975,0.10456625],"category_scores_gemma":[0.01340514,0.0001283826,0.00027697394,0.0010644087,0.000026401874,0.00032118277,0.0004408298,0.00035805555,0.001014778],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000029638042,0.0010169972,0.06568913,0.00020528765,0.00041155558,0.000020681073,0.0027509334,0.0037012936,0.0002946745,0.00026479573,0.88595974,0.039655246],"study_design_scores_gemma":[0.0005841135,0.00034618107,0.5492768,0.0012094519,0.0001389522,0.0000016322336,0.0064059985,0.026744485,0.00034524983,0.0027761078,0.41173953,0.00043143443],"about_ca_topic_score_codex":0.00018614465,"about_ca_topic_score_gemma":0.00016485422,"teacher_disagreement_score":0.4835877,"about_ca_system_score_codex":0.00014447437,"about_ca_system_score_gemma":0.0002500163,"threshold_uncertainty_score":0.9997631},"labels":[],"label_agreement":null},{"id":"W6923411768","doi":"10.1371/journal.pone.0293382.s003","title":"Quality appraisal of longitudinal studies using Newcastle Ottawa Scale.","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Quality (philosophy); Longitudinal study; Government (linguistics); Critical appraisal; Data collection","score_opus":0.6543883855328971,"score_gpt":0.5442819466279222,"score_spread":0.11010643890497496,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6923411768","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5961129,0.07484098,0.0009996052,0.0031289826,0.0067196484,0.0019332683,0.27477786,0.00058246806,0.040904276],"genre_scores_gemma":[0.9977767,0.000008872743,0.00068112416,0.00004362407,0.0002661425,0.000027802402,0.00069354486,0.000011271099,0.00049093785],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99652284,0.00022171916,0.00079202675,0.00049016933,0.0017632054,0.0002100184],"domain_scores_gemma":[0.9961694,0.0024109776,0.00017521049,0.00051169866,0.00066063064,0.00007206966],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0022420455,0.00013927116,0.00033449504,0.00012861504,0.00010072851,0.00018513268,0.00047951957,0.00005916938,0.064312786],"category_scores_gemma":[0.013694063,0.00009477042,0.00020785807,0.0006597961,0.00003904009,0.000368634,0.00023968257,0.00013635437,0.0017189115],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000019715359,0.0001252019,0.013437784,0.0009557956,0.00011494865,0.000019169702,0.0011875562,0.00081987516,0.0012315445,0.0002635919,0.9466826,0.035142247],"study_design_scores_gemma":[0.00040337312,0.00019665588,0.08268202,0.005290664,0.000050774826,0.000017742737,0.0036534232,0.0055252747,0.0062910914,0.010577453,0.88466597,0.0006455635],"about_ca_topic_score_codex":0.00001393505,"about_ca_topic_score_gemma":0.000050339928,"teacher_disagreement_score":0.40166378,"about_ca_system_score_codex":0.000057563804,"about_ca_system_score_gemma":0.000099309604,"threshold_uncertainty_score":0.99905837},"labels":[],"label_agreement":null},{"id":"W6929742458","doi":"10.5061/dryad.qbzkh18p5","title":"Social interactions generate complex selection patterns in virtual worlds","year":2024,"lang":"en","type":"dataset","venue":"DRYAD","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université du Québec à Montréal","funders":"Mitacs; European Commission","keywords":"Selection (genetic algorithm); Natural selection; Natural (archaeology); Social relation; Variation (astronomy); Inclusive fitness; Mechanism (biology)","score_opus":0.23797326387077092,"score_gpt":0.4434277590926073,"score_spread":0.20545449522183637,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6929742458","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0052254796,0.000031848725,0.00022475707,0.0021446885,0.0032934812,0.00033651595,0.9883021,0.000041832245,0.000399293],"genre_scores_gemma":[0.025123073,0.00006400363,0.000035331483,0.0008655769,0.0020868916,0.00012361964,0.9681665,0.000022992172,0.0035120568],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99554443,0.0004925255,0.0010095078,0.00083455595,0.0017844919,0.0003345098],"domain_scores_gemma":[0.998704,0.00033607284,0.00025023337,0.00045341896,0.00018414248,0.000072142124],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0025185815,0.00029684667,0.0004522154,0.00074177835,0.00022374673,0.00069618784,0.00081806397,0.00019876346,0.005860622],"category_scores_gemma":[0.00038993085,0.00023826561,0.00022964235,0.0010932784,0.00006559087,0.0001973019,0.00025520404,0.0009243719,0.01039685],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000017776574,0.00014595297,0.00040101682,0.000021018155,0.000030408013,0.000011794996,0.0001208091,0.0001655835,0.00007710289,0.00003976814,0.99321437,0.00575441],"study_design_scores_gemma":[0.00015049825,0.000055682067,0.002265046,0.000066766785,0.000032603268,0.0000027812066,0.00020900763,0.0006754784,0.000022521892,0.0015277648,0.99474305,0.000248786],"about_ca_topic_score_codex":0.00043555986,"about_ca_topic_score_gemma":0.01702441,"teacher_disagreement_score":0.020135649,"about_ca_system_score_codex":0.00034211972,"about_ca_system_score_gemma":0.00012334985,"threshold_uncertainty_score":0.99504817},"labels":[],"label_agreement":null},{"id":"W6929826709","doi":"10.5281/zenodo.12031831","title":"youth and crime pdf","year":2024,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Commit; Law enforcement; Psychological intervention; Suicide prevention; Poison control; Injury prevention; Enforcement; Human factors and ergonomics","score_opus":0.15798201310407178,"score_gpt":0.32661207697116956,"score_spread":0.16863006386709778,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6929826709","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000086624204,0.00083309296,0.0007028386,0.0006063214,0.00034940342,0.00041626787,0.00035244285,0.0005625057,0.99609053],"genre_scores_gemma":[0.0076063937,0.00025474295,0.00017373773,0.00018034,0.00039596405,2.6629266e-8,0.00036487298,0.0041760323,0.9868479],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9964939,0.00047530883,0.00039600968,0.0008411571,0.0014943851,0.00029924192],"domain_scores_gemma":[0.9984442,0.00003852462,0.00016206899,0.00081576087,0.00035020363,0.0001892508],"candidate_categories":["scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0030763906,0.00020746977,0.000256061,0.00054793904,0.00060498633,0.0023859693,0.0014443657,0.00014084192,0.22966842],"category_scores_gemma":[0.0019117219,0.00016731938,0.0000802744,0.0006022329,0.00024130126,0.00010087867,0.0015677137,0.0003456815,0.269955],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000008834588,0.00004086614,8.4060053e-7,0.000058528334,0.00003498736,0.0000065931463,0.0007579336,8.935005e-7,0.000063584135,0.0011925813,0.9610316,0.03680272],"study_design_scores_gemma":[0.00015519913,0.000101936836,0.000037556183,0.00012222206,0.000029243542,0.000014237769,0.00070522376,0.000025311967,0.000015916092,0.0024911936,0.99611646,0.00018549137],"about_ca_topic_score_codex":0.000023669756,"about_ca_topic_score_gemma":4.855964e-7,"teacher_disagreement_score":0.040286586,"about_ca_system_score_codex":0.00007204194,"about_ca_system_score_gemma":0.000003681115,"threshold_uncertainty_score":0.99864966},"labels":[],"label_agreement":null},{"id":"W6930161627","doi":"10.5203/0096509","title":"Environment Canada - Water Quality Monitoring Stations - East Side Rivers and Lake Outlet Water Quality","year":2017,"lang":"en","type":"dataset","venue":"UMANCEOS","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Hydrology (agriculture); Water quality; River delta; Drainage basin; Shelf ice; Bank","score_opus":0.21747342379488832,"score_gpt":0.39985270430275854,"score_spread":0.18237928050787022,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930161627","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.054520912,0.00008184596,0.000049878065,0.0015874545,0.0020427138,0.000587721,0.9409027,0.0000146783095,0.0002121115],"genre_scores_gemma":[0.06752091,0.00039076098,0.0003188322,0.00039693242,0.00090623193,0.00018623634,0.92420435,0.000033032782,0.0060427],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9926037,0.0006732819,0.0013249,0.0011382384,0.0036062172,0.0006536408],"domain_scores_gemma":[0.9963197,0.00034881147,0.00047703454,0.0024238252,0.00016839114,0.0002622196],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.006005512,0.0004397697,0.0007223951,0.000103214305,0.000943501,0.0005595484,0.0014066957,0.00019988685,0.0025143675],"category_scores_gemma":[0.0006475071,0.0002599966,0.00013528361,0.000031652322,0.00033438526,0.0003461261,0.0006260608,0.00047041336,0.0007832465],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000032755855,0.0000860235,0.005928386,0.000091705,0.000071401446,0.000016617541,0.0004908633,0.00021062847,0.00017245048,0.0000038820485,0.9914321,0.0014631946],"study_design_scores_gemma":[0.00036518896,0.000028475908,0.019725865,0.000051190615,0.00004190227,7.3117405e-7,0.00046115476,0.00000259113,0.00048017563,0.0014990671,0.97690403,0.00043961612],"about_ca_topic_score_codex":0.2438763,"about_ca_topic_score_gemma":0.5662309,"teacher_disagreement_score":0.3223546,"about_ca_system_score_codex":0.0003773745,"about_ca_system_score_gemma":0.00017701842,"threshold_uncertainty_score":0.99999475},"labels":[],"label_agreement":null},{"id":"W6930190331","doi":"10.5281/zenodo.10910123","title":"Capture of the noise from a 120 m long roll-on/roll-off cargo ship","year":2024,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université du Québec à Montréal; Université du Québec à Rimouski","funders":"","keywords":"Noise (video); Hydrophone; Sound (geography); Underwater; Sensitivity (control systems); Software; Ambient noise level; Background noise","score_opus":0.10569522803381616,"score_gpt":0.3096380628105987,"score_spread":0.20394283477678254,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930190331","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011233329,0.0013322252,0.00028672267,0.0021117947,0.00088251196,0.0009700957,0.0014850305,0.0004210536,0.99138725],"genre_scores_gemma":[0.18462633,0.0002977692,0.0001255362,0.0007393192,0.0010125883,2.4570812e-7,0.0005736039,0.007393903,0.8052307],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9939988,0.0010817453,0.0006437415,0.0009798575,0.002928271,0.00036755938],"domain_scores_gemma":[0.9968147,0.00014006477,0.0004163,0.0018937667,0.0005645616,0.0001705504],"candidate_categories":["scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0027976276,0.000311254,0.00041624883,0.0004335457,0.0006194805,0.0011389354,0.0038300816,0.0002419268,0.08927073],"category_scores_gemma":[0.0034032161,0.00020206139,0.0002661148,0.0011653572,0.00033049769,0.00008794551,0.0018585881,0.00069411454,0.04008286],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004359499,0.00011720135,0.000020428222,0.000057356876,0.00009609564,0.00001029069,0.00080223306,0.00006432377,0.0003034113,0.0006631137,0.98406714,0.013754812],"study_design_scores_gemma":[0.00029291763,0.000101220794,0.00046957063,0.0003374094,0.000049233342,0.00000503506,0.00034420207,0.000076774115,0.00017989751,0.0025245794,0.99540883,0.0002103285],"about_ca_topic_score_codex":0.0001404138,"about_ca_topic_score_gemma":0.000012785234,"teacher_disagreement_score":0.18615653,"about_ca_system_score_codex":0.00015193028,"about_ca_system_score_gemma":0.000014856958,"threshold_uncertainty_score":0.99989796},"labels":[],"label_agreement":null},{"id":"W6930262978","doi":"10.5281/zenodo.12253199","title":"Prowirl","year":2024,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Nucleofection; Limiting; Field (mathematics); Noise (video); Work (physics); Subpoena","score_opus":0.1706779805591923,"score_gpt":0.3463615684099742,"score_spread":0.17568358785078186,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930262978","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000017494527,0.00063440297,0.0017117908,0.0014403497,0.00056799094,0.00064211775,0.0003160302,0.0012674775,0.99340236],"genre_scores_gemma":[0.0016154314,0.00013057941,0.00034526753,0.00017750726,0.0005876392,6.034059e-8,0.00040507427,0.007638153,0.9891003],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9955273,0.0005746967,0.0004861857,0.00094389083,0.002103647,0.000364229],"domain_scores_gemma":[0.9978498,0.000040017097,0.00020731417,0.0012089146,0.0005117468,0.00018215623],"candidate_categories":["scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.003954298,0.00022989092,0.00027788623,0.00070669485,0.00067146064,0.0021850734,0.002400903,0.0001678229,0.25631595],"category_scores_gemma":[0.0025805482,0.0001831465,0.00013750026,0.0010011303,0.0002060064,0.00010779054,0.0017081032,0.00042758757,0.34342158],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00000799586,0.000066401524,4.21994e-7,0.00006926711,0.000040580682,0.000010672856,0.00020940835,0.0000029321131,0.00009572638,0.0034941786,0.91193295,0.08406945],"study_design_scores_gemma":[0.00013724888,0.00010498685,0.000012593631,0.00013691341,0.00001878927,0.000018614648,0.00017969876,0.000036530626,0.000025288711,0.003523788,0.99559945,0.00020607795],"about_ca_topic_score_codex":0.000016630816,"about_ca_topic_score_gemma":0.0000010219422,"teacher_disagreement_score":0.08710562,"about_ca_system_score_codex":0.00013158341,"about_ca_system_score_gemma":0.000005225276,"threshold_uncertainty_score":0.99885076},"labels":[],"label_agreement":null},{"id":"W6930279845","doi":"10.5281/zenodo.10892753","title":"Application of Market Timing Theory: Evidence from Canadian Firms","year":2024,"lang":"en","type":"article","venue":"DOAJ (DOAJ: Directory of Open Access Journals)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Market timing; Capital structure; Corporate finance; Leverage (statistics); Equity (law); Valuation (finance); Capital market; Stock market; Volatility (finance); External financing","score_opus":0.5118674764462113,"score_gpt":0.5977812231907639,"score_spread":0.08591374674455254,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930279845","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8852176,0.06654814,0.021314835,0.0027542214,0.0020802312,0.0014558065,0.0002613127,0.00005165773,0.0203162],"genre_scores_gemma":[0.9954047,0.0029788355,0.00044439046,0.00022046146,0.00015185603,0.00004128538,0.000004839397,0.000020084324,0.00073352986],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9942709,0.00080861547,0.0014334251,0.0007282095,0.0024165208,0.00034227976],"domain_scores_gemma":[0.992978,0.004358955,0.000661616,0.0010227899,0.0006213404,0.00035734565],"candidate_categories":["scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.016490875,0.0002367311,0.0006147265,0.0009935598,0.00023189535,0.002365758,0.005086782,0.00010824055,0.03912612],"category_scores_gemma":[0.0044017904,0.00017489148,0.00023875272,0.0018554021,0.00017935797,0.0027551346,0.00057594036,0.00033092007,0.00013375875],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015570121,0.000099753066,0.52053887,0.0000845003,0.0001770077,0.000019420664,0.000654002,0.00055845326,0.04766242,0.0005321024,0.16347253,0.2660452],"study_design_scores_gemma":[0.00013672693,0.000011793787,0.7016467,0.0015647033,0.00009532858,0.0000025633042,0.00032788713,0.0039379527,0.013652672,0.21677755,0.061495632,0.00035052228],"about_ca_topic_score_codex":0.05646765,"about_ca_topic_score_gemma":0.009600033,"teacher_disagreement_score":0.2656947,"about_ca_system_score_codex":0.00022171148,"about_ca_system_score_gemma":0.00048336622,"threshold_uncertainty_score":0.99866986},"labels":[],"label_agreement":null},{"id":"W6930359942","doi":"10.5281/zenodo.12042164","title":"5th grade social studies textbook pdf","year":2024,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Download; Vocabulary; Reading (process); Social studies; Class (philosophy); State (computer science)","score_opus":0.28511173278310015,"score_gpt":0.3982049053351869,"score_spread":0.11309317255208673,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930359942","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000028600929,0.0012938862,0.00038991,0.0026136118,0.0006687275,0.0005949802,0.00025158495,0.0008285058,0.9933302],"genre_scores_gemma":[0.0017211676,0.00038474528,0.00012841281,0.00024897276,0.000972931,1.0443887e-7,0.00023680743,0.004226954,0.9920799],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99481034,0.0008162757,0.0005883504,0.0009835038,0.002368931,0.00043259922],"domain_scores_gemma":[0.9980312,0.00007420031,0.00028463983,0.00077112933,0.00069471425,0.0001441038],"candidate_categories":["sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0039724316,0.00029296524,0.0004284391,0.00067130534,0.0014240118,0.0018907287,0.0022756716,0.00017695905,0.27085918],"category_scores_gemma":[0.0025679446,0.00023446797,0.0001887306,0.00090944726,0.00047222225,0.00012421753,0.0020449478,0.0004939336,0.476291],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000010222404,0.000081699065,4.813335e-7,0.000114158116,0.0001503732,0.000014038299,0.0010992625,0.0000020797825,0.00008939453,0.002654183,0.96124995,0.03453415],"study_design_scores_gemma":[0.00019903504,0.00010545861,0.000018765088,0.00012941884,0.000039942206,0.0000117720265,0.0012619772,0.00001416344,0.000020187614,0.0075387424,0.9904098,0.00025073308],"about_ca_topic_score_codex":0.000010391492,"about_ca_topic_score_gemma":0.0000017697583,"teacher_disagreement_score":0.2054318,"about_ca_system_score_codex":0.00022943398,"about_ca_system_score_gemma":0.000006971138,"threshold_uncertainty_score":0.999876},"labels":[],"label_agreement":null},{"id":"W6930369453","doi":"10.5281/zenodo.13969103","title":"Escherichia and Shigella GTDB species identification MASH sketch for