{"meta":{"query_hash":"bfc1c98d030d","filters":{"venue":"Evaluation Review"},"cohort_total":27,"direct_labels_cover":1,"predictions_cover":27,"exported":27,"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/bfc1c98d030d","api":"https://metacan.xera.ac/api/v1/cohort?venue=Evaluation+Review"},"results":[{"id":"W15985522","doi":"10.1177/0193841x05276654","title":"The construction of social imaginaries on terrorism in newspapers from Montreal and Guadalajara : a comparative study among three cultural groups","year":2008,"lang":"en","type":"dissertation","venue":"Evaluation Review","topic":"Cultural and political discourse analysis","field":"Social 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":"Terrorism; Newspaper; Construct (python library); Media studies; Political science; The Imaginary; Reading (process); Sociology; Order (exchange); Law; Psychology","score_opus":0.07795567414794159,"score_gpt":0.4310312717312078,"score_spread":0.35307559758326623,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W15985522","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":[0.956735,0.022020515,1.7555476e-7,0.002149258,0.0001851312,0.0015133711,0.000019599844,0.000010118162,0.017366838],"genre_scores_gemma":[0.9779975,0.02034815,0.000006882835,0.00005044251,0.00017544902,0.00018342552,0.00045512803,0.0000054192124,0.0007776321],"study_design_codex":"qualitative","study_design_gemma":"observational","domain_scores_codex":[0.9972064,0.0010039958,0.0004975985,0.00026121977,0.0008629038,0.00016789716],"domain_scores_gemma":[0.9988496,0.00020312943,0.00041688446,0.00010490163,0.0003771471,0.00004836514],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007462905,0.00018836877,0.00056231715,0.000041910906,0.00065308413,0.00007595771,0.00014138382,0.00009724654,0.00022699946],"category_scores_gemma":[0.00026930467,0.00011710216,0.0001501446,0.00029736923,0.00047989012,0.00020904666,0.000010613031,0.00021638973,0.000008429128],"study_design_candidate":"qualitative","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":[0.00025960166,0.0005492279,0.033598103,0.00044676356,0.0010260993,0.000007009699,0.69731945,0.000011442919,0.00001896401,0.023554523,0.007796692,0.23541215],"study_design_scores_gemma":[0.00038369984,0.00004776244,0.8121269,0.0007408524,0.00093502883,1.8877103e-7,0.1800536,0.00006220776,0.0000018270754,0.0043554665,0.0010970754,0.0001954067],"about_ca_topic_score_codex":0.028701842,"about_ca_topic_score_gemma":0.33635193,"teacher_disagreement_score":0.77852875,"about_ca_system_score_codex":0.000116011244,"about_ca_system_score_gemma":0.00016186155,"threshold_uncertainty_score":0.9777661},"labels":[],"label_agreement":null},{"id":"W1983148067","doi":"10.1177/0193841x0002400103","title":"An Application of Theory-Driven Evaluation to a Drop-In Youth Center","year":2000,"lang":"en","type":"review","venue":"Evaluation Review","topic":"Youth Development and Social Support","field":"Social Sciences","cited_by":30,"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 Montréal; McGill University","funders":"","keywords":"Drop out; Context (archaeology); Theory of change; Intervention (counseling); Program evaluation; Data collection; Psychology; Center (category theory); Management science; Applied psychology; Sociology; Political science; Engineering; Social science; Economics; Public administration","score_opus":0.1513130385005866,"score_gpt":0.490324462403093,"score_spread":0.3390114239025064,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1983148067","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":[0.000017401622,0.9658009,0.00014435685,0.00010901751,0.00017709103,0.010188374,0.000053349184,0.000033408218,0.023476116],"genre_scores_gemma":[0.0005752715,0.995446,0.00014632611,0.0002272315,0.00019676788,0.0015434364,0.0016713905,0.00003151152,0.00016201285],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9896031,0.006053203,0.0013394618,0.00055232627,0.0020941067,0.00035778826],"domain_scores_gemma":[0.99795395,0.00011410355,0.000710221,0.00047046097,0.0005873983,0.00016384617],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.021642206,0.00029247013,0.0013613058,0.00027328776,0.00018541468,0.000039774113,0.00055073196,0.0002762719,0.0035822936],"category_scores_gemma":[0.0007442302,0.00026771525,0.0003280238,0.0014025111,0.00007198304,0.0002743365,0.00002974352,0.00018292328,0.00047774994],"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":[0.000002028353,0.00007074104,0.000064719476,0.0026796064,0.000022254502,1.18278074e-7,0.0045247376,0.000001997716,1.6424385e-8,0.0010674968,0.00012240498,0.9914439],"study_design_scores_gemma":[0.00019774842,0.000024325935,0.00009975346,0.025783466,0.0015716567,2.4297498e-7,0.00021980974,0.000044480097,3.0175915e-8,0.0007170102,0.9709998,0.00034171395],"about_ca_topic_score_codex":0.00019237453,"about_ca_topic_score_gemma":0.00049584405,"teacher_disagreement_score":0.99110216,"about_ca_system_score_codex":0.0007743801,"about_ca_system_score_gemma":0.0023919686,"threshold_uncertainty_score":0.9999775},"labels":[],"label_agreement":null},{"id":"W1990843122","doi":"10.1177/0193841x07307532","title":"The Interaction of Co-Occurring Mental Disorders and Recovery Management Checkups on Substance Abuse Treatment Participation and Recovery","year":2008,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Substance Abuse Treatment and Outcomes","field":"Medicine","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":"Centre for Addiction and Mental Health","funders":"National Institute on Drug Abuse","keywords":"Substance abuse; Intervention (counseling); Psychiatry; Substance use; Mental health; Psychology; Clinical psychology; Randomized controlled trial; Substance abuse treatment; Medicine; Internal medicine","score_opus":0.08272507338278332,"score_gpt":0.3832698896296803,"score_spread":0.300544816246897,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1990843122","genre_codex":"empirical","genre_gemma":"review","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":[0.94327074,0.0533506,0.000009266117,0.0004695751,0.00012718127,0.0015850398,0.0000018307861,0.000015200399,0.0011705555],"genre_scores_gemma":[0.4513455,0.5478571,0.000081233,0.00022828755,0.000014316473,0.00019642644,0.000036073867,0.0000072107564,0.0002338486],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9989886,0.000108568755,0.00030794187,0.00018976763,0.00029792314,0.00010719617],"domain_scores_gemma":[0.99940073,0.00014322432,0.000174289,0.00019713407,0.000044880897,0.0000397487],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00040174005,0.00012998759,0.00024193576,0.000047329875,0.00012412318,0.000008014927,0.000021779266,0.000020880476,0.000060543458],"category_scores_gemma":[0.00003659648,0.000083253115,0.000068141,0.00010257563,0.000047990427,0.00013701871,0.000003114907,0.000037951628,0.000012366589],"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":[0.00058145274,0.00030736628,0.08114472,0.0004829185,0.00041309386,0.0000047090402,0.0025717795,0.00002977148,0.000050761624,0.00007809319,0.0010349104,0.9133004],"study_design_scores_gemma":[0.02218011,0.0046418654,0.85029155,0.021799698,0.0056766607,0.00008120426,0.0026123975,0.0034747415,0.004313145,0.00073109777,0.08331231,0.0008851928],"about_ca_topic_score_codex":0.0000039459965,"about_ca_topic_score_gemma":0.000051057235,"teacher_disagreement_score":0.9124152,"about_ca_system_score_codex":0.00020655517,"about_ca_system_score_gemma":0.000015999072,"threshold_uncertainty_score":0.33949634},"labels":[],"label_agreement":null},{"id":"W2001415027","doi":"10.1177/0193841x06296947","title":"A Repeated Observation Approach for Estimating the Street Homeless Population","year":2007,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Census and Population Estimation","field":"Mathematics","cited_by":38,"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":"Downtown; Population; Identification (biology); Estimation; Sampling (signal processing); Statistics; Visibility; Geography; Confidence interval; Econometrics; Demography; Psychology; Computer science; Mathematics; Sociology; Engineering","score_opus":0.26078212160035724,"score_gpt":0.44689302585433927,"score_spread":0.18611090425398202,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2001415027","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":[0.060638666,0.0035251558,0.92765695,0.00090749084,0.00029594044,0.0058751074,0.000012304529,0.00015499293,0.00093338743],"genre_scores_gemma":[0.5472116,0.00015899765,0.4492462,0.0005755618,0.00031462117,0.0004829002,0.0018788441,0.000037222384,0.000094061244],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9978265,0.00023193932,0.0008670551,0.00024585676,0.0006385429,0.00019010808],"domain_scores_gemma":[0.99770147,0.0005473845,0.0006670585,0.00039209967,0.0006523204,0.00003969224],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009854336,0.00015241155,0.00028818136,0.00005888226,0.00026826604,0.00003309076,0.00012368856,0.00006864298,0.00007283298],"category_scores_gemma":[0.0038112132,0.00011001456,0.000114891765,0.00043497753,0.000014042186,0.00020077804,0.000015420377,0.00008794235,0.00000445065],"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":[0.000029950435,0.00018455587,0.01592738,0.006773208,0.00006737549,2.1183911e-7,0.00042830262,0.0075394143,0.00004448927,0.079297446,0.007860454,0.8818472],"study_design_scores_gemma":[0.0006104342,0.00003022286,0.07606599,0.0011021057,0.00044297866,0.0000041430826,0.000040360013,0.88578665,0.000020651016,0.03466309,0.0010070802,0.00022630533],"about_ca_topic_score_codex":0.000026449303,"about_ca_topic_score_gemma":0.000046463036,"teacher_disagreement_score":0.8816209,"about_ca_system_score_codex":0.0001429924,"about_ca_system_score_gemma":0.00003691419,"threshold_uncertainty_score":0.45626536},"labels":[],"label_agreement":null},{"id":"W2015434985","doi":"10.1177/0193841x14567313","title":"An