ECTYPER","year":2023,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Agency of Canada","funders":"","keywords":"Sketch; Identification (biology); RefSeq; Shigella; Shigella dysenteriae; Escherichia coli; Taxonomy (biology)","score_opus":0.16846007993272166,"score_gpt":0.32975934050073546,"score_spread":0.1612992605680138,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930369453","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0015958667,0.00049180817,0.012283068,0.0029263035,0.0011806624,0.002572143,0.0010748571,0.0023190393,0.97555625],"genre_scores_gemma":[0.014887617,0.0007123726,0.00071644277,0.00015602165,0.000767918,5.6181375e-7,0.000790704,0.008245775,0.9737226],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99648124,0.00042172775,0.0005402247,0.00088188075,0.0013289368,0.00034600755],"domain_scores_gemma":[0.99777114,0.00017949082,0.0003339799,0.00088176766,0.00068408035,0.00014956223],"candidate_categories":["scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00436034,0.00022227608,0.00029013556,0.00046910878,0.0011526695,0.001989039,0.0014616925,0.0001670384,0.03198432],"category_scores_gemma":[0.0044714925,0.00019478308,0.00010030227,0.0007529522,0.0002318448,0.0001562499,0.0008412419,0.00022227694,0.018077822],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000028787335,0.000056989553,0.000014741462,0.00006617037,0.000042507898,0.0000013673156,0.00030013872,0.000007362085,0.0010862014,0.0021945396,0.9500413,0.046159886],"study_design_scores_gemma":[0.0002697389,0.00011716649,0.001253319,0.000058051544,0.000022023447,0.0000034578895,0.00040382857,0.00013629257,0.00011826285,0.001897478,0.99550533,0.0002150248],"about_ca_topic_score_codex":0.000018851024,"about_ca_topic_score_gemma":0.000010044304,"teacher_disagreement_score":0.045944862,"about_ca_system_score_codex":0.00010947867,"about_ca_system_score_gemma":0.000005917382,"threshold_uncertainty_score":0.999047},"labels":[],"label_agreement":null},{"id":"W6930399018","doi":"10.5281/zenodo.12234255","title":"+91-8094774404 «««₪₪»» DEATH SPELL / REVENGE SPELLS CASTER IN U.K, SWEDEN SWITZERLAND,AMERICA, CANADA.","year":2024,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Spell; Wife; Caster; Miracle; MAGIC (telescope); Happening; Mistake","score_opus":0.1377311826636261,"score_gpt":0.3242768853356639,"score_spread":0.1865457026720378,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930399018","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0001551965,0.0007560753,0.00033742658,0.00157868,0.0006137475,0.0008065583,0.00055523287,0.00013128879,0.9950658],"genre_scores_gemma":[0.004645379,0.00023605768,0.00019766354,0.00039739546,0.00046567753,2.3465166e-8,0.00025421116,0.0037464332,0.9900572],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9939611,0.000841065,0.0008266823,0.0012108969,0.00253348,0.00062680297],"domain_scores_gemma":[0.9975958,0.00009362489,0.0003282647,0.0013070796,0.00038857976,0.00028663906],"candidate_categories":["metaepi_narrow","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0028473495,0.0003735786,0.0005144651,0.00076830114,0.00021540614,0.0014164865,0.0025665748,0.00018584204,0.31405354],"category_scores_gemma":[0.0015055961,0.00031457492,0.000121235906,0.0012707801,0.00014381646,0.0001303837,0.0014404686,0.0006861556,0.38323632],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000022956252,0.00008233482,0.000012772063,0.00009447871,0.000043824097,0.00009585504,0.00030125453,0.000020830057,0.00006934515,0.00041395143,0.9829334,0.015908971],"study_design_scores_gemma":[0.00030114356,0.000104862644,0.00009778063,0.00027570254,0.000021532676,0.000025614148,0.00038723391,0.0000679884,0.000023635766,0.0011654866,0.9971769,0.00035212399],"about_ca_topic_score_codex":0.01817763,"about_ca_topic_score_gemma":0.0072629103,"teacher_disagreement_score":0.06918279,"about_ca_system_score_codex":0.0006098674,"about_ca_system_score_gemma":0.00004809062,"threshold_uncertainty_score":0.9999306},"labels":[],"label_agreement":null},{"id":"W6930469099","doi":"10.5281/zenodo.12645768","title":"Download Pdf Shuni","year":2024,"lang":"fr","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Passion; Period (music)","score_opus":0.10406544078213886,"score_gpt":0.3131305898442519,"score_spread":0.20906514906211304,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930469099","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00029033926,0.0019588284,0.0022740047,0.0065837614,0.0017469055,0.0010380798,0.00072549103,0.0006940483,0.9846885],"genre_scores_gemma":[0.0063054,0.00072526373,0.00038312274,0.0004788748,0.0008581555,1.0142022e-7,0.0009421075,0.003981138,0.98632586],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9925036,0.0014874035,0.0009730265,0.0015857148,0.0026338021,0.0008164891],"domain_scores_gemma":[0.9962161,0.00012944249,0.0003255643,0.0016218195,0.0012407454,0.0004663033],"candidate_categories":["metaepi_narrow","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.005686478,0.00048407106,0.0005081797,0.0008041616,0.0020759166,0.005094458,0.003980049,0.00033980122,0.82743126],"category_scores_gemma":[0.00395491,0.00043013925,0.00029979122,0.0017114978,0.0006933213,0.000302985,0.0031727282,0.0009786781,0.875607],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000037815953,0.00020113704,0.0000012472735,0.0001947804,0.00010394017,0.00003520885,0.00081837794,0.000037964237,0.0005549377,0.00764808,0.7104614,0.2799051],"study_design_scores_gemma":[0.00035178728,0.000341468,0.00007297403,0.00039860915,0.00007688492,0.000068197485,0.00066081394,0.00023073914,0.0000941896,0.0053539732,0.99188614,0.00046419594],"about_ca_topic_score_codex":0.00004241395,"about_ca_topic_score_gemma":0.0000023461346,"teacher_disagreement_score":0.28142476,"about_ca_system_score_codex":0.0005249058,"about_ca_system_score_gemma":0.000018077833,"threshold_uncertainty_score":0.99981505},"labels":[],"label_agreement":null},{"id":"W6930685898","doi":"10.5281/zenodo.15753378","title":"Fig. 50 in A revision of Parapterogramma Papp and Pseudopterogramma Papp, with a review of the Parapterogramma genus group of the Pacific and Indomalayan regions (Diptera, Sphaeroceridae, Limosininae)","year":2025,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph","funders":"","keywords":"Dorsum; Genus; Group (periodic table); Scale (ratio); Morphology (biology)","score_opus":0.05291101311911732,"score_gpt":0.282416591565892,"score_spread":0.22950557844677466,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930685898","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.12553918,0.15502088,0.0066243457,0.015615462,0.0012168702,0.03092477,0.0024515735,0.00084221584,0.66176474],"genre_scores_gemma":[0.93870705,0.04227064,0.0014716729,0.0003805192,0.00009340203,0.0000018675318,0.0001809636,0.0016328932,0.015261015],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99499553,0.0014724695,0.0010571156,0.0006963353,0.0014692727,0.00030924418],"domain_scores_gemma":[0.9965722,0.00015336217,0.0011022239,0.0014835832,0.000583801,0.000104807346],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0036586279,0.00029868275,0.00069145474,0.00033407132,0.00039463773,0.0002502811,0.0018041965,0.00014620085,0.0016656487],"category_scores_gemma":[0.0014888678,0.00017227653,0.00016825242,0.00158439,0.0009804798,0.000098477154,0.0013707898,0.00037529678,0.000029355026],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028658562,0.0005783479,0.00455796,0.005738761,0.0001989442,0.0000064841693,0.0011980978,0.0000053726976,0.00017618708,0.0027980744,0.19400944,0.79044574],"study_design_scores_gemma":[0.00068021996,0.00039213346,0.0063776863,0.010113155,0.00007369927,0.000022428067,0.0003530468,0.00001392817,0.000040677205,0.00021742951,0.9815436,0.00017196876],"about_ca_topic_score_codex":0.00012581298,"about_ca_topic_score_gemma":0.000021709246,"teacher_disagreement_score":0.81316787,"about_ca_system_score_codex":0.00003874876,"about_ca_system_score_gemma":0.00001610608,"threshold_uncertainty_score":0.99924695},"labels":[],"label_agreement":null},{"id":"W6930711784","doi":"10.5281/zenodo.12180681","title":"cathay pacific pdf","year":2024,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"China; Aviation; Subsidiary; Asia pacific; Air travel","score_opus":0.12806388315481362,"score_gpt":0.32441335470004007,"score_spread":0.19634947154522644,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930711784","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000019278332,0.0006143761,0.001052878,0.0012986665,0.00070085813,0.00054984645,0.00032335153,0.0008854516,0.9945553],"genre_scores_gemma":[0.0027462973,0.00017273422,0.00015779736,0.000103315244,0.0004980025,5.5220056e-8,0.00046425159,0.005102897,0.99075466],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9950785,0.0006741316,0.0005427811,0.0010438013,0.002241594,0.0004191718],"domain_scores_gemma":[0.99759734,0.000055075747,0.00023748836,0.0013588312,0.0005267046,0.00022454385],"candidate_categories":["scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.004159912,0.000266089,0.00031776648,0.0007688972,0.00077288074,0.002800018,0.002644598,0.00019246872,0.49130452],"category_scores_gemma":[0.0023415652,0.00021629517,0.00016137291,0.0011006597,0.0002473773,0.00011592207,0.0015067733,0.00048607818,0.6722786],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000008866201,0.00007545053,8.6595264e-7,0.000065158456,0.000044957684,0.000015004076,0.00031282537,0.0000033004267,0.00011545262,0.0013989088,0.9518851,0.046074092],"study_design_scores_gemma":[0.0001464453,0.00010679177,0.000016203918,0.00012843362,0.000021693377,0.000022021428,0.00045327182,0.000033829045,0.000024675785,0.003153237,0.9956536,0.00023978548],"about_ca_topic_score_codex":0.000020828962,"about_ca_topic_score_gemma":0.0000011823543,"teacher_disagreement_score":0.18097404,"about_ca_system_score_codex":0.00015991068,"about_ca_system_score_gemma":0.0000064382716,"threshold_uncertainty_score":0.99823517},"labels":[],"label_agreement":null},{"id":"W6930828234","doi":"10.5281/zenodo.1421111","title":"A Transcultural Approach To The Subaltern Identity Of The Metis/ Halfbreeds In Beatrice Culleton Mosionier'S In Search Of April Raintree","year":2018,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Subaltern; Transculturation; Ethnic group; Identity (music); Hegemony; Cultural hegemony; Cultural identity; Cultural studies","score_opus":0.13341797200804287,"score_gpt":0.3248457347361717,"score_spread":0.1914277627281288,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930828234","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9457265,0.000061224135,0.010244446,0.00422103,0.00015871334,0.0013017633,0.000065887085,0.000035872596,0.03818455],"genre_scores_gemma":[0.9989919,0.000016410782,0.00021506102,0.0000795847,0.00005705523,1.293527e-7,0.00001293092,0.00009256273,0.00053433946],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99514014,0.0015177426,0.00065510836,0.0004189363,0.0019704243,0.00029766245],"domain_scores_gemma":[0.9978006,0.00010754855,0.00014198685,0.0008211566,0.0010476937,0.000081001635],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009377305,0.00010955438,0.00021261252,0.0002985879,0.0006490597,0.0004115947,0.0031866676,0.00004860318,0.0016480444],"category_scores_gemma":[0.0026075018,0.0000630195,0.00010040126,0.0030152334,0.00038196333,0.00033858846,0.0011634625,0.00026047492,0.0005590781],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019201403,0.005228402,0.016015947,0.00050838216,0.00022058476,0.000009517982,0.20981118,0.018387558,0.100994706,0.04048407,0.21685418,0.38956532],"study_design_scores_gemma":[0.003619007,0.0013090152,0.38055953,0.0002533282,0.00004560209,0.000048636815,0.02318088,0.008261147,0.02149147,0.004989815,0.55564624,0.0005953496],"about_ca_topic_score_codex":0.00044913564,"about_ca_topic_score_gemma":0.00009900305,"teacher_disagreement_score":0.38896996,"about_ca_system_score_codex":0.00013052713,"about_ca_system_score_gemma":0.000009593504,"threshold_uncertainty_score":0.9992646},"labels":[],"label_agreement":null},{"id":"W6930901674","doi":"10.5281/zenodo.15596103","title":"RovHer: heritability-optimized scores for the functional prioritization of rare missense variants","year":2025,"lang":"en","type":"dataset","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Population Health Research Institute","funders":"","keywords":"Prioritization; Missense mutation; Mutation; Disease; Identification (biology)","score_opus":0.12105340158777163,"score_gpt":0.32464463375748503,"score_spread":0.20359123216971342,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6930901674","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00012618848,0.00040260365,0.025113778,0.002554715,0.0009174297,0.0024949333,0.96612936,0.00014718823,0.0021138175],"genre_scores_gemma":[0.0010187482,0.000536654,0.000737793,0.00027792944,0.00035192323,9.487433e-7,0.9949274,0.00030500413,0.0018435634],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9944294,0.0011790472,0.0010794349,0.0008773262,0.0020745345,0.00036027696],"domain_scores_gemma":[0.9925557,0.0013002851,0.00055701646,0.0017338797,0.0037328547,0.00012027362],"candidate_categories":["metaresearch","sts","scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.007458658,0.00027085457,0.00044979766,0.00036510566,0.0024372574,0.001225205,0.002776441,0.00019763815,0.017379433],"category_scores_gemma":[0.022280237,0.00019830072,0.0002459283,0.0011180881,0.00035476836,0.00023808291,0.0013184244,0.0003825666,0.0005081774],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003772329,0.0002473894,0.0000016178315,0.00019741901,0.00009580977,0.0000017213513,0.00008890602,0.00080470485,0.00013134297,0.00027643802,0.9831755,0.014601913],"study_design_scores_gemma":[0.0008451193,0.00019458412,0.0003341021,0.00010425031,0.000084154424,0.000010125626,0.00027914817,0.0006332197,0.000050812825,0.0007776688,0.9965013,0.00018546337],"about_ca_topic_score_codex":0.000040120733,"about_ca_topic_score_gemma":0.0000028304253,"teacher_disagreement_score":0.028798088,"about_ca_system_score_codex":0.00017493858,"about_ca_system_score_gemma":0.00004404597,"threshold_uncertainty_score":0.9998116},"labels":[],"label_agreement":null},{"id":"W6931042798","doi":"10.5281/zenodo.15425140","title":"Additive manufacturing: a tool for engineering microstructures and mechanical behavior","year":2025,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Geomechanica (Canada)","funders":"","keywords":"Austenite; Microstructure; European union; Plasticity; Crystal twinning; Crystal plasticity; Austenitic stainless steel","score_opus":0.06290811341146287,"score_gpt":0.3089826469860265,"score_spread":0.24607453357456366,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931042798","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.77519035,0.00019854649,0.2014399,0.0033172409,0.0008350957,0.0036791966,0.0012330444,0.0008596979,0.013246901],"genre_scores_gemma":[0.9963493,0.000022722958,0.0020553835,0.00019907199,0.00007493818,4.4402813e-7,0.00016230726,0.00022400856,0.00091184914],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99825346,0.00014895294,0.0003386295,0.00047995124,0.0005241438,0.0002548531],"domain_scores_gemma":[0.99881196,0.00018627553,0.000072607574,0.00037728727,0.00046845246,0.00008343966],"candidate_categories":["scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0017696603,0.00011901755,0.00015603272,0.00021955042,0.0011100274,0.0010551368,0.0007898357,0.00006063189,0.002900856],"category_scores_gemma":[0.0031304825,0.00010206955,0.000064207416,0.0002595225,0.0000808083,0.00017008501,0.000755218,0.00015786709,0.0002488678],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019153899,0.00017848115,0.000022931337,0.00008100244,0.00007095479,0.0000065778604,0.00059338973,0.0001082399,0.05244881,0.021605683,0.20716906,0.71752334],"study_design_scores_gemma":[0.000464544,0.00012346118,0.0031925603,0.00002504593,0.000019713669,0.000012427509,0.0002139979,0.0003863075,0.024917793,0.003157761,0.9673602,0.00012620204],"about_ca_topic_score_codex":0.0000024080703,"about_ca_topic_score_gemma":1.8325865e-7,"teacher_disagreement_score":0.76019114,"about_ca_system_score_codex":0.0000812073,"about_ca_system_score_gemma":0.0000036124602,"threshold_uncertainty_score":0.9999819},"labels":[],"label_agreement":null},{"id":"W6931197449","doi":"10.5281/zenodo.4312906","title":"The Role of Data in AI","year":2020,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Data governance; Data quality; General partnership; Summit; Transparency (behavior); Corporate governance; Data curation; Information governance; Work (physics)","score_opus":0.23507285848324538,"score_gpt":0.3541728027813011,"score_spread":0.1190999442980557,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931197449","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18713234,0.0020414186,0.01739472,0.13032992,0.00043791838,0.0029818504,0.0011733002,0.00079890416,0.6577096],"genre_scores_gemma":[0.9990628,0.00005845383,0.000119210046,0.0003268915,0.00005671902,9.31001e-9,0.00010581796,0.00014714617,0.00012297387],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9972643,0.0005810559,0.00042871246,0.0003857313,0.0011536606,0.00018654144],"domain_scores_gemma":[0.99815863,0.00015456136,0.00011542677,0.00097612804,0.000498871,0.00009635049],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.005059225,0.0000627935,0.00011019202,0.00006988007,0.0007876605,0.0006463483,0.003955458,0.000024374101,0.004741611],"category_scores_gemma":[0.007639094,0.000042914195,0.00002733078,0.0008415944,0.0001568585,0.00034175123,0.002495948,0.0001805905,0.003969246],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011195489,0.00012225159,0.00030105264,0.00001251498,0.000015947142,0.0000024973635,0.0019292816,0.00015956035,0.006499995,0.0085794395,0.21870275,0.76356274],"study_design_scores_gemma":[0.00017032892,0.00008905979,0.0013695519,0.000006526966,0.0000021143512,0.0000016008589,0.0012280038,0.003455778,0.00045100396,0.003939262,0.9892377,0.00004905979],"about_ca_topic_score_codex":0.000011716112,"about_ca_topic_score_gemma":0.0000011434428,"teacher_disagreement_score":0.8119304,"about_ca_system_score_codex":0.000028776602,"about_ca_system_score_gemma":0.000004406863,"threshold_uncertainty_score":0.99680626},"labels":[],"label_agreement":null},{"id":"W6931277097","doi":"10.5281/zenodo.4306644","title":"tanjapm/DIB_flatflame: DIB complementary cantera calculations","year":2020,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Field (mathematics); Perspective (graphical); Visualization; Feature (linguistics)","score_opus":0.18276707337552794,"score_gpt":0.34344563022740227,"score_spread":0.16067855685187432,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931277097","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000894178,0.000251948,0.008377354,0.0048108203,0.00046568515,0.0011000326,0.0016948523,0.0009936703,0.98221624],"genre_scores_gemma":[0.082462594,0.00070916157,0.004325689,0.004246946,0.00376114,5.992359e-7,0.018885883,0.023943814,0.8616642],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.994604,0.0008569273,0.00075570564,0.0010011251,0.0023217152,0.00046055487],"domain_scores_gemma":[0.9974963,0.00009682091,0.00040756553,0.0011521769,0.000508192,0.0003389266],"candidate_categories":["metaepi_narrow","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0019990061,0.0003120775,0.0004174841,0.00051594886,0.001603531,0.0015748533,0.0028795064,0.0001363672,0.31983212],"category_scores_gemma":[0.0017612483,0.00027739326,0.00017303262,0.0010644404,0.0002536121,0.00016967938,0.0018815241,0.0004620526,0.042960808],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00001994496,0.00009296481,0.000011627851,0.000033401935,0.000074364165,0.000012744053,0.00041859524,0.00002645968,0.0002505219,0.0016479034,0.951707,0.045704443],"study_design_scores_gemma":[0.000411695,0.00013620629,0.00026979105,0.00008119532,0.000029096414,0.000012943619,0.00039470042,0.00033636094,0.000021088594,0.00068642304,0.9973281,0.00029239402],"about_ca_topic_score_codex":0.00014490254,"about_ca_topic_score_gemma":0.000013659488,"teacher_disagreement_score":0.27687132,"about_ca_system_score_codex":0.00021528918,"about_ca_system_score_gemma":0.000010744356,"threshold_uncertainty_score":0.9999678},"labels":[],"label_agreement":null},{"id":"W6931288864","doi":"10.5281/zenodo.5428565","title":"Figure 3 in Elasmobranchs from the Lower Triassic Sulphur Mountain Formation near Wapiti Lake (BC, Canada)","year":2007,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Holotype; Single specimen; Table (database); Assemblage (archaeology); Sulfur","score_opus":0.07940355660982276,"score_gpt":0.28752318615575584,"score_spread":0.20811962954593308,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931288864","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0046407445,0.00045345392,0.002277394,0.005133532,0.00090232893,0.001757183,0.0019491807,0.00038881446,0.9824974],"genre_scores_gemma":[0.19286515,0.0007635022,0.000728573,0.004987746,0.0026542712,0.0000011076846,0.013153966,0.011190774,0.77365494],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9938578,0.0011366616,0.0008138195,0.00069852045,0.00297251,0.0005207075],"domain_scores_gemma":[0.99735695,0.00032232836,0.0004447816,0.001177679,0.0005292484,0.00016899456],"candidate_categories":["scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.005574031,0.00028782955,0.00035629576,0.00036913762,0.0012247945,0.0015962105,0.0026992,0.00021861574,0.1674194],"category_scores_gemma":[0.0034406865,0.0002095311,0.00010222082,0.0012686941,0.00020912122,0.0002469344,0.000731628,0.0006491582,0.013366706],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000067394605,0.00009755792,0.000023393659,0.000015459005,0.00002483765,0.000016949465,0.000436782,0.00008449205,0.00002620604,0.00032969288,0.9551752,0.043701995],"study_design_scores_gemma":[0.0006601433,0.000075824275,0.0011296108,0.00012070485,0.000012143622,0.0000040444224,0.00059246045,0.00035254296,0.000024422969,0.0005067716,0.9962837,0.0002376217],"about_ca_topic_score_codex":0.019816272,"about_ca_topic_score_gemma":0.040646285,"teacher_disagreement_score":0.20884246,"about_ca_system_score_codex":0.0004424715,"about_ca_system_score_gemma":0.00004866356,"threshold_uncertainty_score":0.99944025},"labels":[],"label_agreement":null},{"id":"W6931377715","doi":"10.5281/zenodo.3460765","title":"drkupi/APWS: APWS Web Portal","year":2019,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph","funders":"","keywords":"Web application; Process (computing); Software; The Internet; Web service","score_opus":0.12533603815264985,"score_gpt":0.3213343729286678,"score_spread":0.19599833477601794,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931377715","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00017223202,0.0002710462,0.00092652964,0.0008757955,0.0006924226,0.00096806476,0.000516828,0.0008493191,0.9947278],"genre_scores_gemma":[0.014409762,0.0002770458,0.00031363682,0.0003614023,0.00069744006,8.9675716e-8,0.00108154,0.0065797204,0.9762794],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99378264,0.00083521346,0.0007178588,0.0011471385,0.0029751842,0.0005419505],"domain_scores_gemma":[0.9965273,0.000094860865,0.0005084673,0.0018357841,0.00077222794,0.0002613619],"candidate_categories":["metaepi_narrow","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0044350345,0.00033522007,0.00045648566,0.0007430611,0.0009819313,0.0018868465,0.0036513817,0.00025719515,0.26654693],"category_scores_gemma":[0.002974383,0.00028250425,0.0001963583,0.0009607095,0.00025898145,0.00020147688,0.0018224211,0.0005162528,0.20568582],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000019194016,0.00012897227,0.000008362698,0.00004620948,0.00005261798,0.000011795992,0.00016833373,0.0000141932,0.00027551988,0.0019331409,0.9563556,0.04098611],"study_design_scores_gemma":[0.00043581787,0.00016862193,0.000079280086,0.00009820561,0.000019481553,0.000024867957,0.00034847404,0.000107887485,0.000021025613,0.0005296718,0.9978425,0.00032419953],"about_ca_topic_score_codex":0.000025701725,"about_ca_topic_score_gemma":0.0000027537594,"teacher_disagreement_score":0.06086111,"about_ca_system_score_codex":0.00013272195,"about_ca_system_score_gemma":0.000016911119,"threshold_uncertainty_score":0.9999627},"labels":[],"label_agreement":null},{"id":"W6931405649","doi":"10.5281/zenodo.4096981","title":"cb01-(ITA) Falling (2020) - Streaming ALTADEFINIZIONE rjb","year":2020,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Falling (accident); Daughter; The Internet; Falling in love; Link (geometry); Streaming current; Window (computing)","score_opus":0.20107666931437554,"score_gpt":0.3219271611802912,"score_spread":0.12085049186591565,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931405649","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4945632,0.0007503496,0.06162058,0.06378927,0.000922202,0.0023223944,0.00058298104,0.003128137,0.3723209],"genre_scores_gemma":[0.9964343,0.000080486636,0.0009032497,0.0009835152,0.00032706547,3.784687e-8,0.0002096131,0.00049541483,0.000566359],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9955099,0.00063967105,0.0006521666,0.000802461,0.0019646543,0.00043111626],"domain_scores_gemma":[0.9975953,0.00020251606,0.0002090979,0.00064332475,0.0009778243,0.00037191887],"candidate_categories":["metaresearch","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0034479403,0.00017498646,0.00024142973,0.0001472778,0.0021518439,0.0017802466,0.0021060337,0.000069026435,0.02550346],"category_scores_gemma":[0.010159278,0.00015509731,0.00010728448,0.0012722968,0.0001540276,0.00042964643,0.0013656757,0.00034995677,0.02802791],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013483955,0.00025682172,0.00014804788,0.000060225433,0.000063962776,0.00003559669,0.004695109,0.0012551352,0.01715565,0.005102264,0.5493067,0.42178568],"study_design_scores_gemma":[0.00042586154,0.0002754428,0.0013922343,0.000023122764,0.00001133762,0.000014867517,0.0016307834,0.0019293776,0.0006150434,0.0013748911,0.9921042,0.00020284402],"about_ca_topic_score_codex":0.000009588001,"about_ca_topic_score_gemma":3.3359015e-7,"teacher_disagreement_score":0.50187105,"about_ca_system_score_codex":0.000101392245,"about_ca_system_score_gemma":0.0000066406405,"threshold_uncertainty_score":0.999256},"labels":[],"label_agreement":null},{"id":"W6931450987","doi":"10.5281/zenodo.5706364","title":"E3DMT version 1 manual","year":2021,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Magnetotellurics; Inversion (geology); Software; Data collection","score_opus":0.14903162127798597,"score_gpt":0.33910508978367876,"score_spread":0.1900734685056928,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931450987","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.229937,0.00025990923,0.012814964,0.010367544,0.0008150981,0.00056127785,0.0002587921,0.0007719296,0.74421346],"genre_scores_gemma":[0.98592573,0.00008532492,0.0007737121,0.0006311228,0.00022790232,2.1870704e-8,0.00059749983,0.0005071057,0.011251554],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9962238,0.0007844707,0.00038774227,0.00057469925,0.0017303728,0.0002988896],"domain_scores_gemma":[0.99722666,0.00010482975,0.00010528907,0.00079292455,0.0015970662,0.0001732319],"candidate_categories":["sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.003620784,0.000105922394,0.00014596885,0.00016230052,0.0019888182,0.001435979,0.0013319803,0.000054012355,0.087236166],"category_scores_gemma":[0.0050724675,0.00009372152,0.0000878927,0.0009132155,0.0001223691,0.00031552644,0.0014426651,0.00020517991,0.050069943],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000046874717,0.00020872432,0.00003997053,0.00001337631,0.000020684989,0.000032432,0.0006288789,0.00012667914,0.00981724,0.0038367356,0.7948361,0.19039233],"study_design_scores_gemma":[0.00032475355,0.0001019228,0.0018548893,0.0000141037235,0.000006239891,0.00004032698,0.0010050558,0.00019651659,0.00243849,0.0028564571,0.99104786,0.00011341232],"about_ca_topic_score_codex":0.000003826435,"about_ca_topic_score_gemma":3.225625e-7,"teacher_disagreement_score":0.7559888,"about_ca_system_score_codex":0.00012364311,"about_ca_system_score_gemma":0.0000064942383,"threshold_uncertainty_score":0.99960065},"labels":[],"label_agreement":null},{"id":"W6931508920","doi":"10.5281/zenodo.3791116","title":"Gyrophaena gilvicollis Casey 1906","year":2009,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Canadian Forest Service; Natural Resources Canada","funders":"","keywords":"Maple; Ridge; Section (typography); Hardwood","score_opus":0.15759798700454228,"score_gpt":0.33777645769193776,"score_spread":0.18017847068739548,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931508920","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19637686,0.00030692414,0.0137731545,0.013907494,0.0005935479,0.0014684431,0.0002646379,0.001428027,0.7718809],"genre_scores_gemma":[0.99490315,0.00004353045,0.00043034827,0.00083727384,0.0001897068,2.1362366e-8,0.0001485948,0.00030601464,0.0031413545],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99582726,0.0006674642,0.0005522648,0.00064675504,0.0018820984,0.00042418367],"domain_scores_gemma":[0.997336,0.00009711339,0.00016530746,0.00095747976,0.0011893036,0.00025484705],"candidate_categories":["sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0044974764,0.00015763794,0.00020576811,0.00029741743,0.0023431752,0.0018515447,0.0020708977,0.00006613349,0.028094294],"category_scores_gemma":[0.0046059233,0.00013099711,0.00010590614,0.001185451,0.00014480557,0.00041401933,0.000574887,0.00024870134,0.026224857],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008145045,0.00028431407,0.000013030937,0.000007036932,0.000014774253,0.000024491199,0.0007290495,0.00020455517,0.0053352886,0.0058481353,0.52156854,0.46588933],"study_design_scores_gemma":[0.00038325408,0.00036868308,0.0031146952,0.000013385074,0.000007213056,0.000043755277,0.00032068946,0.00033388878,0.00067594775,0.0054501668,0.9891235,0.00016478555],"about_ca_topic_score_codex":0.000009794474,"about_ca_topic_score_gemma":3.82434e-7,"teacher_disagreement_score":0.7985263,"about_ca_system_score_codex":0.00015815363,"about_ca_system_score_gemma":0.0000048921465,"threshold_uncertainty_score":0.9991846},"labels":[],"label_agreement":null},{"id":"W6931537636","doi":"10.5281/zenodo.4458985","title":"Hemp Max Lab Oil { CA } # No.1 Pain Killer Hemp Oil ? Read-Pros &amp; Cons","year":2021,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Point (geometry); Limiting; Filter (signal processing); Work (physics); Product (mathematics); Paraphernalia","score_opus":0.14053998126506084,"score_gpt":0.32287187951133245,"score_spread":0.1823318982462716,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931537636","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04397739,0.00039836115,0.0044855746,0.018258356,0.0010702398,0.00032702135,0.00043246473,0.0008893815,0.93016124],"genre_scores_gemma":[0.33015773,0.0014331766,0.0063961195,0.006980654,0.0013306779,9.0254997e-7,0.0039716237,0.0030071933,0.64672196],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99300605,0.0023017107,0.00076216296,0.0009927351,0.002348935,0.00058840285],"domain_scores_gemma":[0.99413043,0.00034422145,0.00021767511,0.00150909,0.0034511995,0.0003473603],"candidate_categories":["metaresearch","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.009670456,0.00023642469,0.0003181863,0.00021892149,0.0022500525,0.002026981,0.0018230145,0.00011427293,0.15473731],"category_scores_gemma":[0.021313505,0.00020856525,0.0001512654,0.0013293419,0.00030514263,0.00035420177,0.0017193098,0.00041846564,0.087860905],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00005311661,0.00027238345,0.000048467587,0.00005432394,0.00003730531,0.000031768875,0.00072740304,0.00006582423,0.008673715,0.000855169,0.78036034,0.20882016],"study_design_scores_gemma":[0.00052110944,0.00010494046,0.00027481318,0.00005848673,0.000011045807,0.000040412007,0.00067971786,0.00020742991,0.00060509145,0.000838625,0.99640554,0.00025277864],"about_ca_topic_score_codex":0.00003028963,"about_ca_topic_score_gemma":0.000007483298,"teacher_disagreement_score":0.28618032,"about_ca_system_score_codex":0.00021539183,"about_ca_system_score_gemma":0.000028928054,"threshold_uncertainty_score":0.9990489},"labels":[],"label_agreement":null},{"id":"W6931538020","doi":"10.5281/zenodo.7963601","title":"reds2401/Hydrotel-Telemac_results: V1","year":2023,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Sherbrooke","funders":"","keywords":"Nucleofection; Gestational period; TSG101; Dysgeusia; Diafiltration; Liquation; Emperipolesis; Triacetin; Demotion","score_opus":0.1593613973510762,"score_gpt":0.3395902359502935,"score_spread":0.18022883859921726,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931538020","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000083879306,0.0002180385,0.0015204968,0.0014684858,0.0006768589,0.0009781609,0.00076343626,0.0031019202,0.9911887],"genre_scores_gemma":[0.003843647,0.00083406223,0.00037974823,0.00027815794,0.0010797224,2.0944091e-7,0.0010761475,0.019240154,0.97326815],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9931377,0.0010233994,0.00083029375,0.0012285062,0.0031449415,0.0006351331],"domain_scores_gemma":[0.9963723,0.00018125103,0.0004974923,0.0018467377,0.0008024435,0.000299779],"candidate_categories":["metaresearch","metaepi_narrow","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0061208196,0.00033697623,0.00045359274,0.0009484079,0.0014839213,0.0020833255,0.003839541,0.00026690858,0.13580151],"category_scores_gemma":[0.009010331,0.00028571358,0.00019953954,0.0017138179,0.00029888147,0.00019068546,0.0020330756,0.0005281041,0.20441912],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00002426401,0.00008417179,0.0000020309726,0.00003299872,0.000052312833,0.000018054698,0.00019191,0.000026440244,0.00012043803,0.0007386479,0.87383157,0.12487714],"study_design_scores_gemma":[0.000399366,0.00014781779,0.00009028409,0.00014947107,0.000017090597,0.000016606517,0.00024057677,0.00009372366,0.000047544603,0.0014650542,0.9970196,0.00031288262],"about_ca_topic_score_codex":0.000076560624,"about_ca_topic_score_gemma":0.000007703059,"teacher_disagreement_score":0.12456425,"about_ca_system_score_codex":0.00022085161,"about_ca_system_score_gemma":0.000008508504,"threshold_uncertainty_score":0.9999595},"labels":[],"label_agreement":null},{"id":"W6931554602","doi":"10.5281/zenodo.8412257","title":"ComputeCanada/magic_castle: Magic Castle 12.6.7","year":2023,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University; Toronto Dementia Research Alliance; Compute Canada","funders":"","keywords":"MAGIC (telescope); History of computing; Key (lock)","score_opus":0.19707770640107322,"score_gpt":0.3325461461616843,"score_spread":0.1354684397606111,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931554602","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000040558654,0.0001630908,0.0027141578,0.0014504426,0.00088429166,0.00067507237,0.00063013606,0.0019906955,0.99145156],"genre_scores_gemma":[0.0030519518,0.00024354615,0.0003649926,0.00031084337,0.0008262551,6.8399366e-8,0.0006490682,0.012442614,0.9821107],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99466157,0.00065590447,0.00059321383,0.0009848124,0.0025843254,0.00052018155],"domain_scores_gemma":[0.9972211,0.00013605959,0.00034936806,0.0013850247,0.00062463473,0.0002838014],"candidate_categories":["metaepi_narrow","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0036668223,0.00028572237,0.00038615003,0.00074102124,0.0011001594,0.0016539706,0.002883353,0.0001739909,0.18188283],"category_scores_gemma":[0.0027620492,0.00025227905,0.00014474151,0.0010779353,0.00026997467,0.00012690102,0.0018518028,0.00043487147,0.13753283],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000009979093,0.000061049475,0.0000022470167,0.000025549616,0.000036872523,0.00002996943,0.00013464502,0.000029152008,0.000036891794,0.000787852,0.8828668,0.11597901],"study_design_scores_gemma":[0.00030075002,0.00011241012,0.00014316833,0.00008899695,0.000015093342,0.000037512054,0.00021856939,0.00021457998,0.000009351084,0.000855087,0.99773496,0.0002695258],"about_ca_topic_score_codex":0.00037504704,"about_ca_topic_score_gemma":0.000087968045,"teacher_disagreement_score":0.11570948,"about_ca_system_score_codex":0.00026434814,"about_ca_system_score_gemma":0.000012493079,"threshold_uncertainty_score":0.99999297},"labels":[],"label_agreement":null},{"id":"W6931637154","doi":"10.5683/sp3/gm65t8","title":"Survey of Household Spending, 1999 [Canada] [Excel]","year":2015,"lang":"en","type":"dataset","venue":"Borealis","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Household income; Principal (computer security); Spouse; Survey data collection; Payment; Pension; Family income","score_opus":0.284473020106409,"score_gpt":0.3688724870377677,"score_spread":0.08439946693135869,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931637154","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00020137825,0.00042760544,0.000007797135,0.00013739361,0.00087374815,0.00031807047,0.99659365,0.0000115258745,0.0014288484],"genre_scores_gemma":[0.0007401901,0.0001315861,0.000027093576,0.00018722517,0.00015575622,0.000010146652,0.99841845,0.000014670536,0.00031490545],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9895134,0.000913517,0.0015466197,0.00078933727,0.006827796,0.00040931857],"domain_scores_gemma":[0.9936534,0.0012955986,0.0010673013,0.0027435687,0.00093616586,0.00030392403],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.015786348,0.00036849923,0.00094061927,0.00026233576,0.00008729546,0.00016604587,0.002913767,0.00027299786,0.0007759029],"category_scores_gemma":[0.010491507,0.00026570034,0.0001540717,0.00064611965,0.00016134005,0.00011245496,0.00039426863,0.0003591501,0.00004082583],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000027290112,0.00009507932,0.004688051,0.000027306318,0.00003838645,0.000014786513,0.000008433053,0.00004755266,8.653229e-7,0.0000053496724,0.99433136,0.00071553455],"study_design_scores_gemma":[0.00020133548,0.00006905137,0.063942514,0.000048094156,0.00003509752,0.0000012838692,0.00003074944,0.000005702679,0.000019615856,0.0004338762,0.934943,0.00026965095],"about_ca_topic_score_codex":0.9904264,"about_ca_topic_score_gemma":0.99493194,"teacher_disagreement_score":0.05938834,"about_ca_system_score_codex":0.00029123964,"about_ca_system_score_gemma":0.0016389486,"threshold_uncertainty_score":0.9999795},"labels":[],"label_agreement":null},{"id":"W6931648760","doi":"10.5683/sp3/sz1lis","title":"Replication Data and Code for: The Canadian Income Taxation: Statistical Analysis and Parametric Estimates","year":2021,"lang":"en","type":"dataset","venue":"Borealis","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McMaster University","funders":"","keywords":"Replicate; Replication (statistics); Parametric statistics; Statistical analysis; Code (set theory); Nonparametric statistics; Data file","score_opus":0.20212963802193457,"score_gpt":0.42286689708361697,"score_spread":0.2207372590616824,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931648760","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000041340907,0.00077707676,0.008846552,0.0027486973,0.00005997367,0.00041337372,0.98708683,0.000005027849,0.000021124848],"genre_scores_gemma":[0.00066343177,0.0005821957,0.003192817,0.0002813206,0.00006332085,0.000064958585,0.9951289,0.0000044279536,0.000018579984],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99677575,0.00019207019,0.0006213568,0.0010439863,0.0011589837,0.00020786801],"domain_scores_gemma":[0.9886234,0.005817655,0.0004888772,0.004422456,0.00043361253,0.00021402191],"candidate_categories":["metaresearch","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.006897932,0.00017026682,0.00040240132,0.00045401778,0.00051500776,0.0010487792,0.0012817285,0.00015113218,0.00012948552],"category_scores_gemma":[0.033997495,0.00010494791,0.000047931917,0.0012632948,0.00021644798,0.00012483475,0.00032466312,0.00016181452,0.0000030541935],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000049255427,0.00001242144,0.0072243116,0.000019140178,0.00018002251,0.0000021889011,0.0000096109925,0.000029597586,8.207794e-8,0.00013325688,0.9841636,0.008220854],"study_design_scores_gemma":[0.000074727686,0.000018829382,0.11847987,0.000009242316,0.00082846807,0.0000022081272,0.000055424556,0.008767959,8.3283385e-7,0.00282515,0.86882216,0.000115145514],"about_ca_topic_score_codex":0.47000617,"about_ca_topic_score_gemma":0.97274333,"teacher_disagreement_score":0.50273716,"about_ca_system_score_codex":0.000087187036,"about_ca_system_score_gemma":0.00032146418,"threshold_uncertainty_score":0.9999882},"labels":[],"label_agreement":null},{"id":"W6931672917","doi":"10.5281/zenodo.8194413","title":"Fig. 8 in Metabolites isolated from the human intestinal fungus Penicillium oxalicum SL2 and their agonistic effects on PXR and FXR","year":2022,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"CAE (Canada)","funders":"","keywords":"Agonistic behaviour; Pregnane X receptor; Mutant; Metabolite; Fungus; Strain (injury); Structure–activity relationship","score_opus":0.075895808489994,"score_gpt":0.2957144855974231,"score_spread":0.21981867710742908,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931672917","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.14287788,0.0072391713,0.0006937287,0.0053373287,0.00087720173,0.004757974,0.0023132104,0.0010402219,0.8348633],"genre_scores_gemma":[0.91319114,0.00079839915,0.00015020432,0.0013982428,0.00072279415,0.0000012615318,0.0019902396,0.0046278853,0.07711985],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.995568,0.0018402791,0.00043404865,0.0008378038,0.000997526,0.0003223437],"domain_scores_gemma":[0.99791557,0.000659225,0.0002647459,0.0008386474,0.00019171624,0.00013011589],"candidate_categories":["sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0031350644,0.00027859624,0.0003620183,0.00034741973,0.0014902315,0.001299577,0.0015514035,0.00009324363,0.041549783],"category_scores_gemma":[0.004294886,0.00018265545,0.000059798716,0.00064156216,0.00026649813,0.00006729678,0.001762973,0.00061997003,0.0012214113],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011522552,0.00023328462,0.0003430879,0.00008151972,0.00011588497,0.000028753982,0.0029871548,0.000017054059,0.004724587,0.0024275305,0.9052592,0.0836667],"study_design_scores_gemma":[0.0004718577,0.00033648786,0.0044624484,0.00011527618,0.000022025026,0.000011430762,0.0006598801,0.000153092,0.00006659983,0.0021740836,0.9913154,0.00021137197],"about_ca_topic_score_codex":0.00023288661,"about_ca_topic_score_gemma":0.000017146393,"teacher_disagreement_score":0.77031326,"about_ca_system_score_codex":0.00010186797,"about_ca_system_score_gemma":0.000005484272,"threshold_uncertainty_score":0.9998097},"labels":[],"label_agreement":null},{"id":"W6931675262","doi":"10.5683/sp3/m2iv1y","title":"Federated Identity Management: Spring 2006","year":2006,"lang":"en","type":"dataset","venue":"Borealis","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Identity (music); Presentation (obstetrics); General partnership; Identity management; Spring (device)","score_opus":0.08751611499267774,"score_gpt":0.35519868742695326,"score_spread":0.2676825724342755,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931675262","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00003258423,0.00022509722,0.0007905998,0.0002573592,0.0006823094,0.00054431095,0.9835323,0.00005842019,0.013877004],"genre_scores_gemma":[0.000039200648,0.00017545967,0.00021637375,0.0004134846,0.0003589989,0.000050106246,0.99666595,0.000016161674,0.0020642355],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9918786,0.00036433633,0.0013620753,0.0011169076,0.0047457754,0.00053225784],"domain_scores_gemma":[0.9969208,0.00021405969,0.0006033854,0.001694951,0.0004195805,0.00014718345],"candidate_categories":["metaepi_narrow","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.005316784,0.00043864755,0.00060869823,0.00048285216,0.00041628562,0.0017160246,0.0020976267,0.0003069152,0.00098122],"category_scores_gemma":[0.00063070876,0.00034459328,0.00031289263,0.00079594,0.00011738557,0.0005038312,0.0005866333,0.00039113557,0.0010276879],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00001179054,0.00015922764,0.00029867725,0.000060893737,0.000059265272,0.000056813275,0.0000025449965,0.000058103265,0.0000012490807,0.00015624738,0.99696654,0.002168633],"study_design_scores_gemma":[0.00026845007,0.00002335343,0.007447476,0.00008118079,0.00009300382,0.0000014873004,0.000029337645,0.000022685159,0.000010558604,0.005685581,0.9859599,0.00037700293],"about_ca_topic_score_codex":0.10278401,"about_ca_topic_score_gemma":0.116023906,"teacher_disagreement_score":0.013239901,"about_ca_system_score_codex":0.0001837567,"about_ca_system_score_gemma":0.000055331446,"threshold_uncertainty_score":0.999932},"labels":[],"label_agreement":null},{"id":"W6931747030","doi":"10.5683/sp2/lutyat","title":"Data from: Rates of hypoxia induction alter mechanisms of O2 uptake and the critical O2 tension of goldfish","year":2021,"lang":"en","type":"dataset","venue":"Borealis","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Hypoxia (environmental); Confounding; Fish <Actinopterygii>; Peripheral","score_opus":0.22109727069254453,"score_gpt":0.39117526646829454,"score_spread":0.17007799577575,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931747030","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0015285818,0.00052544073,0.00041881076,0.0013186509,0.00043551877,0.0002964343,0.9954083,0.0000031989116,0.00006506663],"genre_scores_gemma":[0.0038752393,0.00071239047,0.0012498136,0.00022491766,0.00013345198,0.000010347762,0.99376494,0.000008668637,0.00002022545],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9945338,0.00075288414,0.0014638322,0.0007238514,0.0023571705,0.00016849983],"domain_scores_gemma":[0.9926587,0.0025793102,0.0007045407,0.0030415023,0.00094644615,0.0000695319],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0066454387,0.00021840139,0.0008707731,0.00013098329,0.000070579765,0.00008649572,0.0015752037,0.00025472304,0.00050476566],"category_scores_gemma":[0.012587219,0.0001234229,0.00013731561,0.0003082384,0.00068534305,0.00021359118,0.00091330445,0.00026079442,0.0000039978368],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011512559,0.00014396916,0.000019159042,0.000072995324,0.000061387596,0.0000026820608,0.00004017598,0.000001815171,0.0019318168,0.0004042758,0.9960478,0.0011588278],"study_design_scores_gemma":[0.0011392382,0.00019531962,0.0042448356,0.0005591958,0.00053242897,0.0000062386557,0.0014020494,0.0003156095,0.008995022,0.09029685,0.892028,0.000285221],"about_ca_topic_score_codex":0.013984207,"about_ca_topic_score_gemma":0.003312231,"teacher_disagreement_score":0.104019776,"about_ca_system_score_codex":0.000015646714,"about_ca_system_score_gemma":0.00013998887,"threshold_uncertainty_score":0.99573016},"labels":[],"label_agreement":null},{"id":"W6931767154","doi":"10.5281/zenodo.7065821","title":"YichengDWu/Sophon.jl: v0.5.0","year":2023,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"CUDA; Code (set theory); Work (physics); Macro","score_opus":0.19229175428168743,"score_gpt":0.34275185059852215,"score_spread":0.1504600963168347,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931767154","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00008329499,0.00024281719,0.0013704005,0.0016831497,0.000852575,0.0007495124,0.00045989643,0.0031251172,0.99143326],"genre_scores_gemma":[0.0042589563,0.0004685994,0.00033091445,0.00030889385,0.0009653328,1.4456172e-7,0.00065189006,0.017791344,0.9752239],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99394655,0.0008854077,0.0006693987,0.0010868547,0.002862301,0.00054949033],"domain_scores_gemma":[0.99669987,0.0001686552,0.00039452437,0.0016323613,0.0008269099,0.00027769603],"candidate_categories":["metaepi_narrow","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0052909073,0.0003087421,0.00040373357,0.0008319646,0.0013800072,0.001815747,0.0035679445,0.00024301109,0.18190111],"category_scores_gemma":[0.0072345124,0.00026532655,0.00018162558,0.0015733923,0.00026126282,0.0001716383,0.0020068856,0.00049619674,0.2388476],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000014250516,0.00008869791,0.000004062,0.000030353434,0.000049935614,0.000012903919,0.00023603263,0.000016632514,0.00011837985,0.0010483006,0.90844125,0.089939184],"study_design_scores_gemma":[0.00030077822,0.000121146506,0.00019701476,0.00009976244,0.000017427006,0.000011057343,0.00027662818,0.00005638676,0.00003675623,0.0015485926,0.99704725,0.00028722046],"about_ca_topic_score_codex":0.00006888944,"about_ca_topic_score_gemma":0.000007475513,"teacher_disagreement_score":0.089651965,"about_ca_system_score_codex":0.00020500073,"about_ca_system_score_gemma":0.000007947082,"threshold_uncertainty_score":0.9999799},"labels":[],"label_agreement":null},{"id":"W6931916823","doi":"10.5281/zenodo.7343269","title":"Adventures with SPIRou: The APERO pipeline and LBL RV analysis tool","year":2022,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université de Montréal","funders":"","keywords":"Pipeline (software); Outlier; Data reduction; Line (geometry); Component (thermodynamics); Reduction (mathematics)","score_opus":0.08621879554823171,"score_gpt":0.29663387015278164,"score_spread":0.21041507460454995,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931916823","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.77873456,0.00077768747,0.03745391,0.029727718,0.00033328775,0.0023960734,0.0006300149,0.0007970826,0.14914964],"genre_scores_gemma":[0.99663305,0.000030947274,0.00016285157,0.00056462066,0.000059096823,1.5665204e-7,0.00015468792,0.00018720282,0.0022073726],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9959505,0.0009880224,0.0003643702,0.0005174991,0.0019298644,0.00024971145],"domain_scores_gemma":[0.998336,0.00014826622,0.00015608064,0.00077171094,0.0004956601,0.00009225659],"candidate_categories":["sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.005530946,0.00011507939,0.00017316549,0.00027395715,0.0041810093,0.0011246353,0.0015468736,0.000019230058,0.033480156],"category_scores_gemma":[0.0014021948,0.000071431066,0.00007859235,0.002019187,0.00020840758,0.00017886535,0.0016862898,0.00027981788,0.0008150587],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00055205764,0.0006853911,0.0014949777,0.000029503692,0.00053075084,0.000033483724,0.0062757866,0.013916279,0.0023795296,0.010996833,0.5698001,0.3933053],"study_design_scores_gemma":[0.0003085854,0.0002611855,0.0069229417,0.0000027954452,0.00007216654,0.000024197154,0.002501321,0.0019206693,0.00005388218,0.0008723974,0.9869433,0.000116537594],"about_ca_topic_score_codex":0.000019279418,"about_ca_topic_score_gemma":0.000002138405,"teacher_disagreement_score":0.41714323,"about_ca_system_score_codex":0.00008255141,"about_ca_system_score_gemma":0.0000042127863,"threshold_uncertainty_score":0.9999629},"labels":[],"label_agreement":null},{"id":"W6931924740","doi":"10.5281/zenodo.7629383","title":"Fig. 16 in Revision of the Nearctic species of the Lasioglossum (Dialictus) gemmatum species complex (Hymenoptera: Halictidae)","year":2023,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"","keywords":"Nearctic ecozone; Species complex; Scale (ratio); Taxonomy (biology); Ecological succession","score_opus":0.15850588085818568,"score_gpt":0.31084743807642784,"score_spread":0.15234155721824216,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931924740","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0035654502,0.00018445718,0.00014544796,0.0035885787,0.0008144067,0.001554139,0.00058365025,0.0002786348,0.98928523],"genre_scores_gemma":[0.2888231,0.0008412972,0.00012067726,0.0002588183,0.00045176575,1.5524859e-7,0.00014165904,0.0039783884,0.70538414],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99405104,0.0014469583,0.00092262507,0.000579755,0.0026527047,0.0003468907],"domain_scores_gemma":[0.9965774,0.00019416602,0.000758838,0.0017568481,0.0006380825,0.000074670315],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0037463906,0.00024042376,0.0004664389,0.00043067505,0.00073374383,0.00045243354,0.003918501,0.00013587264,0.054489832],"category_scores_gemma":[0.005400295,0.00014306472,0.00023160779,0.0020312949,0.00083600334,0.00010954372,0.0029023543,0.0004164857,0.0029453435],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000027559467,0.00012756078,0.00030317475,0.00012630655,0.000037136197,0.0000013186573,0.0008437811,0.00005827919,0.0009005757,0.0033786478,0.9843917,0.009803978],"study_design_scores_gemma":[0.0002800379,0.000072261486,0.017292842,0.0004964824,0.000015485257,0.0000030318383,0.0006305224,0.00003913711,0.00027419475,0.00063329603,0.98012584,0.0001368679],"about_ca_topic_score_codex":0.00017177645,"about_ca_topic_score_gemma":0.000047774185,"teacher_disagreement_score":0.28525764,"about_ca_system_score_codex":0.0002205844,"about_ca_system_score_gemma":0.000013515911,"threshold_uncertainty_score":0.997831},"labels":[],"label_agreement":null},{"id":"W6931969718","doi":"10.5683/sp/p8yxp5","title":"Canadian