Alternative Framework to Investigating and Understanding Intraindividual Processes in Substance Abuse Recovery","year":2015,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Substance Abuse Treatment and Outcomes","field":"Medicine","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":"Simon Fraser University","funders":"","keywords":"Nomothetic and idiographic; Substance abuse; Craving; Psychology; Abstinence; Affect (linguistics); Clinical psychology; Addiction; Developmental psychology; Social psychology; Psychiatry","score_opus":0.28539059405864137,"score_gpt":0.4279131534136509,"score_spread":0.14252255935500951,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2015434985","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":[0.93557084,0.059764795,0.0010775994,0.0016210213,0.000096191965,0.0014618714,0.0000013506076,0.000037243146,0.00036908407],"genre_scores_gemma":[0.9621578,0.022114266,0.010519773,0.0048301257,0.00010655909,0.00018477763,0.000042684966,0.000020886053,0.000023156575],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9985747,0.00017157504,0.00030390004,0.00027065355,0.00052013615,0.00015907019],"domain_scores_gemma":[0.9990587,0.00021282857,0.00011413752,0.00018779641,0.00022325493,0.0002032525],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0017018163,0.00013673631,0.00032976674,0.00010811983,0.000030435054,0.000026704794,0.000058524638,0.000042939868,0.000060085433],"category_scores_gemma":[0.0019460886,0.00011217791,0.000020216681,0.0006285322,0.000033203454,0.00031103592,0.0000046881737,0.00011553906,0.00001915859],"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":[0.00022182794,0.0003412537,0.7392154,0.0036669062,0.00022513278,0.000063037354,0.16325645,0.00052293506,0.00008951478,0.0027746798,0.0019063529,0.08771653],"study_design_scores_gemma":[0.031220695,0.0054734484,0.17659321,0.33605936,0.0046749506,0.0002433903,0.061021138,0.007075288,0.0030719228,0.36680007,0.0040473575,0.003719165],"about_ca_topic_score_codex":0.000004576074,"about_ca_topic_score_gemma":0.00018499553,"teacher_disagreement_score":0.5626222,"about_ca_system_score_codex":0.00040556391,"about_ca_system_score_gemma":0.0002287174,"threshold_uncertainty_score":0.45744824},"labels":[],"label_agreement":null},{"id":"W2025194027","doi":"10.1177/0193841x12458103","title":"Measuring Stakeholder Participation in Evaluation","year":2012,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":30,"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; Université Laval","funders":"","keywords":"Intraclass correlation; Reliability (semiconductor); Stakeholder; Sample (material); Convergent validity; Psychology; Generalizability theory; Sample size determination; Scale (ratio); Statistics; Applied psychology; Psychometrics; Clinical psychology; Mathematics; Economics; Developmental psychology; Geography","score_opus":0.8475988642781547,"score_gpt":0.6249450003263772,"score_spread":0.22265386395177744,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2025194027","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":[0.6679611,0.187464,0.002532478,0.009724199,0.0025818646,0.008335941,0.000006760418,0.00008862073,0.121305004],"genre_scores_gemma":[0.99438477,0.0023743007,0.00033534778,0.0015767487,0.00014983848,0.0008956148,0.000044016375,0.000010488492,0.00022888386],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.98644745,0.0035799912,0.0015105358,0.00038788392,0.0076792208,0.00039490336],"domain_scores_gemma":[0.9963818,0.00048631395,0.000517922,0.00065549224,0.0018209134,0.00013756091],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.104222745,0.00015728186,0.00033885756,0.00031055574,0.00010964375,0.00009458444,0.00027786006,0.000057266698,0.024950214],"category_scores_gemma":[0.011957306,0.00012115324,0.000092047856,0.0014593095,0.00002389697,0.0015652125,0.000044175038,0.00013966387,0.0027040427],"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":[0.000005376884,0.00017200192,0.08618295,0.000083770836,0.000008492564,7.5323506e-8,0.0007156972,0.0021308307,0.00006354203,0.0008978101,0.005772951,0.9039665],"study_design_scores_gemma":[0.0011915267,0.00003484978,0.8213325,0.0009253917,0.00022461775,0.0000026840094,0.00023044505,0.12060682,0.00017534448,0.0048244963,0.050149556,0.00030178123],"about_ca_topic_score_codex":0.000008936398,"about_ca_topic_score_gemma":0.00006749839,"teacher_disagreement_score":0.9036647,"about_ca_system_score_codex":0.00041595098,"about_ca_system_score_gemma":0.00045854147,"threshold_uncertainty_score":0.99807245},"labels":[],"label_agreement":null},{"id":"W2034770421","doi":"10.1177/0193841x05279182","title":"Facilitating Self-Determination in Adolescents with Intellectual Disabilities","year":2005,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Disability Education and Employment","field":"Social 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":true,"ca_institutions":"Université du Québec à Trois-Rivières","funders":"","keywords":"Curriculum; Identity (music); Psychology; Self-determination; Intellectual disability; Field (mathematics); Pedagogy; Medical education; Applied psychology; Mathematics education; Political science; Medicine","score_opus":0.09025164283672198,"score_gpt":0.42462906028296526,"score_spread":0.3343774174462433,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2034770421","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":[0.89155287,0.019306548,0.00022822102,0.028483903,0.0003186532,0.0049523963,0.000002861554,0.00021907745,0.05493545],"genre_scores_gemma":[0.99543035,0.0023037062,0.0006892208,0.00084787403,0.00007784949,0.00024701047,0.0000069471184,0.0000050711396,0.00039194885],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99790275,0.0007831537,0.0003065381,0.00018700842,0.00065929256,0.00016125984],"domain_scores_gemma":[0.99927384,0.00024510798,0.00006875413,0.00013967237,0.00021492923,0.000057702095],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0035732181,0.000077543606,0.00013427097,0.000044982473,0.00012749867,0.00003568128,0.00010312224,0.000026336602,0.0019843339],"category_scores_gemma":[0.0059414366,0.00006775757,0.000033824195,0.00032717545,0.000094132374,0.00028382815,0.000010344178,0.00006699379,0.0002974708],"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":[0.000007304116,0.00049395656,0.018078852,0.0012278849,0.0000047732688,4.7452346e-8,0.17886709,0.000052451218,0.000001178165,0.00096900394,0.0016711324,0.7986263],"study_design_scores_gemma":[0.0023907968,0.00031039453,0.03754926,0.029048258,0.00030891248,0.0000027922795,0.25809845,0.010371348,0.000031873275,0.002583935,0.65803325,0.0012707282],"about_ca_topic_score_codex":0.00013734163,"about_ca_topic_score_gemma":0.002827264,"teacher_disagreement_score":0.7973556,"about_ca_system_score_codex":0.0006122967,"about_ca_system_score_gemma":0.00031741895,"threshold_uncertainty_score":0.998928},"labels":[],"label_agreement":null},{"id":"W2047617457","doi":"10.1177/0193841x02026004003","title":"The Development of Science Achievement in Middle and High Schoolr","year":2002,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Educational Environments and Student Outcomes","field":"Social Sciences","cited_by":68,"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; Canadian Institute for Advanced Research","funders":"","keywords":"Socioeconomic status; Multilevel model; Academic achievement; Context (archaeology); Psychology; Mathematics education; Autonomy; Science education; Demography; Sociology; Geography; Political science; Mathematics; Statistics","score_opus":0.14697315987348716,"score_gpt":0.41420795312102526,"score_spread":0.26723479324753807,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2047617457","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":[0.7474932,0.17109361,0.000022078168,0.04270848,0.00041417032,0.0019035704,7.193343e-7,0.000008044585,0.036356095],"genre_scores_gemma":[0.9453784,0.051746566,0.0010182989,0.0004151953,0.000017865925,0.00009268076,8.8208526e-7,0.0000017835478,0.0013283344],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9986176,0.00013354067,0.00022894815,0.0000918991,0.00083499105,0.0000930133],"domain_scores_gemma":[0.99968344,0.00007343141,0.00007987999,0.000077342316,0.000055831966,0.00003007712],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0061018225,0.000032195894,0.000066894965,0.000024546558,0.000441427,0.000019919977,0.00014031211,0.0000075091025,0.0005000825],"category_scores_gemma":[0.0005146426,0.000021569143,0.000009624521,0.00022777237,0.00025648193,0.00008718786,0.000027311655,0.000026747048,0.00002424927],"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":[8.2764626e-7,0.00024575047,0.07872472,0.0001772896,0.000013490786,5.5687746e-8,0.019838192,0.00000871645,0.000029054001,0.07124815,0.0011794299,0.8285343],"study_design_scores_gemma":[0.00016151556,0.000007820983,0.7566562,0.00046231213,0.000016379528,5.0302592e-8,0.0018519883,0.000053252592,0.000011830828,0.0010001657,0.2397086,0.00006986756],"about_ca_topic_score_codex":0.000029468138,"about_ca_topic_score_gemma":0.00022828995,"teacher_disagreement_score":0.82846445,"about_ca_system_score_codex":0.0001307115,"about_ca_system_score_gemma":0.00014707551,"threshold_uncertainty_score":0.54755527},"labels":[],"label_agreement":null},{"id":"W2065805300","doi":"10.1177/0193841x06293411","title":"Does Receiving an Earnings Supplement Affect Union Formation? Estimating Effects for Program Participants Using Propensity Score Matching","year":2006,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Gender, Labor, and Family Dynamics","field":"Social