Business Patterns, December 2012 [B2020]","year":2023,"lang":"en","type":"dataset","venue":"Borealis","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Census; Metropolitan area; Subdivision; Sampling frame; Line of business; Classification scheme; Small business","score_opus":0.1196570038811735,"score_gpt":0.35295985099336336,"score_spread":0.23330284711218985,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6931969718","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00003308088,0.00013278826,0.000023654646,0.004672189,0.0018464049,0.000410462,0.99167264,0.00004390484,0.001164893],"genre_scores_gemma":[0.000010676146,0.0004961882,0.000020217156,0.0013467652,0.0006184575,0.00007072936,0.99434763,0.000027414155,0.0030619304],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9934541,0.000329801,0.0010384667,0.0010203468,0.0034058455,0.0007514155],"domain_scores_gemma":[0.9949947,0.00060080073,0.00035028663,0.0025630556,0.0008799952,0.0006111652],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.004372786,0.0004279311,0.00061759335,0.0007218496,0.00027499453,0.00064693467,0.0025238495,0.00045714722,0.002805142],"category_scores_gemma":[0.0034136574,0.0003171217,0.00021404937,0.0012056082,0.00007370602,0.00025874755,0.00029211957,0.00041030804,0.010500845],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000058256346,0.00003286671,0.0006706107,0.000038099653,0.000034344066,0.000093395596,0.000013671448,0.000028024833,2.6859504e-7,0.000007936564,0.9964412,0.0026337882],"study_design_scores_gemma":[0.00013177916,0.000016652637,0.026000815,0.00010252399,0.000047773075,0.0000036442575,0.000066419816,0.000005588154,0.0000012029768,0.0012165377,0.9720276,0.0003794813],"about_ca_topic_score_codex":0.96570534,"about_ca_topic_score_gemma":0.99626833,"teacher_disagreement_score":0.030562941,"about_ca_system_score_codex":0.0004085749,"about_ca_system_score_gemma":0.0009637519,"threshold_uncertainty_score":0.99992806},"labels":[],"label_agreement":null},{"id":"W6939986338","doi":"10.6084/m9.figshare.c.3911695_d7","title":"Additional file 7: of Barriers and facilitators to the implementation of a school-based physical activity policy in Canada: application of the theoretical domains framework","year":2017,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Coding (social sciences); Data collection; Measure (data warehouse); Context (archaeology); Work (physics); Key (lock)","score_opus":0.04988355207348834,"score_gpt":0.3659384260501186,"score_spread":0.3160548739766303,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6939986338","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.122949824,9.988651e-7,0.000018602514,0.0016355346,0.000008241715,0.0003469684,0.87488097,9.366978e-7,0.00015792757],"genre_scores_gemma":[0.99490726,4.9421406e-8,0.00008109921,0.00009089694,0.000024547773,0.00031431168,0.004575954,0.000001932239,0.000003954889],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99839985,0.00015057954,0.00022715563,0.00017293874,0.00094865187,0.00010081363],"domain_scores_gemma":[0.9959438,0.0029058137,0.0003132138,0.00059620175,0.0001725045,0.00006846407],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00029164102,0.00006412784,0.00013863237,0.00003711974,0.0001383545,0.000024331071,0.00059932953,0.000026898002,0.5210727],"category_scores_gemma":[0.030323736,0.00003559734,0.000056679575,0.00021059027,0.000093406175,0.000069488146,0.00014017119,0.000098325756,0.00001574283],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000029270861,0.0000406836,0.005960369,0.000028924702,0.000008087355,8.291882e-8,0.00022586147,0.0001785947,0.00007530016,0.0014577566,0.9788288,0.013166268],"study_design_scores_gemma":[0.00013526347,0.000041545816,0.95515937,0.0002986312,0.0000029062112,7.384339e-8,0.0008939524,0.00091327,0.001936885,0.009053132,0.031506263,0.000058690093],"about_ca_topic_score_codex":0.028282076,"about_ca_topic_score_gemma":0.09956644,"teacher_disagreement_score":0.949199,"about_ca_system_score_codex":0.000114540126,"about_ca_system_score_gemma":0.0011652507,"threshold_uncertainty_score":0.9781887},"labels":[],"label_agreement":null},{"id":"W6958274219","doi":"10.6084/m9.figshare.11960187","title":"Additional file 2 of Development, validation and application of a 3D printed model depicting adenoid hypertrophy in comparison to a Nasoendoscopy","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Flowchart; 3d printed; Adenoid hypertrophy; Sampling (signal processing); Muscle hypertrophy","score_opus":0.1782430192798449,"score_gpt":0.3467289463942152,"score_spread":0.1684859271143703,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6958274219","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.012645319,0.000014146878,0.0018294853,0.00016455649,0.0000049940177,0.00049820106,0.9842595,0.00001428595,0.000569498],"genre_scores_gemma":[0.69376475,2.0095064e-7,0.05160801,0.00019793733,0.00002390263,0.0008620992,0.25350273,0.0000074888912,0.00003290904],"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99839485,0.000049130675,0.0005557037,0.00026729805,0.0006458032,0.00008722333],"domain_scores_gemma":[0.99833375,0.0009307295,0.00026544186,0.00013300135,0.0002752167,0.00006188086],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00014659065,0.000071126524,0.00018198669,0.00008658647,0.000035334197,0.000022367778,0.00021581573,0.000033190634,0.5080384],"category_scores_gemma":[0.009821159,0.00006245068,0.000023123814,0.00042742834,0.0000063535294,0.000103846636,0.00011575381,0.0000655575,0.000268874],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000037268004,0.00007347049,0.00059518765,0.000073175106,0.0000055315368,1.3379287e-7,0.0009183931,0.010944743,0.0019061917,0.0000033183555,0.97026974,0.015172837],"study_design_scores_gemma":[0.00054938404,0.0001770069,0.02866825,0.0017911256,0.00000520533,6.3630955e-7,0.0006597261,0.5774868,0.028305853,0.00028711866,0.36178765,0.0002812505],"about_ca_topic_score_codex":0.000002245634,"about_ca_topic_score_gemma":0.000009316389,"teacher_disagreement_score":0.7307568,"about_ca_system_score_codex":0.000025589603,"about_ca_system_score_gemma":0.000086138745,"threshold_uncertainty_score":0.99851954},"labels":[],"label_agreement":null},{"id":"W6958379795","doi":"10.6084/m9.figshare.13290662.v1","title":"Additional file 1 of Developing a coding taxonomy to analyze dental regulatory complaints","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; University of Toronto","funders":"","keywords":"Coding (social sciences); Reliability (semiconductor); Taxonomy (biology); Table (database); Data collection","score_opus":0.3230252684778103,"score_gpt":0.33593518630117825,"score_spread":0.012909917823367922,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6958379795","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00014790095,0.000005316052,0.000053456075,0.00034546186,0.000021264308,0.000308121,0.9950951,0.000019313557,0.0040041176],"genre_scores_gemma":[0.045239776,1.142504e-7,0.017244214,0.0031151683,0.00025537208,0.0018276954,0.93186873,0.000012812053,0.00043611167],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.997749,0.00007392329,0.00046222008,0.0003515109,0.0012044912,0.000158907],"domain_scores_gemma":[0.99682593,0.0022284032,0.00020094149,0.00022277239,0.00035887968,0.00016304172],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00018224971,0.00010036486,0.00020190384,0.00008211557,0.000108220825,0.00007641119,0.0005695676,0.00003751969,0.9962192],"category_scores_gemma":[0.020659372,0.000085940446,0.00010627683,0.00055963,0.0000114265595,0.00014438122,0.00025229345,0.000078923134,0.012992066],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000006845692,0.000010600812,0.000053609096,0.00001813448,0.000008555138,0.000002458452,0.0000455853,0.000071149414,0.00002437868,0.000011541001,0.99667794,0.003069193],"study_design_scores_gemma":[0.00006726864,0.000028688815,0.010576479,0.0008211927,0.0000013388287,0.0000010530664,0.00022799178,0.00044596862,0.00029162635,0.00007578002,0.9873519,0.0001107308],"about_ca_topic_score_codex":8.3936163e-7,"about_ca_topic_score_gemma":0.000009262806,"teacher_disagreement_score":0.9832272,"about_ca_system_score_codex":0.00006401347,"about_ca_system_score_gemma":0.00013570239,"threshold_uncertainty_score":0.98777646},"labels":[],"label_agreement":null},{"id":"W6958384052","doi":"10.6084/m9.figshare.24309955","title":"Additional file 1 of New approaches and technical considerations in detecting outlier measurements and trajectories in longitudinal children growth data","year":2023,"lang":"en","type":"article","venue":"Open MIND","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; University of Toronto","funders":"","keywords":"Outlier; Table (database); Anomaly detection; Cluster analysis; Measure (data warehouse); Lookup table; Section (typography); Type I and type II errors","score_opus":0.554596492596195,"score_gpt":0.40107526726793846,"score_spread":0.15352122532825657,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6958384052","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9004455,0.000095762385,0.0000709735,0.001679823,0.000075571836,0.0016356937,0.09083053,0.00000647415,0.0051596593],"genre_scores_gemma":[0.9799872,0.000003281043,0.016875528,0.000012085663,0.00003207532,0.00006338281,0.0027647477,0.000005270425,0.00025644657],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9979187,0.00013965038,0.00052490615,0.00054708996,0.0007190568,0.00015056756],"domain_scores_gemma":[0.99714404,0.0022679397,0.00012039504,0.00035269407,0.000054996064,0.00005990959],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0029940263,0.00009393738,0.0002241159,0.00017814746,0.000095680596,0.00024936142,0.00044845932,0.000052567255,0.10782245],"category_scores_gemma":[0.011406742,0.00007632736,0.00001515335,0.00050757255,0.0001242483,0.00049860554,0.00060915423,0.00012399512,0.00004057103],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000026582435,0.00014714307,0.5030606,0.0000053827484,0.000022202992,0.0000043139485,0.00055674376,0.00006416797,0.00008658508,0.000021971146,0.46668977,0.029314563],"study_design_scores_gemma":[0.0004346962,0.000041059713,0.9917748,0.00010934989,0.000006469429,0.00000943524,0.0005028766,0.00022054127,0.00015123823,0.004663342,0.0019752448,0.000110942034],"about_ca_topic_score_codex":0.0001335726,"about_ca_topic_score_gemma":0.0036837321,"teacher_disagreement_score":0.48871422,"about_ca_system_score_codex":0.000022312372,"about_ca_system_score_gemma":0.00016497559,"threshold_uncertainty_score":0.9969206},"labels":[],"label_agreement":null},{"id":"W6958470861","doi":"10.6084/m9.figshare.19335927.v1","title":"Additional file 1 of An observational cohort study of health outcomes and costs associated with early pregnancy assessment units in the UK","year":2022,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Economics; University of Alberta","funders":"","keywords":"Observational study; Cohort study; Early pregnancy factor; Pregnancy; Cohort; Retrospective cohort study","score_opus":0.36383295418599004,"score_gpt":0.4007051523165506,"score_spread":0.03687219813056053,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6958470861","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.20553157,0.000009649707,1.1216715e-7,0.00009771481,0.000008148373,0.0007485392,0.79332614,0.0000034735588,0.00027467308],"genre_scores_gemma":[0.67097694,9.484095e-8,0.000051405714,0.000117649404,0.0000027339795,0.0013910132,0.32735243,0.0000027591816,0.0001049938],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99547803,0.0010137017,0.00052375067,0.00024424848,0.0026273916,0.00011286924],"domain_scores_gemma":[0.9937813,0.0049087363,0.0005206963,0.00029115644,0.00046892496,0.000029134713],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0014491591,0.00007977929,0.00023602675,0.00007744216,0.00019200647,0.000039089082,0.00043717882,0.000015672184,0.8499918],"category_scores_gemma":[0.0059596677,0.00004913211,0.000021019867,0.0007586498,0.000012512605,0.00012971029,0.000120752346,0.00014419935,0.000005555289],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000004821994,0.0006054219,0.4024442,0.000004325756,0.00001854659,0.0000014005119,0.00047604137,0.00014711077,7.852894e-8,0.0000069279904,0.5955067,0.0007843813],"study_design_scores_gemma":[0.0003449467,0.0008621715,0.9916357,0.00043163076,0.0000024950455,3.5255948e-7,0.0028566888,0.00012292128,1.5503088e-7,0.000112945665,0.0035776151,0.0000524085],"about_ca_topic_score_codex":0.000075250486,"about_ca_topic_score_gemma":0.0005929497,"teacher_disagreement_score":0.84998626,"about_ca_system_score_codex":0.00010111476,"about_ca_system_score_gemma":0.0003411288,"threshold_uncertainty_score":0.71347094},"labels":[],"label_agreement":null},{"id":"W6958483370","doi":"10.6084/m9.figshare.12134709","title":"Additional file 1 of Prevalence of diabetic peripheral neuropathy in Africa: a systematic review and meta-analysis","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Peripheral neuropathy; Diabetes mellitus; Diabetic neuropathy; Scale (ratio); MEDLINE; Quality (philosophy)","score_opus":0.2885511541966848,"score_gpt":0.32672869211839756,"score_spread":0.038177537921712745,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6958483370","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000011584852,0.008796059,8.064744e-7,0.00029650677,0.0000015058151,0.0004987638,0.99001455,0.0000039122942,0.00037633604],"genre_scores_gemma":[0.37128854,0.00033200928,0.0028060842,0.008367175,0.00007815273,0.017865328,0.59031385,0.00005904657,0.00888977],"study_design_codex":"not_applicable","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9971484,0.00048778055,0.00089000043,0.00031049998,0.0010644643,0.00009888702],"domain_scores_gemma":[0.9954222,0.0033305902,0.0005039365,0.00035294303,0.00032597987,0.00006439532],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005099966,0.00009950377,0.0011379144,0.00007506794,0.000017357997,0.000022507023,0.00042824933,0.000022820906,0.993749],"category_scores_gemma":[0.04799211,0.00006246905,0.00044982793,0.001155523,0.000016290885,0.00010424475,0.00011305552,0.000067078436,0.00027529008],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000010755533,0.000027333925,0.000008786716,0.07005474,0.00088416267,0.0000012552825,0.00010210602,0.000032714473,0.0000017326214,6.952488e-7,0.92888075,0.0000046398354],"study_design_scores_gemma":[0.001123468,0.0019418411,0.050970394,0.20439763,0.47594166,0.000011533298,0.0013507092,0.020343194,0.00022453871,0.001757797,0.2394428,0.0024944304],"about_ca_topic_score_codex":5.268888e-7,"about_ca_topic_score_gemma":0.00000177424,"teacher_disagreement_score":0.9934737,"about_ca_system_score_codex":0.0000078793055,"about_ca_system_score_gemma":0.00003164824,"threshold_uncertainty_score":0.96002704},"labels":[],"label_agreement":null},{"id":"W6958634776","doi":"10.6084/m9.figshare.12072534","title":"Additional file 4 of Sedentary behavior patterns and adiposity in children: a study based on compositional data analysis","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Children's Hospital of Eastern Ontario","funders":"","keywords":"Table (database); Compositional data; Physical activity; Sedentary behavior; Sedentary lifestyle; Data file","score_opus":0.21652186579229066,"score_gpt":0.3554182507638054,"score_spread":0.13889638497151474,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6958634776","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.035370346,0.0000035516064,0.0000030031601,0.00014623922,0.0000040297023,0.00036263926,0.96403396,0.00000718003,0.00006902498],"genre_scores_gemma":[0.42110962,2.8460859e-8,0.00012208793,0.00033507033,0.000019757468,0.00013378407,0.5782725,0.0000018195972,0.000005291174],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9973815,0.00019597397,0.00038762408,0.0005466415,0.0013909882,0.000097235046],"domain_scores_gemma":[0.9974363,0.0016901215,0.00015040506,0.00051974505,0.0001138551,0.00008958279],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00022347386,0.00009990648,0.00022792666,0.00015622367,0.00006335331,0.00006551029,0.0006843503,0.000029985475,0.98675007],"category_scores_gemma":[0.0022579813,0.0000824278,0.000073031806,0.0006404683,0.000010071615,0.00015204356,0.00029682694,0.000117842035,0.00031873782],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00001935814,0.0005593097,0.20479894,0.0000048008037,0.0000464984,0.0000054957227,0.000029968616,0.00027207294,0.0000010923917,5.903264e-8,0.7937841,0.0004782917],"study_design_scores_gemma":[0.00029209498,0.00018824816,0.9856691,0.0001363259,0.000053163792,3.636579e-7,0.000074686766,0.008511269,0.000005890246,0.0000041556314,0.004979851,0.00008490222],"about_ca_topic_score_codex":0.000029752437,"about_ca_topic_score_gemma":0.0001571515,"teacher_disagreement_score":0.9864313,"about_ca_system_score_codex":0.000018387089,"about_ca_system_score_gemma":0.000045930836,"threshold_uncertainty_score":0.40968361},"labels":[],"label_agreement":null},{"id":"W6958772513","doi":"10.6084/m9.figshare.14833317.v1","title":"Additional file 1 of How to analyze work productivity loss due to health problems in randomized controlled trials? A simulation study","year":2021,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; University of British Columbia","funders":"","keywords":"Work (physics); Standard deviation; Binomial distribution; Productivity; Calibration; Distribution (mathematics)","score_opus":0.21223548743212642,"score_gpt":0.39869921775178757,"score_spread":0.18646373031966115,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6958772513","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0020442724,0.00006804464,0.000085482236,0.003812179,0.00007327256,0.008414527,0.98529494,0.000016053504,0.00019121727],"genre_scores_gemma":[0.73580825,4.4830125e-7,0.0011620373,0.00047412954,0.00029342668,0.015793076,0.24398553,0.000014802557,0.0024683238],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98916197,0.0058163875,0.0018593677,0.0007115364,0.0022011132,0.00024965536],"domain_scores_gemma":[0.92618716,0.06950907,0.0013220115,0.0008062173,0.0019615497,0.00021398849],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.014958236,0.00017438176,0.0028241752,0.00031118758,0.00008787892,0.00021343984,0.00033394867,0.000048370504,0.94166356],"category_scores_gemma":[0.5770684,0.00011338587,0.000492332,0.0020480058,0.00000940151,0.000180106,0.00015007168,0.00012805431,0.00063569454],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0068398523,0.0006676972,0.000045021417,0.000026389023,0.000066901644,0.0000048238508,0.00055994873,0.04251528,0.0000022052147,5.329373e-7,0.94481766,0.004453666],"study_design_scores_gemma":[0.2965261,0.0011131855,0.074775405,0.012457931,0.00009097893,0.0000027070137,0.0027176712,0.008000516,0.000085743675,0.0029432885,0.600315,0.0009714564],"about_ca_topic_score_codex":0.0000073969745,"about_ca_topic_score_gemma":0.00036704095,"teacher_disagreement_score":0.9410279,"about_ca_system_score_codex":0.00010858782,"about_ca_system_score_gemma":0.0005060042,"threshold_uncertainty_score":0.8170779},"labels":[],"label_agreement":null},{"id":"W6959055379","doi":"10.7939/r3-rsd0-2690","title":"Comparison of the concordance between clinical and histopathologic diagnosis of oral mucosal lesions in an Oral Medicine graduate program and the Oral Pathology biopsy service.","year":2024,"lang":"en","type":"dissertation","venue":"University of Alberta Library","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Concordance; Medical diagnosis; Oral medicine; Gold standard (test); Biopsy; Histopathology; Oral and maxillofacial pathology; Disease; Incidence (geometry)","score_opus":0.20804367192973114,"score_gpt":0.4069750722572452,"score_spread":0.19893140032751408,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6959055379","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9910834,0.0006299653,0.000002936,0.006563483,0.00035649588,0.00067674863,0.00003185188,0.000011478951,0.0006436689],"genre_scores_gemma":[0.99808997,0.00033071378,0.00067565765,0.00007866105,0.00002000708,0.000002425038,0.00011551965,0.00000912509,0.00067789905],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99607265,0.0015004809,0.0010235888,0.0005501357,0.0006889595,0.00016418395],"domain_scores_gemma":[0.99601877,0.0023742116,0.00080239645,0.00055047905,0.00017118662,0.00008294038],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0022761906,0.00021469181,0.0011574264,0.00020507822,0.00012065668,0.00002505463,0.001084315,0.00026616806,0.00009914767],"category_scores_gemma":[0.0006321565,0.00012152704,0.00016601662,0.00070982135,0.0019404744,0.00030593047,0.00037105268,0.00050271925,0.0000023976106],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00044280136,0.00027738686,0.9642874,0.0001466837,0.000033199685,0.000008514116,0.01077128,0.0000048632414,0.000015297752,0.0008952174,0.0017335045,0.021383874],"study_design_scores_gemma":[0.0014441009,0.00064077653,0.97048736,0.0003821158,0.0002751217,0.0000018672015,0.015953982,0.0003972676,0.00005620568,0.008294068,0.0019264993,0.00014061108],"about_ca_topic_score_codex":0.0023664632,"about_ca_topic_score_gemma":0.008569971,"teacher_disagreement_score":0.021243263,"about_ca_system_score_codex":0.000010436523,"about_ca_system_score_gemma":0.00013717667,"threshold_uncertainty_score":0.7149759},"labels":[],"label_agreement":null},{"id":"W6961256053","doi":"10.1371/journal.pone.0315590.s009","title":"Study quality evaluation via the Newcastle-Ottawa scale.","year":2024,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Quality (philosophy); Data collection; Quality assurance; Data quality; Quality assessment","score_opus":0.605830071897582,"score_gpt":0.5189129255716202,"score_spread":0.08691714632596181,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6961256053","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.65937144,0.007780004,0.00045021554,0.027634725,0.006659507,0.01753688,0.12967977,0.0010262764,0.1498612],"genre_scores_gemma":[0.9965472,3.9819656e-7,0.000018631152,0.00024008492,0.00023298948,0.0003738336,0.0013653671,0.0000074823447,0.0012140083],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9932215,0.001215617,0.0006375255,0.00055632746,0.004177013,0.00019202434],"domain_scores_gemma":[0.99633783,0.0018754367,0.000107658925,0.0009318531,0.00068208313,0.00006514375],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.011328247,0.00012766683,0.00015449613,0.00007666134,0.00023724954,0.0008049204,0.00085514324,0.000047541605,0.30312476],"category_scores_gemma":[0.011266147,0.00006623574,0.00014323395,0.00075586076,0.000012806269,0.00028621848,0.00018127327,0.00019642724,0.020561824],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000051140187,0.00014195408,0.0024520943,0.00002043067,0.000020032867,0.0000037485095,0.0016535477,0.00025144016,0.00004638847,0.000020982347,0.85773367,0.1376506],"study_design_scores_gemma":[0.00038293863,0.0002125771,0.31017205,0.00041076916,0.00004915665,0.000004445281,0.0046681645,0.01354529,0.0002229492,0.015363231,0.6546639,0.0003045262],"about_ca_topic_score_codex":0.000025609394,"about_ca_topic_score_gemma":0.00027657655,"teacher_disagreement_score":0.3371758,"about_ca_system_score_codex":0.00009124476,"about_ca_system_score_gemma":0.00012442561,"threshold_uncertainty_score":0.9970624},"labels":[],"label_agreement":null},{"id":"W6976708570","doi":"10.6084/m9.figshare.12233396","title":"NOS","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Scale (ratio); Field (mathematics); Process (computing)","score_opus":0.5584769089976009,"score_gpt":0.40391553303996286,"score_spread":0.15456137595763808,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6976708570","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011571144,0.00086269685,0.00023050002,0.09447224,0.00067485886,0.0019049539,0.3900353,0.0006240234,0.49962428],"genre_scores_gemma":[0.9841181,6.7430597e-7,0.00026530505,0.009994936,0.00039132722,0.000060365903,0.0042890906,0.000008153838,0.0008720193],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9982058,0.00006000025,0.00024510923,0.00027568324,0.0010942183,0.00011917998],"domain_scores_gemma":[0.9990261,0.00032145125,0.000067589215,0.000267979,0.0001858871,0.0001309778],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00025531813,0.00006129949,0.000100324476,0.000018960736,0.000058541762,0.0001460236,0.0006046065,0.000030054343,0.68612653],"category_scores_gemma":[0.017098676,0.000041870346,0.000076153294,0.0003180307,0.0000033491615,0.00012891178,0.00013032988,0.000079056146,0.08712512],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000029298178,0.000008563244,0.00023921505,0.000006163445,0.0000017019398,0.0000027304022,0.00012049738,0.000046363257,0.000084901745,0.000018571931,0.99030215,0.009166219],"study_design_scores_gemma":[0.00008520784,0.000037376238,0.004114649,0.000046399015,8.5778163e-7,2.6057074e-7,0.00007786632,0.0006778191,0.0007813898,0.0009414135,0.9931691,0.00006764705],"about_ca_topic_score_codex":5.569714e-7,"about_ca_topic_score_gemma":0.0000010910662,"teacher_disagreement_score":0.972547,"about_ca_system_score_codex":0.000009050245,"about_ca_system_score_gemma":0.000031689004,"threshold_uncertainty_score":0.9911807},"labels":[],"label_agreement":null},{"id":"W6976956266","doi":"10.6084/m9.figshare.11960187.v1","title":"Additional file 2 of Development, validation and application of a 3D printed model depicting adenoid hypertrophy in comparison to a Nasoendoscopy","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Flowchart; 3d printed; Adenoid hypertrophy; Sampling (signal processing); Muscle hypertrophy","score_opus":0.1782430192798449,"score_gpt":0.3467289463942152,"score_spread":0.1684859271143703,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6976956266","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.012645319,0.000014146878,0.0018294853,0.00016455649,0.0000049940177,0.00049820106,0.9842595,0.00001428595,0.000569498],"genre_scores_gemma":[0.69376475,2.0095064e-7,0.05160801,0.00019793733,0.00002390263,0.0008620992,0.25350273,0.0000074888912,0.00003290904],"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99839485,0.000049130675,0.0005557037,0.00026729805,0.0006458032,0.00008722333],"domain_scores_gemma":[0.99833375,0.0009307295,0.00026544186,0.00013300135,0.0002752167,0.00006188086],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00014659065,0.000071126524,0.00018198669,0.00008658647,0.000035334197,0.000022367778,0.00021581573,0.000033190634,0.5080384],"category_scores_gemma":[0.009821159,0.00006245068,0.000023123814,0.00042742834,0.0000063535294,0.000103846636,0.00011575381,0.0000655575,0.000268874],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000037268004,0.00007347049,0.00059518765,0.000073175106,0.0000055315368,1.3379287e-7,0.0009183931,0.010944743,0.0019061917,0.0000033183555,0.97026974,0.015172837],"study_design_scores_gemma":[0.00054938404,0.0001770069,0.02866825,0.0017911256,0.00000520533,6.3630955e-7,0.0006597261,0.5774868,0.028305853,0.00028711866,0.36178765,0.0002812505],"about_ca_topic_score_codex":0.000002245634,"about_ca_topic_score_gemma":0.000009316389,"teacher_disagreement_score":0.7307568,"about_ca_system_score_codex":0.000025589603,"about_ca_system_score_gemma":0.000086138745,"threshold_uncertainty_score":0.99851954},"labels":[],"label_agreement":null},{"id":"W6977002126","doi":"10.6084/m9.figshare.26600314","title":"Additional file 9 of MetaPro: a scalable and reproducible data processing and analysis pipeline for metatranscriptomic investigation of microbial communities","year":2024,"lang":"en","type":"dataset","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; SickKids Foundation; Hospital for Sick Children","funders":"","keywords":"Pipeline (software); Table (database); Scalability; Set (abstract data type); Data set; Data processing","score_opus":0.29670691952791073,"score_gpt":0.3689802987462378,"score_spread":0.07227337921832705,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6977002126","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000007570013,0.0013034911,0.000004431414,0.00007868861,0.000026063295,0.00041134644,0.998146,0.0000075444814,0.000014873382],"genre_scores_gemma":[0.00001682417,0.000017144894,0.0022273536,0.000050314226,0.00006141305,0.0003235678,0.9970755,0.000007700523,0.0002201608],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99727714,0.00017275978,0.0009141704,0.0007021431,0.00079930015,0.00013446705],"domain_scores_gemma":[0.9938827,0.0031214317,0.0006831182,0.0014580219,0.00079899374,0.00005570787],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0015111443,0.0001968589,0.0006757199,0.00049229653,0.00011917949,0.00023591217,0.0009183388,0.00012811855,0.8231011],"category_scores_gemma":[0.011780261,0.00015249097,0.00013477026,0.0008998932,0.000105798645,0.000360078,0.00048264558,0.00019180552,0.00009852731],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000016656908,0.00002572252,0.0000017171849,0.0019072503,0.00021142722,2.2823083e-7,0.00006778204,0.0000107335945,0.000022997683,9.811007e-8,0.996966,0.000769379],"study_design_scores_gemma":[0.000095327625,0.000033344277,0.00005704546,0.0029599555,0.0006753496,0.0000012963452,0.00022221521,0.0034074695,0.00012870283,0.0004733795,0.9918034,0.000142518],"about_ca_topic_score_codex":0.000049425926,"about_ca_topic_score_gemma":0.00055441604,"teacher_disagreement_score":0.8230026,"about_ca_system_score_codex":0.000013935041,"about_ca_system_score_gemma":0.0002786355,"threshold_uncertainty_score":0.99654394},"labels":[],"label_agreement":null},{"id":"W6977218444","doi":"10.6084/m9.figshare.24309955.v1","title":"Additional file 1 of New approaches and technical considerations in detecting outlier measurements and trajectories in longitudinal children growth data","year":2023,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; University of Toronto","funders":"","keywords":"Outlier; Table (database); Anomaly detection; Cluster analysis; Measure (data warehouse); Lookup table; Section (typography); Type I and type II errors","score_opus":0.5566174924804334,"score_gpt":0.3690764838847402,"score_spread":0.18754100859569317,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6977218444","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006449631,0.000060742033,0.0000015176693,0.00023575856,0.000011496917,0.00029079916,0.99251616,0.00002047844,0.00041344322],"genre_scores_gemma":[0.6510483,0.0000016898855,0.0016950506,0.000030502935,0.000064322165,0.00024538048,0.34681663,0.000008500933,0.00008961682],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9980454,0.00010054051,0.000443289,0.0004647029,0.0007893073,0.00015679016],"domain_scores_gemma":[0.9962159,0.0032048284,0.000117860785,0.00032010188,0.000081213715,0.000060141832],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00078289345,0.000098132965,0.000188823,0.00019930296,0.00008237781,0.000109309076,0.00026662904,0.00006224171,0.6151086],"category_scores_gemma":[0.047348652,0.00008171226,0.000020501433,0.0005344325,0.000030703573,0.00032622914,0.00034765623,0.00014205625,0.000092994516],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000037643379,0.000028886467,0.035218433,0.000012000025,0.00000553456,0.0000015681875,0.000077589146,0.000024862646,0.0000054861525,0.000004512318,0.96332586,0.0012915231],"study_design_scores_gemma":[0.00023983073,0.000026082285,0.99207306,0.0005403739,0.0000026741425,0.0000075262483,0.00013813867,0.00031569385,0.000040556086,0.003364395,0.0031486254,0.00010307052],"about_ca_topic_score_codex":0.000030020286,"about_ca_topic_score_gemma":0.0010159353,"teacher_disagreement_score":0.96017724,"about_ca_system_score_codex":0.000022871975,"about_ca_system_score_gemma":0.00012679871,"threshold_uncertainty_score":0.96067595},"labels":[],"label_agreement":null},{"id":"W6977232621","doi":"10.6084/m9.figshare.c.3911695_d7.v1","title":"Additional file 7: of Barriers and facilitators to the implementation of a school-based physical activity policy in Canada: application of the theoretical domains framework","year":2017,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Coding (social sciences); Data collection; Measure (data warehouse); Context (archaeology); Work (physics); Key (lock)","score_opus":0.04988355207348834,"score_gpt":0.3659384260501186,"score_spread":0.3160548739766303,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6977232621","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.122949824,9.988651e-7,0.000018602514,0.0016355346,0.000008241715,0.0003469684,0.87488097,9.366978e-7,0.00015792757],"genre_scores_gemma":[0.99490726,4.9421406e-8,0.00008109921,0.00009089694,0.000024547773,0.00031431168,0.004575954,0.000001932239,0.000003954889],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99839985,0.00015057954,0.00022715563,0.00017293874,0.00094865187,0.00010081363],"domain_scores_gemma":[0.9959438,0.0029058137,0.0003132138,0.00059620175,0.0001725045,0.00006846407],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00029164102,0.00006412784,0.00013863237,0.00003711974,0.0001383545,0.000024331071,0.00059932953,0.000026898002,0.5210727],"category_scores_gemma":[0.030323736,0.00003559734,0.000056679575,0.00021059027,0.000093406175,0.000069488146,0.00014017119,0.000098325756,0.00001574283],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000029270861,0.0000406836,0.005960369,0.000028924702,0.000008087355,8.291882e-8,0.00022586147,0.0001785947,0.00007530016,0.0014577566,0.9788288,0.013166268],"study_design_scores_gemma":[0.00013526347,0.000041545816,0.95515937,0.0002986312,0.0000029062112,7.384339e-8,0.0008939524,0.00091327,0.001936885,0.009053132,0.031506263,0.000058690093],"about_ca_topic_score_codex":0.028282076,"about_ca_topic_score_gemma":0.09956644,"teacher_disagreement_score":0.949199,"about_ca_system_score_codex":0.000114540126,"about_ca_system_score_gemma":0.0011652507,"threshold_uncertainty_score":0.9781887},"labels":[],"label_agreement":null},{"id":"W6977304791","doi":"10.6084/m9.figshare.19335927","title":"Additional file 1 of An observational cohort study of health outcomes and costs associated with early pregnancy assessment units in the UK","year":2022,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Economics; University of Alberta","funders":"","keywords":"Observational study; Cohort study; Early pregnancy factor; Pregnancy; Cohort; Retrospective cohort