Sciences","cited_by":18,"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":"Propensity score matching; Receipt; Earnings; Matching (statistics); Disadvantaged; Affect (linguistics); Context (archaeology); Demographic economics; Random assignment; Average treatment effect; Psychology; Economics; Econometrics; Medicine; Accounting; Economic growth; Geography","score_opus":0.13375186684758997,"score_gpt":0.42968491301862294,"score_spread":0.295933046171033,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2065805300","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":[0.9271009,0.0018580318,0.061220545,0.00055012415,0.00059826206,0.008306177,0.00000671827,0.00016598399,0.00019327526],"genre_scores_gemma":[0.91992295,0.0004073479,0.078158915,0.00017793331,0.00031182452,0.0007438354,0.00021637206,0.000022048893,0.000038796657],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9966604,0.0014616837,0.0004892058,0.00029085533,0.0006599687,0.00043790054],"domain_scores_gemma":[0.9986667,0.0001811265,0.00043342015,0.00019093849,0.00043943367,0.000088381814],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0110513205,0.00015562186,0.0003067193,0.00005026462,0.0009921237,0.0001824539,0.00014602668,0.000059536425,0.00004394076],"category_scores_gemma":[0.0010393461,0.000115058276,0.000084314364,0.00031533767,0.000046828834,0.0007478382,0.00003129736,0.00009276564,0.000005804692],"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":[0.000024586525,0.0009662483,0.30116588,0.026228214,0.000110480316,0.0000035344517,0.03651638,0.023901196,0.0019608438,0.011553559,0.0014906701,0.5960784],"study_design_scores_gemma":[0.001755582,0.00043288778,0.33287528,0.015312954,0.0012317265,0.0000030840336,0.0025944293,0.6151956,0.00013768388,0.023500439,0.0058712084,0.0010891639],"about_ca_topic_score_codex":0.00081436767,"about_ca_topic_score_gemma":0.001430289,"teacher_disagreement_score":0.59498924,"about_ca_system_score_codex":0.00028773426,"about_ca_system_score_gemma":0.00020797554,"threshold_uncertainty_score":0.76307136},"labels":[],"label_agreement":null},{"id":"W2093553995","doi":"10.1177/0193841x03261484","title":"Picking Up the Challenge","year":2004,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Homelessness and Social Issues","field":"Health Professions","cited_by":14,"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 Ottawa; Child Welfare League of Canada; Royal Ottawa Mental Health Centre","funders":"","keywords":"Enumeration; Population; Computer science; Data science; Management science; Sociology; Mathematics; Engineering; Demography","score_opus":0.2812821955724736,"score_gpt":0.5541621417145391,"score_spread":0.27287994614206557,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2093553995","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":[0.043049347,0.80264527,0.00042226625,0.062232673,0.0024036043,0.0057361377,0.0000035381706,0.00014594331,0.08336125],"genre_scores_gemma":[0.8842191,0.11259202,0.000023959663,0.002004816,0.00031703777,0.000540075,0.000009717712,0.000009529813,0.00028371712],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99810106,0.00088075636,0.00033735082,0.00011895519,0.0003922992,0.00016958464],"domain_scores_gemma":[0.9991457,0.00013625863,0.00016149915,0.000245657,0.00027742173,0.000033489156],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0039427076,0.0000692418,0.00024500617,0.000012605617,0.0006200742,0.000004860008,0.00012243488,0.000054361073,0.0036730017],"category_scores_gemma":[0.00056700007,0.000041176405,0.000058443704,0.0001556098,0.000019760266,0.000060370374,0.000031863314,0.00022703609,0.0015245546],"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":[0.0000037755449,0.00009372372,0.00397015,0.008367574,0.00006661751,0.0000017891512,0.10223774,0.000037744412,0.000007466545,0.27022052,0.002682806,0.6123101],"study_design_scores_gemma":[0.0024697853,0.000077862955,0.035667162,0.03417498,0.00057383813,2.927097e-7,0.03554402,0.000066485714,0.000010046448,0.08293495,0.8079817,0.00049885537],"about_ca_topic_score_codex":0.000111088964,"about_ca_topic_score_gemma":0.00059403846,"teacher_disagreement_score":0.8411698,"about_ca_system_score_codex":0.00012862876,"about_ca_system_score_gemma":0.00028023284,"threshold_uncertainty_score":0.99925286},"labels":[],"label_agreement":null},{"id":"W2099407725","doi":"10.1177/0193841x0102500602","title":"Characteristics of Responding-, Nonresponding- and Refusing-Parents in an Adolescent Lifestyle Choice Study","year":2001,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Adolescent Sexual and Reproductive Health","field":"Health Professions","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":"Brock University; Queen's University","funders":"","keywords":"Parental consent; Informed consent; Psychology; Scale (ratio); Clinical psychology; Developmental psychology; Medicine; Alternative