study","score_opus":0.36383295418599004,"score_gpt":0.4007051523165506,"score_spread":0.03687219813056053,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6977304791","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.20553157,0.000009649707,1.1216715e-7,0.00009771481,0.000008148373,0.0007485392,0.79332614,0.0000034735588,0.00027467308],"genre_scores_gemma":[0.67097694,9.484095e-8,0.000051405714,0.000117649404,0.0000027339795,0.0013910132,0.32735243,0.0000027591816,0.0001049938],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99547803,0.0010137017,0.00052375067,0.00024424848,0.0026273916,0.00011286924],"domain_scores_gemma":[0.9937813,0.0049087363,0.0005206963,0.00029115644,0.00046892496,0.000029134713],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0014491591,0.00007977929,0.00023602675,0.00007744216,0.00019200647,0.000039089082,0.00043717882,0.000015672184,0.8499918],"category_scores_gemma":[0.0059596677,0.00004913211,0.000021019867,0.0007586498,0.000012512605,0.00012971029,0.000120752346,0.00014419935,0.000005555289],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000004821994,0.0006054219,0.4024442,0.000004325756,0.00001854659,0.0000014005119,0.00047604137,0.00014711077,7.852894e-8,0.0000069279904,0.5955067,0.0007843813],"study_design_scores_gemma":[0.0003449467,0.0008621715,0.9916357,0.00043163076,0.0000024950455,3.5255948e-7,0.0028566888,0.00012292128,1.5503088e-7,0.000112945665,0.0035776151,0.0000524085],"about_ca_topic_score_codex":0.000075250486,"about_ca_topic_score_gemma":0.0005929497,"teacher_disagreement_score":0.84998626,"about_ca_system_score_codex":0.00010111476,"about_ca_system_score_gemma":0.0003411288,"threshold_uncertainty_score":0.71347094},"labels":[],"label_agreement":null},{"id":"W6977890608","doi":"10.6084/m9.figshare.12476702","title":"Additional file 1 of Using multiple agreement methods for continuous repeated measures data: a tutorial for practitioners","year":2020,"lang":"en","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Measure (data warehouse); Data file; Data collection; Term (time)","score_opus":0.6110904157845687,"score_gpt":0.4802538677778531,"score_spread":0.13083654800671557,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6977890608","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000046099476,0.00003099837,0.009500506,0.000330173,0.00018427489,0.0013028845,0.98837024,0.000026268732,0.00025005656],"genre_scores_gemma":[0.00025189997,2.428838e-7,0.19273658,0.00026789226,0.00063038705,0.0012723894,0.8046659,0.000011508184,0.00016316683],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99697274,0.0002695734,0.0007846205,0.00067457295,0.001069267,0.00022921393],"domain_scores_gemma":[0.976388,0.02017554,0.0007664432,0.0006909906,0.0018369481,0.0001421],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0013488733,0.00015339255,0.00036152906,0.00006211853,0.00018865541,0.00013671965,0.00083606184,0.00008822398,0.9149437],"category_scores_gemma":[0.283754,0.00012494957,0.00020213854,0.00030912863,0.000019294743,0.00042027226,0.0002315963,0.000087115404,0.00019293901],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017147289,0.00006486032,0.0000024123863,0.000059263875,0.00004783434,3.8678473e-7,0.000075157186,0.00020985604,0.0005260927,0.000003221664,0.98350763,0.015331811],"study_design_scores_gemma":[0.00062935957,0.00014773168,0.000034789457,0.00034664114,0.000021318509,6.9135376e-7,0.00024406986,0.042291574,0.0007697892,0.00035574634,0.9550139,0.00014442774],"about_ca_topic_score_codex":0.000005518309,"about_ca_topic_score_gemma":0.000009090574,"teacher_disagreement_score":0.91475075,"about_ca_system_score_codex":0.000052779586,"about_ca_system_score_gemma":0.0002837599,"threshold_uncertainty_score":0.72227925},"labels":[],"label_agreement":null},{"id":"W7000381172","doi":"","title":"On estimation of the four-parameter kappa distribution","year":2000,"lang":"en","type":"other","venue":"Library and Archives Canada (Government of Canada)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"","keywords":"Distribution (mathematics); Estimation; Kappa; Measure (data warehouse); Cohen's kappa","score_opus":0.016595010817677364,"score_gpt":0.19865254961267975,"score_spread":0.1820575387950024,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7000381172","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006962571,0.00024151341,0.00030029606,0.0033327295,0.0004710721,0.00041541253,0.001221241,0.000007707248,0.98704743],"genre_scores_gemma":[0.64361143,0.00014756498,0.0003859514,0.0008221175,0.00005795167,0.000009820785,0.000017497116,0.000048067013,0.3548996],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9936667,0.00019194648,0.00049358734,0.00031578582,0.005150193,0.00018175937],"domain_scores_gemma":[0.9980591,0.00079478894,0.00048544942,0.0005564456,6.782146e-7,0.00010353546],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000054490385,0.00019737873,0.00032023742,0.000029085351,0.00009694014,0.000030520503,0.00058158016,0.00005220889,0.00070803886],"category_scores_gemma":[0.000054926877,0.0001242443,0.000069876754,0.00014490835,0.00013622419,0.000095795556,0.00010627787,0.00014516665,2.1828185e-8],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00053987774,0.00013734546,0.0064250175,0.00025639913,0.00016953841,0.0000127829135,0.000053000695,0.0031765671,0.0006866749,0.059923846,0.79436433,0.13425462],"study_design_scores_gemma":[0.0005854911,0.00018411563,0.038804516,0.0012330142,0.00006879216,0.0000027390797,0.00035247553,0.005834,0.021032397,0.042751484,0.88862956,0.0005214397],"about_ca_topic_score_codex":0.0015196169,"about_ca_topic_score_gemma":0.011337034,"teacher_disagreement_score":0.63664883,"about_ca_system_score_codex":0.0000075708317,"about_ca_system_score_gemma":0.00071789714,"threshold_uncertainty_score":0.77525294},"labels":[],"label_agreement":null},{"id":"W7001214307","doi":"","title":"Interobserver agreement in grading of colorectal cancer - findings from a nation-wide web-based survey of histopathologists","year":2009,"lang":"en","type":"article","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Population and Public Health","funders":"","keywords":"Colorectal cancer; Grading (engineering); Neoplasm staging; MEDLINE","score_opus":0.19801129263271033,"score_gpt":0.3727435666499713,"score_spread":0.174732274017261,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7001214307","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9967881,0.00007545256,0.0009125906,0.00086109055,0.0002785485,0.00027133976,0.00006331414,0.000009819865,0.00073976786],"genre_scores_gemma":[0.9981423,0.00001064736,0.0013948373,0.0002911979,0.0000058974833,0.000011796541,0.00001307304,0.0000029696466,0.00012730001],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99621856,0.00048252306,0.001205001,0.00043919787,0.0014573814,0.00019731019],"domain_scores_gemma":[0.9970989,0.00167021,0.00037026845,0.00034331813,0.00046606045,0.00005123444],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.007153096,0.00014082811,0.00043414117,0.00029281166,0.00003705083,0.000032536227,0.0006300295,0.000075402626,0.0010983277],"category_scores_gemma":[0.0039067823,0.000103153085,0.00011054758,0.0006698572,0.00011520087,0.00015599307,0.000052468822,0.00010932154,0.00001319307],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013877984,0.0002931797,0.9678939,0.0000048818215,0.000006113698,0.0000018758121,0.0002555515,0.0010709168,0.025224984,0.00007873024,0.0024924753,0.0025386128],"study_design_scores_gemma":[0.0007845115,0.0001923529,0.95297027,0.00007850562,0.0000053874687,5.4455988e-8,0.00008098315,0.0032064829,0.040102903,0.0023767524,0.000096747826,0.00010504256],"about_ca_topic_score_codex":0.0092616165,"about_ca_topic_score_gemma":0.02648293,"teacher_disagreement_score":0.017221313,"about_ca_system_score_codex":0.00024950344,"about_ca_system_score_gemma":0.00014661132,"threshold_uncertainty_score":0.9998148},"labels":[],"label_agreement":null},{"id":"W7017358622","doi":"","title":"Arrhythmia risk associated with the use of bronchodilators in patients with chronic obstructive pulmonary disease : cohort studies and methodological issues","year":2008,"lang":"en","type":"dissertation","venue":"eScholarship@McGill (McGill)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"McGill University","keywords":"COPD; Ipratropium bromide; Cohort; Pulmonary disease; Cohort study; Disease; Tiotropium bromide; Cardiac arrhythmia","score_opus":0.1250337899961993,"score_gpt":0.33541205763739645,"score_spread":0.21037826764119716,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7017358622","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99500597,0.0016557217,4.896069e-7,0.000047124322,0.0002955499,0.0018768067,0.00077269005,0.000042535445,0.00030312326],"genre_scores_gemma":[0.9965444,0.0013420279,0.00063238514,0.000039350365,0.0000252495,0.00018673888,0.0001741477,0.00005540804,0.0010002429],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9885746,0.003994437,0.0014063236,0.0016038393,0.0038370017,0.0005837839],"domain_scores_gemma":[0.9892698,0.005648073,0.0018901695,0.0010706823,0.0018646157,0.00025665533],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0050750514,0.0007793591,0.0014754988,0.0003788547,0.0010541881,0.00011580542,0.0007473721,0.00034247636,0.000051261148],"category_scores_gemma":[0.019119723,0.0004192569,0.0002279439,0.001238661,0.0006639911,0.00094495394,0.00019902983,0.001375522,0.000009826841],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003657413,0.00051121606,0.9080123,0.00016417447,0.0013433364,0.00014233484,0.00012071926,0.0019452735,0.00007785662,0.0006459235,0.00002102215,0.08335848],"study_design_scores_gemma":[0.0009248716,0.0004917977,0.98633945,0.0005797926,0.0004534009,0.0000024705407,0.00064429943,0.000027248057,0.00028757282,0.0072819227,0.0023866158,0.0005805826],"about_ca_topic_score_codex":0.000244522,"about_ca_topic_score_gemma":0.0022255538,"teacher_disagreement_score":0.082777895,"about_ca_system_score_codex":0.00088212406,"about_ca_system_score_gemma":0.0001150352,"threshold_uncertainty_score":0.99982595},"labels":[],"label_agreement":null},{"id":"W7017895069","doi":"","title":"Comparison of Two Dependent within Subject Coefficients of Variation to Evaluate the Reproducibility of Measurement Devices","year":2008,"lang":"en","type":"article","venue":"Scholarship@Western (Western University)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Reproducibility; Reliability (semiconductor); Coefficient of variation; Statistical power; Statistical hypothesis testing; Sample size determination; Monte Carlo method; Sample (material); Regression analysis","score_opus":0.45195550238889165,"score_gpt":0.4384273986723294,"score_spread":0.013528103716562268,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7017895069","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99187386,0.00009065498,0.006329924,0.00022227729,0.00038678639,0.0009315687,0.000024753255,0.00002049493,0.00011969181],"genre_scores_gemma":[0.9995444,0.0000056423623,0.00015119601,0.00005553634,0.000023441804,0.000001864631,0.0000022166134,0.000010796863,0.00020489578],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9885284,0.0022342727,0.0017246435,0.0013400242,0.0058463914,0.00032628907],"domain_scores_gemma":[0.9921418,0.0005716234,0.001225548,0.002884802,0.0030171126,0.00015910564],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02812497,0.0002651707,0.0007625134,0.0005020488,0.000273324,0.00006442656,0.002103525,0.00009829628,0.000043994052],"category_scores_gemma":[0.0039678398,0.00019495426,0.00024116148,0.001524548,0.00031021977,0.0006245185,0.00054929894,0.00030480445,0.00004247333],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00034295427,0.00064811826,0.9618928,0.000049653478,0.000080041966,0.0000032379753,0.004593458,0.008500188,0.02339866,0.000049223716,0.0000014969808,0.0004401625],"study_design_scores_gemma":[0.0009303809,0.000365075,0.8633426,0.00014485608,0.00010540932,0.0000024974636,0.00087348715,0.000012869058,0.13371056,0.00029066452,0.000058806014,0.00016278744],"about_ca_topic_score_codex":0.00006387161,"about_ca_topic_score_gemma":0.0027779979,"teacher_disagreement_score":0.1103119,"about_ca_system_score_codex":0.0002738958,"about_ca_system_score_gemma":0.00031128738,"threshold_uncertainty_score":0.9747609},"labels":[],"label_agreement":null},{"id":"W7034376527","doi":"","title":"Validation of water vapour profiles (version 13) retrieved by the IMK/IAA scientific retrieval processor based on full resolution spectra measrued by MIPAS on board ENVISAT","year":2009,"lang":"en","type":"article","venue":"JuSER (Forschungszentrum Jülich)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Canadian Space Agency; Ministry of Education, India; Bundesministerium für Bildung und Forschung","keywords":"Water vapor; Resolution (logic); Calibration; Spectral line; On board; High resolution","score_opus":0.045279761892022624,"score_gpt":0.29059089321575293,"score_spread":0.2453111313237303,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7034376527","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9723782,0.00010089527,0.0006619549,0.022229552,0.0009470488,0.0017154498,0.00016985694,0.00010461957,0.0016924174],"genre_scores_gemma":[0.9946399,0.000014654778,0.00017013613,0.0009803824,0.00014515882,0.000019707977,0.00027379647,0.000029486766,0.003726789],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9887156,0.00088520657,0.0015149947,0.001499195,0.006462428,0.0009226063],"domain_scores_gemma":[0.995924,0.00057348266,0.00073176256,0.0017896882,0.00072412094,0.0002569423],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.010282926,0.0005228323,0.0005887184,0.00035407243,0.0010034038,0.00078548666,0.0016133959,0.00030552558,0.0009875889],"category_scores_gemma":[0.002264877,0.00029769185,0.0003745292,0.0014362674,0.00036682928,0.0006451864,0.000114945746,0.00052815233,0.0007630271],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0041941376,0.0013828198,0.0016103414,0.000048981896,0.00003809197,0.0000027384192,0.00087242824,0.0017301901,0.59812456,0.00018788694,0.3901327,0.0016751043],"study_design_scores_gemma":[0.0019468789,0.0018733173,0.0036218371,0.00015387901,0.000081208455,0.0000015994472,0.0004905038,0.0055606794,0.93559974,0.0037219108,0.04647129,0.00047716257],"about_ca_topic_score_codex":0.00003405476,"about_ca_topic_score_gemma":0.000007673869,"teacher_disagreement_score":0.3436614,"about_ca_system_score_codex":0.00039630526,"about_ca_system_score_gemma":0.00014707173,"threshold_uncertainty_score":0.99994755},"labels":[],"label_agreement":null},{"id":"W7036192077","doi":"","title":"Bibliometric Analysis on GABA-A Receptors Research Based on CiteSpace and VOSviewer","year":2023,"lang":"en","type":"article","venue":"DOAJ (DOAJ: Directory of Open Access Journals)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Bibliometrics; Web of science; Scientometrics; Chinese academy of sciences; DECIPHER; Anxiety","score_opus":0.7385120060079285,"score_gpt":0.674947508126148,"score_spread":0.06356449788178054,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7036192077","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9796246,0.0033674506,0.00029998997,0.0034124728,0.00086692505,0.0008420016,0.00005723848,0.000052397285,0.011476909],"genre_scores_gemma":[0.9884474,0.008009584,0.00013520026,0.0005255976,0.00013069557,0.00005781691,0.000009721564,0.00003076886,0.0026531734],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98381364,0.0027583123,0.0015720263,0.0012698128,0.009833203,0.0007530324],"domain_scores_gemma":[0.9846726,0.01023805,0.00083531317,0.0015846755,0.0021257717,0.0005436062],"candidate_categories":["metaresearch","metaepi_narrow","bibliometrics","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","bibliometrics"],"category_scores_codex":[0.053732943,0.00035977789,0.0011506629,0.20118225,0.00072796835,0.005019012,0.005063482,0.00015771888,0.026083222],"category_scores_gemma":[0.022773858,0.0002538884,0.00046659433,0.3987013,0.00038738307,0.0015089881,0.0012007914,0.00081293593,0.0005473729],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023364347,0.00034116188,0.62777877,0.000023749624,0.00021472051,0.000017882607,0.0000804624,0.0046688733,0.002846856,0.000045268644,0.30096704,0.06278161],"study_design_scores_gemma":[0.00042179087,0.00007895175,0.92903733,0.00021267642,0.00010028963,3.7718405e-7,0.00021002672,0.0023476803,0.003442742,0.008730695,0.055100106,0.00031735006],"about_ca_topic_score_codex":0.00030083707,"about_ca_topic_score_gemma":0.0000730469,"teacher_disagreement_score":0.30125856,"about_ca_system_score_codex":0.00018652482,"about_ca_system_score_gemma":0.0001772415,"threshold_uncertainty_score":0.99999136},"labels":[],"label_agreement":null},{"id":"W7095506909","doi":"","title":"Selective and Automatic editing with CADI-applications","year":2011,"lang":"en","type":"article","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Benchmark (surveying); Selection (genetic algorithm); Inclusion (mineral); Baseline (sea); Image editing","score_opus":0.16645321318107173,"score_gpt":0.3371844289019,"score_spread":0.17073121572082825,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7095506909","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.46609598,0.000044015756,0.10483201,0.00077640027,0.000078293175,0.0008653816,0.0000026299776,0.00012674664,0.42717856],"genre_scores_gemma":[0.9839182,0.0000021177402,0.0151096275,0.00014691569,0.00003393359,0.00006508358,2.1197576e-7,0.0000027289796,0.00072117784],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9987575,0.0000549738,0.0002277129,0.00025330676,0.0005897107,0.000116805924],"domain_scores_gemma":[0.99918616,0.00021715029,0.000079638754,0.0002484687,0.00020526575,0.000063310545],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0014019111,0.00006236889,0.00009675981,0.00006582142,0.00014122804,0.00006536526,0.00018433653,0.000020875039,0.0007571006],"category_scores_gemma":[0.00022224714,0.000034409462,0.000015876989,0.00035202815,0.00008053612,0.00015655818,0.00003456796,0.000055209344,0.00013922436],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00002689083,0.00046885517,0.57098615,0.00003482631,0.00008819351,0.0000029818361,0.009651006,0.000023923427,0.00078711496,0.039715003,0.01576084,0.3624542],"study_design_scores_gemma":[0.00080684037,0.0005828157,0.7424203,0.000054346427,0.000052452022,0.000020820493,0.01074589,0.006111228,0.008773034,0.20470104,0.025253711,0.0004774844],"about_ca_topic_score_codex":0.00006625045,"about_ca_topic_score_gemma":0.00012213342,"teacher_disagreement_score":0.5178222,"about_ca_system_score_codex":0.000015205228,"about_ca_system_score_gemma":0.000024287401,"threshold_uncertainty_score":0.82897204},"labels":[],"label_agreement":null},{"id":"W7099412294","doi":"","title":"Contributors","year":2003,"lang":"en","type":"article","venue":"","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Christian ministry; Work (physics); Government (linguistics); Agriculture; Local government","score_opus":0.23506219794522168,"score_gpt":0.40688131997094334,"score_spread":0.17181912202572167,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7099412294","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2335203,0.00008271044,0.027645344,0.0018422415,0.0007391292,0.00017566979,0.0000014217048,0.000041852367,0.7359513],"genre_scores_gemma":[0.98412794,0.000001994559,0.000972574,0.0005848903,0.000013580122,0.0000032394573,1.2346821e-7,0.0000013291319,0.0142943505],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.99806714,0.00018579555,0.0002937157,0.0002034912,0.001110812,0.00013907127],"domain_scores_gemma":[0.9988997,0.0004408642,0.00004264944,0.0003595754,0.00019271483,0.00006450605],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0055926656,0.00004854195,0.00009575339,0.000042899464,0.00007226313,0.000107249056,0.00024324699,0.000023162518,0.009851176],"category_scores_gemma":[0.0059510153,0.000027452905,0.000059567512,0.00026923415,0.00003109352,0.00009286884,0.000014503498,0.000039394636,0.0024052402],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000076239107,0.0001612286,0.17544173,0.000001058353,0.000012099638,0.0000028712702,0.000114024275,0.00011481338,0.0014855799,0.5942506,0.20787418,0.020534137],"study_design_scores_gemma":[0.00018339226,0.000025298388,0.0068552103,9.799003e-7,0.000001520821,8.761463e-7,0.00026235677,0.000035359124,0.0053564743,0.15785795,0.82935935,0.000061249964],"about_ca_topic_score_codex":0.000004213007,"about_ca_topic_score_gemma":0.000007263727,"teacher_disagreement_score":0.7506076,"about_ca_system_score_codex":0.000017646786,"about_ca_system_score_gemma":0.000024524306,"threshold_uncertainty_score":0.9983715},"labels":[],"label_agreement":null},{"id":"W7110938138","doi":"10.1371/journal.pntd.0013073.s002","title":"NEWCASTLE—OTTAWA QUALITY ASSESSMENT SCALE for cohort studies.","year":2025,"lang":"","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Cohort; Scale (ratio); Quality (philosophy); Quality assessment; Cohort study","score_opus":0.5112718175797546,"score_gpt":0.5431478145066313,"score_spread":0.031875996926876704,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7110938138","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006358691,0.015466295,0.002066069,0.021598179,0.008601935,0.015660819,0.8157962,0.00026045626,0.114191376],"genre_scores_gemma":[0.72175556,0.0008244325,0.012676185,0.016662683,0.0024364733,0.017874528,0.07689792,0.00016450642,0.15070769],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9900101,0.00094995793,0.0028268762,0.0019405015,0.0033113922,0.0009611872],"domain_scores_gemma":[0.98494226,0.007443336,0.00093518954,0.0022003613,0.004186006,0.0002928462],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.009340777,0.0006013661,0.0013373438,0.00029039432,0.0010872976,0.00086544204,0.002044106,0.00033903844,0.19436932],"category_scores_gemma":[0.034384985,0.00049246044,0.00082420785,0.0014482269,0.000098090466,0.00048357635,0.0010093445,0.00047912914,0.0025611147],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000048246733,0.0004596512,0.012196396,0.0010238162,0.0004311484,0.0000020593325,0.00022423721,0.00015368867,0.000048745685,0.00030841003,0.9430762,0.0420274],"study_design_scores_gemma":[0.0012945823,0.00023378734,0.093816705,0.0042132223,0.00017115446,4.7700104e-7,0.0022558556,0.0013114923,0.0005758543,0.0097933365,0.88576496,0.00056859304],"about_ca_topic_score_codex":0.00002654902,"about_ca_topic_score_gemma":0.00019135336,"teacher_disagreement_score":0.7388983,"about_ca_system_score_codex":0.00088220835,"about_ca_system_score_gemma":0.001279442,"threshold_uncertainty_score":0.9997527},"labels":[],"label_agreement":null},{"id":"W7111126593","doi":"10.1371/journal.pone.0322293.s003","title":"Newcastle Ottawa scale (NOS) based quality assessment of included cohort studies.","year":2025,"lang":"","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Cohort; Scale (ratio); Quality assessment; Cohort study; Quality (philosophy)","score_opus":0.39541674345228983,"score_gpt":0.5128512836640952,"score_spread":0.11743454021180533,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7111126593","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06445217,0.013707635,0.0012715679,0.015272026,0.006276993,0.011803365,0.650837,0.00028530174,0.23609395],"genre_scores_gemma":[0.9845662,0.000057442197,0.0015389486,0.0018150979,0.00013359779,0.0005677841,0.007049204,0.000022959986,0.0042487374],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9874121,0.0018597511,0.0035536143,0.0015817098,0.00486796,0.0007248697],"domain_scores_gemma":[0.98606366,0.005583245,0.001459812,0.0025239068,0.004109404,0.00025996813],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.009374518,0.00057320006,0.0016027146,0.00043513722,0.0005940072,0.00032290455,0.0022091137,0.0003313142,0.27147603],"category_scores_gemma":[0.026764458,0.00047414776,0.0007012317,0.0023593875,0.00016854207,0.00039264114,0.0013484915,0.0005636783,0.0010971734],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000059357197,0.0013172287,0.08515667,0.0018386795,0.00052502967,0.000005921248,0.00027249634,0.0014033472,0.00042496336,0.0001701675,0.89306104,0.015765078],"study_design_scores_gemma":[0.0024664265,0.00040216133,0.5855315,0.0113687925,0.00025913888,4.2195902e-7,0.0024304418,0.009121517,0.0076808794,0.0028558793,0.3770362,0.00084663933],"about_ca_topic_score_codex":0.00009836394,"about_ca_topic_score_gemma":0.0002730428,"teacher_disagreement_score":0.92011404,"about_ca_system_score_codex":0.00071200595,"about_ca_system_score_gemma":0.0023724684,"threshold_uncertainty_score":0.999771},"labels":[],"label_agreement":null},{"id":"W7111223050","doi":"10.1371/journal.pone.0325723.s002","title":"Methodological quality assessment of 14 included studies using the Newcastle-Ottawa quality assessment scale.","year":2025,"lang":"","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Quality assessment; Quality (philosophy); Risk assessment; Quality assurance; Conformity assessment; Quantitative assessment","score_opus":0.8977663646866403,"score_gpt":0.6523932223448586,"score_spread":0.24537314234178165,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7111223050","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.32417533,0.030140538,0.045938022,0.054638825,0.014494965,0.021736136,0.4195766,0.0005123807,0.08878722],"genre_scores_gemma":[0.9618341,0.00018512852,0.03129048,0.0022214737,0.00031917842,0.0005073909,0.0019634091,0.00003061473,0.00164827],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.95753235,0.025068209,0.006565836,0.002262959,0.0074291136,0.0011415394],"domain_scores_gemma":[0.93530947,0.048653033,0.004376615,0.004455375,0.0068524857,0.0003530393],"candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.058654774,0.0008782174,0.0028080218,0.0003396498,0.001663435,0.0006855537,0.0036183165,0.00053333066,0.08035503],"category_scores_gemma":[0.0833898,0.0005323274,0.0012949271,0.0027168612,0.00052237604,0.00053727085,0.004161048,0.0014070697,0.00012155779],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005823129,0.007829572,0.14075652,0.010270585,0.00484163,0.000032822696,0.006597658,0.017166061,0.013734067,0.013414996,0.6562522,0.12852155],"study_design_scores_gemma":[0.002709627,0.00066413166,0.80679214,0.009980191,0.00053390145,0.000004010527,0.028928274,0.011551227,0.003980526,0.041248985,0.092079274,0.001527692],"about_ca_topic_score_codex":0.00018978467,"about_ca_topic_score_gemma":0.00026008778,"teacher_disagreement_score":0.66603565,"about_ca_system_score_codex":0.0012468771,"about_ca_system_score_gemma":0.002142347,"threshold_uncertainty_score":0.9997128},"labels":[],"label_agreement":null},{"id":"W7126640062","doi":"","title":"Polytomous regression did not outperform dichotomous logistic regression in diagnosing serious bacterial infections in febrile children","year":2008,"lang":"","type":"article","venue":"EUR Research Repository (Erasmus University Rotterdam)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Pediatric Oncology Group","funders":"","keywords":"Logistic regression; Polytomous Rasch model; Regression analysis; Regression; Cross-sectional regression; Linear regression","score_opus":0.1637317756781169,"score_gpt":0.3677283708463741,"score_spread":0.20399659516825722,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7126640062","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9877211,0.00041913913,0.000062844454,0.0014236389,0.0026713836,0.003053095,0.0000690464,0.000107411724,0.004472339],"genre_scores_gemma":[0.991195,0.0016326355,0.00011210193,0.00007660988,0.0006999827,0.000027070233,0.00003583839,0.00006080379,0.0061599216],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.97883576,0.0074477177,0.0021431677,0.002885231,0.0063145044,0.0023736353],"domain_scores_gemma":[0.9921193,0.002723995,0.0007607497,0.0023978616,0.001062709,0.0009353729],"candidate_categories":["metaepi_narrow","sts","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.009276715,0.0009307538,0.0013885904,0.003833515,0.004723284,0.0008305797,0.0028416992,0.0008094263,0.0005216665],"category_scores_gemma":[0.0034079622,0.0008458691,0.0005180325,0.004619079,0.0020495418,0.0020595489,0.0022007348,0.003730192,0.00038091608],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0042940183,0.00412978,0.93738335,0.000111223024,0.0000817052,0.008294836,0.006013031,0.0012073278,0.02459845,0.000034719833,0.0066921446,0.007159427],"study_design_scores_gemma":[0.010489625,0.0022846125,0.9557827,0.0027512007,0.000053566,0.0008620191,0.0022582605,0.0019971167,0.012895496,0.00016573741,0.009032333,0.0014272959],"about_ca_topic_score_codex":0.0114202,"about_ca_topic_score_gemma":0.0022214388,"teacher_disagreement_score":0.018399399,"about_ca_system_score_codex":0.003963689,"about_ca_system_score_gemma":0.0016561344,"threshold_uncertainty_score":0.9998599},"labels":[],"label_agreement":null},{"id":"W7135411397","doi":"10.6084/m9.figshare.5531614","title":"Additional file 7: of Barriers and facilitators to the implementation of a school-based physical activity policy in Canada: application of the theoretical domains framework","year":2017,"lang":"","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Coding (social sciences); Data collection; Measure (data warehouse); Context (archaeology); Work (physics); Key (lock)","score_opus":0.04273739852549995,"score_gpt":0.3551625362591402,"score_spread":0.3124251377336402,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7135411397","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.087862745,0.0000048277134,0.00002396161,0.002802677,0.000024264997,0.000880288,0.90824217,0.0000011292531,0.00015794988],"genre_scores_gemma":[0.98928356,3.8671394e-7,0.00009408365,0.00013595939,0.000056608864,0.0006293654,0.009788989,0.000004819821,0.00000620105],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9967153,0.00038908218,0.0005310366,0.00038033992,0.0017488657,0.00023538926],"domain_scores_gemma":[0.9918572,0.005605332,0.0008300262,0.0011611211,0.0003730011,0.00017331875],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005202174,0.00016340514,0.00031977287,0.00007454879,0.00036189563,0.00006268309,0.0012060762,0.000074279225,0.78640634],"category_scores_gemma":[0.04831533,0.000099319484,0.00014059631,0.00044405268,0.0002906821,0.00014782614,0.00035990146,0.00025440974,0.00003179113],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009948199,0.00013470941,0.0049936967,0.0001427322,0.000029444682,2.3548772e-7,0.00060409366,0.0003750041,0.00008482704,0.0016404788,0.9630029,0.028892353],"study_design_scores_gemma":[0.00035028308,0.00014595593,0.9331137,0.0017050242,0.000014678274,1.9297467e-7,0.0026763014,0.003265757,0.0028958821,0.008070998,0.047593437,0.00016777051],"about_ca_topic_score_codex":0.06549466,"about_ca_topic_score_gemma":0.16080534,"teacher_disagreement_score":0.92812,"about_ca_system_score_codex":0.00032331602,"about_ca_system_score_gemma":0.004358845,"threshold_uncertainty_score":0.9597011},"labels":[],"label_agreement":null},{"id":"W7135411763","doi":"10.6084/m9.figshare.5531614.v1","title":"Additional file 7: of Barriers and facilitators to the implementation of a school-based physical activity policy in Canada: application of the theoretical domains framework","year":2017,"lang":"","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Coding (social sciences); Data collection; Measure (data warehouse); Context (archaeology); Work (physics); Key (lock)","score_opus":0.04273739852549995,"score_gpt":0.3551625362591402,"score_spread":0.3124251377336402,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7135411763","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.087862745,0.0000048277134,0.00002396161,0.002802677,0.000024264997,0.000880288,0.90824217,0.0000011292531,0.00015794988],"genre_scores_gemma":[0.98928356,3.8671394e-7,0.00009408365,0.00013595939,0.000056608864,0.0006293654,0.009788989,0.000004819821,0.00000620105],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9967153,0.00038908218,0.0005310366,0.00038033992,0.0017488657,0.00023538926],"domain_scores_gemma":[0.9918572,0.005605332,0.0008300262,0.0011611211,0.0003730011,0.00017331875],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005202174,0.00016340514,0.00031977287,0.00007454879,0.00036189563,0.00006268309,0.0012060762,0.000074279225,0.78640634],"category_scores_gemma":[0.04831533,0.000099319484,0.00014059631,0.00044405268,0.0002906821,0.00014782614,0.00035990146,0.00025440974,0.00003179113],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009948199,0.00013470941,0.0049936967,0.0001427322,0.000029444682,2.3548772e-7,0.00060409366,0.0003750041,0.00008482704,0.0016404788,0.9630029,0.028892353],"study_design_scores_gemma":[0.00035028308,0.00014595593,0.9331137,0.0017050242,0.000014678274,1.9297467e-7,0.0026763014,0.003265757,0.0028958821,0.008070998,0.047593437,0.00016777051],"about_ca_topic_score_codex":0.06549466,"about_ca_topic_score_gemma":0.16080534,"teacher_disagreement_score":0.92812,"about_ca_system_score_codex":0.00032331602,"about_ca_system_score_gemma":0.004358845,"threshold_uncertainty_score":0.9597011},"labels":[],"label_agreement":null},{"id":"W7155592100","doi":"","title":"RAPID-TEST (Rapid respiratory microbiological point-of-care-testing in primary care): Statistical Analysis Plan","year":2024,"lang":"en","type":"book","venue":"Bristol Research (University of Bristol)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Infection and Immunity","funders":"","keywords":"Statistical analysis; Primary care; Plan (archaeology); MEDLINE","score_opus":0.2515944795773787,"score_gpt":0.36885111426384026,"score_spread":0.11725663468646158,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7155592100","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09690136,0.04971807,0.003397998,0.0009580524,0.0007670167,0.004567804,0.009199443,0.00021436009,0.8342759],"genre_scores_gemma":[0.7258789,0.0023771094,0.02541073,0.000316771,0.00065851875,0.000019774869,0.0041221296,0.00024923024,0.24096683],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9884044,0.00215805,0.0014975809,0.0019631523,0.0051033795,0.00087347376],"domain_scores_gemma":[0.9841393,0.009793829,0.000762136,0.0016570241,0.0032630083,0.00038465427],"candidate_categories":["metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.012766559,0.00047609158,0.0020249186,0.0038759816,0.0004133645,0.00016314768,0.0032928453,0.00072121236,0.0013192644],"category_scores_gemma":[0.0064087557,0.0004787651,0.0008157582,0.0037387472,0.002721455,0.0002496219,0.0017518897,0.0021132382,0.00043811373],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019675537,0.0014631335,0.05774032,0.006338406,0.0017214323,0.003674285,0.009096038,0.00033532197,0.003583946,0.0020176168,0.61274475,0.2993172],"study_design_scores_gemma":[0.0024187944,0.0046548964,0.091402054,0.0027620308,0.00148217,0.000016495545,0.028051568,0.0005147685,0.00012526152,0.011841279,0.854917,0.0018136796],"about_ca_topic_score_codex":0.0008534838,"about_ca_topic_score_gemma":0.0015187507,"teacher_disagreement_score":0.62897754,"about_ca_system_score_codex":0.002753006,"about_ca_system_score_gemma":0.0031522429,"threshold_uncertainty_score":0.99999255},"labels":[],"label_agreement":null}]}