medicine","score_opus":0.2809426709750991,"score_gpt":0.5545025882861809,"score_spread":0.2735599173110818,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2099407725","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":[0.98445433,0.009123866,0.000026477164,0.0010264665,0.00043994936,0.004688606,0.000007672382,0.000027010692,0.00020559576],"genre_scores_gemma":[0.9846287,0.012620551,0.000046306315,0.0012233154,0.00029779307,0.00018473982,0.000045573932,0.00002400896,0.00092901685],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9928931,0.0045105955,0.0012105561,0.0005065288,0.0005418982,0.00033731482],"domain_scores_gemma":[0.9978727,0.00022408446,0.0006875162,0.00051249424,0.0005502262,0.00015298389],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009240041,0.0001807196,0.000587937,0.0002929891,0.00025948603,0.0000072254957,0.00013968743,0.00008785821,0.00030120678],"category_scores_gemma":[0.00254147,0.00014859423,0.000028811823,0.00079274795,0.000036502865,0.00025917005,0.000094573166,0.0004498969,0.000048972175],"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":[0.0001492905,0.0010583157,0.97514886,0.002907385,0.000006946454,0.0000031371308,0.0014816552,7.606637e-7,0.00009879198,0.000052692514,0.00017371375,0.018918443],"study_design_scores_gemma":[0.0010931145,0.00030215344,0.977722,0.0071975794,0.00008282934,0.0000020173559,0.0007319268,0.00010763018,0.0000014194418,0.000031818436,0.012587144,0.00014036418],"about_ca_topic_score_codex":0.00007560711,"about_ca_topic_score_gemma":0.00018177512,"teacher_disagreement_score":0.01877808,"about_ca_system_score_codex":0.00028479547,"about_ca_system_score_gemma":0.00032861735,"threshold_uncertainty_score":0.6059497},"labels":[],"label_agreement":null},{"id":"W2112067598","doi":"10.1177/0193841x0002400601","title":"The Evaluation of Information Campaigns to Promote Voluntary Household Water Conservation","year":2000,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Environmental Education and Sustainability","field":"Environmental Science","cited_by":246,"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":"Persuasion; Popularity; Summative assessment; Water conservation; Public relations; Marketing; Business; Environmental resource management; Psychology; Environmental economics; Political science; Social psychology; Formative assessment; Economics; Ecology; Water resources; Pedagogy","score_opus":0.03931785094118342,"score_gpt":0.31836773981942185,"score_spread":0.27904988887823845,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2112067598","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":[0.9716162,0.0016118372,0.00007630087,0.012182726,0.00011201078,0.0045010047,0.000006875099,0.000023942808,0.009869121],"genre_scores_gemma":[0.9887487,0.0020355561,0.0004205857,0.006955296,0.000023229712,0.00092492317,0.00019614458,0.000010877049,0.000684663],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9971328,0.00057454203,0.00054789166,0.0001679962,0.0014070168,0.00016979873],"domain_scores_gemma":[0.99924254,0.0000341724,0.00011603207,0.00043109708,0.00010515638,0.00007101958],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.008522069,0.000108014225,0.0001273912,0.000020518499,0.00019791746,0.000025365563,0.00017557717,0.00003353524,0.029099679],"category_scores_gemma":[0.0004146308,0.00006898178,0.000056866793,0.00024569055,0.000086922875,0.00054135907,0.00004205388,0.00006523972,0.0017140377],"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":[0.000014651482,0.000076084565,0.0043551438,0.000127443,0.000008371612,2.3168223e-8,0.0014514391,0.006513578,0.00017299343,0.000024058234,0.030023169,0.9572331],"study_design_scores_gemma":[0.00056370656,0.00009860824,0.33430982,0.00024479197,0.00028413284,0.0000032102382,0.0003605195,0.015172895,0.00096338737,0.002433717,0.64530855,0.00025667972],"about_ca_topic_score_codex":0.00013794338,"about_ca_topic_score_gemma":0.00003626954,"teacher_disagreement_score":0.95697635,"about_ca_system_score_codex":0.0006102493,"about_ca_system_score_gemma":0.00006443036,"threshold_uncertainty_score":0.99906325},"labels":[],"label_agreement":null},{"id":"W2112624513","doi":"10.1177/0193841x06293412","title":"Addressing Program Fidelity Using Onsite Observations and Program Provider Descriptions of Program Delivery","year":2006,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Youth Development and Social Support","field":"Social Sciences","cited_by":82,"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":"Fidelity; Program evaluation; Data collection; Quarter (Canadian coin); Computer science; Program Design Language; Population; Medical education; Medicine; Environmental health; Political science; Software engineering","score_opus":0.41773105899385166,"score_gpt":0.49081063493772575,"score_spread":0.0730795759438741,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2112624513","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":[0.88958275,0.072753444,0.00018368743,0.0024494627,0.00035883713,0.019893892,0.000028847207,0.0006886286,0.014060435],"genre_scores_gemma":[0.84454954,0.012200333,0.13819124,0.00048192262,0.0003642499,0.0031349757,0.0004073691,0.000035625955,0.00063473923],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9971701,0.000601035,0.00063551706,0.00025959098,0.0010418992,0.00029184698],"domain_scores_gemma":[0.99802387,0.000057362893,0.00035728814,0.00014916153,0.0013272027,0.000085113155],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0039338223,0.0001354788,0.00029415713,0.000061087856,0.0007281493,0.0001407482,0.000114127986,0.00009806615,0.00021703204],"category_scores_gemma":[0.00072425505,0.00013306394,0.00010576308,0.00077456026,0.0002839718,0.0005010892,0.00003760715,0.000102793805,0.0000038788335],"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":[0.0000022460147,0.00041137644,0.05159513,0.0006373801,0.000023833454,5.8791954e-7,0.0010751173,0.0000063296525,0.000035382003,0.0015071441,0.0016909795,0.9430145],"study_design_scores_gemma":[0.0008434724,0.00018190125,0.6391178,0.0075514233,0.0013333389,0.0000030698839,0.0011006619,0.005366595,0.000024602712,0.0048655258,0.3387843,0.00082734285],"about_ca_topic_score_codex":0.0016400192,"about_ca_topic_score_gemma":0.0014009987,"teacher_disagreement_score":0.94218713,"about_ca_system_score_codex":0.0002222355,"about_ca_system_score_gemma":0.0013242804,"threshold_uncertainty_score":0.56004095},"labels":[],"label_agreement":null},{"id":"W2113827930","doi":"10.1177/0193841x0002400501","title":"Health Outcomes of Elementary School Students in New Brunswick","year":2000,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Obesity, Physical Activity, Diet","field":"Medicine","cited_by":32,"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 Alberta; Canadian Institute for Advanced Research","funders":"","keywords":"Socioeconomic status; Psychology; Mental health; Physical health; Multilevel model; Healthy eating; Physical activity; Gerontology; Environmental health; Medicine; Physical therapy; Psychiatry; Population","score_opus":0.05501168353014899,"score_gpt":0.43534261547478337,"score_spread":0.38033093194463435,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2113827930","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":[0.9364591,0.04542128,0.0000022776621,0.013219347,0.000052827276,0.0028501528,0.000002143777,0.000021163683,0.0019716872],"genre_scores_gemma":[0.939469,0.043514516,0.00042467692,0.014366648,0.000094540395,0.000029843766,0.00007136937,0.00001796179,0.0020114384],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9977014,0.0003254355,0.00056928734,0.00020385388,0.0010400079,0.00015998915],"domain_scores_gemma":[0.9992254,0.000048513954,0.00015877342,0.00031930668,0.000080050886,0.00016799176],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0015893474,0.00011610148,0.00068086456,0.000053143205,0.000017433402,0.0000046582427,0.00011130136,0.000022531329,0.008334897],"category_scores_gemma":[0.0002105204,0.00009584516,0.0001092726,0.00029998767,0.000014163923,0.00010468967,0.000028042457,0.00013578952,0.00049863517],"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":[0.0000069921516,0.0019995752,0.82236516,0.0012596295,0.000039877803,6.562985e-7,0.000023683517,0.0000026091845,0.0000026469452,0.000055836445,0.00558662,0.16865668],"study_design_scores_gemma":[0.0014921288,0.00018563383,0.989897,0.0026971158,0.00015227197,0.0000012155882,0.0000049611176,0.000034190805,0.000013018332,0.0003604232,0.005094992,0.00006704271],"about_ca_topic_score_codex":0.0013529467,"about_ca_topic_score_gemma":0.00055732776,"teacher_disagreement_score":0.16858964,"about_ca_system_score_codex":0.00022735227,"about_ca_system_score_gemma":0.0009999975,"threshold_uncertainty_score":0.9925716},"labels":[],"label_agreement":null},{"id":"W2124187889","doi":"10.1177/0193841x05284090","title":"How to Begin a New Topic in Mathematics","year":2006,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Mathematics Education and Teaching Techniques","field":"Social Sciences","cited_by":10,"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":"Mathematics education; Computer science; Contrast (vision); Multilevel model; Mathematics","score_opus":0.12386235465861056,"score_gpt":0.4557607482728444,"score_spread":0.33189839361423384,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2124187889","genre_codex":"other","genre_gemma":"methods","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":[0.0041674683,0.03211106,0.07969458,0.2883002,0.00043430223,0.006902133,0.0000015050332,0.00040405017,0.58798474],"genre_scores_gemma":[0.32909364,0.017554399,0.48497364,0.011457508,0.0018463834,0.0016468551,0.00005735518,0.000069286456,0.15330094],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.9988692,0.00024247817,0.00021463046,0.0001066029,0.00045350037,0.00011363319],"domain_scores_gemma":[0.9995085,0.00007547003,0.000089971036,0.00018079669,0.00009063133,0.00005463135],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0034354343,0.00005834465,0.00014600015,0.000071159026,0.000066738445,0.000070019836,0.0001475111,0.000029780656,0.0008564319],"category_scores_gemma":[0.0016913651,0.000053796044,0.000035899007,0.00034236995,0.000012916883,0.00008065976,0.000012204508,0.00005380819,0.0000778363],"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":[1.5783355e-7,0.0000917476,0.00016696447,0.0003922849,0.0000017520232,1.85785e-7,0.004447951,0.0000013629459,0.000009687007,0.7345206,0.1733908,0.086976476],"study_design_scores_gemma":[0.00006917994,0.000006779366,0.00026454387,0.0013713534,0.000022256581,3.573014e-7,0.0002074795,0.000087881286,0.000015589529,0.3725164,0.62534004,0.00009811601],"about_ca_topic_score_codex":0.00054005114,"about_ca_topic_score_gemma":0.0013238363,"teacher_disagreement_score":0.45194924,"about_ca_system_score_codex":0.00013135035,"about_ca_system_score_gemma":0.00025269957,"threshold_uncertainty_score":0.9377328},"labels":[],"label_agreement":null},{"id":"W2131308570","doi":"10.1177/0193841x03254349","title":"Outpatient Marijuana Treatment for Adolescents","year":2003,"lang":"en","type":"review","venue":"Evaluation Review","topic":"Prenatal Substance Exposure Effects","field":"Medicine","cited_by":60,"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":"National Institute on Drug Abuse; U.S. Public Health Service","keywords":"Economic evaluation; Economic cost; Economic analysis; Substance abuse; Intervention (counseling); Drug treatment; Quarter (Canadian coin); Medicine; Treatment and control groups; Cost–benefit analysis; Substance abuse treatment; Cost analysis; Total cost; Environmental health; Demography; Psychiatry; Economics; Internal medicine; Agricultural economics; Geography; Political science; Operations research","score_opus":0.16997458196408038,"score_gpt":0.45066889266219834,"score_spread":0.28069431069811795,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2131308570","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":[2.3150162e-7,0.9705626,0.00009261779,0.000051718536,0.0006850623,0.027203301,0.00005663485,0.000048773305,0.0012991058],"genre_scores_gemma":[0.0000019507543,0.98820615,0.0005082043,0.000577131,0.00019417555,0.007862859,0.0017409847,0.00009171775,0.00081682386],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9962899,0.0008321722,0.0010904428,0.0006450634,0.00081116566,0.00033122103],"domain_scores_gemma":[0.9976773,0.00015504201,0.00071583386,0.00087597955,0.00038016037,0.00019566667],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0018183454,0.00060158496,0.0030933733,0.00012514551,0.00005772881,0.000014709636,0.00012940897,0.0002488247,0.00046260195],"category_scores_gemma":[0.00091802346,0.00041192005,0.0013018366,0.00038159167,0.000022273916,0.00005114497,0.000019641653,0.00019277298,0.00056734413],"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":[0.000006601637,0.000162373,0.0000023067182,0.21432778,0.00014694108,0.000002659793,0.000004877371,5.1116878e-8,3.3147238e-8,0.000017857412,0.005423326,0.7799052],"study_design_scores_gemma":[0.0009141027,0.00039843525,0.000004814814,0.29847318,0.01045236,0.000027292424,3.3682383e-7,0.0000079086885,4.1353235e-7,0.000012985789,0.6894883,0.00021991663],"about_ca_topic_score_codex":0.0000015701149,"about_ca_topic_score_gemma":0.0000025108918,"teacher_disagreement_score":0.77968526,"about_ca_system_score_codex":0.0016591739,"about_ca_system_score_gemma":0.0010402115,"threshold_uncertainty_score":0.9998333},"labels":[],"label_agreement":null},{"id":"W2166236442","doi":"10.1177/0193841x14531260","title":"Grief, Anger, and Relationality","year":2014,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Emotional Labor in Professions","field":"Social 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":"Public Health Ontario; Toronto Rehabilitation Institute; University of Toronto; University Health Network","funders":"Institute of Gender and Health; Canadian Institutes of Health Research; Toronto Rehabilitation Institute; Ontario Ministry of Health and Long-Term Care","keywords":"Grief; Anger; Empathy; Psychology; Context (archaeology); Neurorehabilitation; Nursing; Thematic analysis; Focus group; Psychosocial; Rehabilitation; Health care; Feeling; Psychotherapist; Medicine; Qualitative research; Clinical psychology; Psychiatry; Physical therapy; Social psychology","score_opus":0.08182174885329145,"score_gpt":0.47375205392300074,"score_spread":0.3919303050697093,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2166236442","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":[0.004490334,0.0669491,0.00033911524,0.05057369,0.00044162574,0.00093123113,0.0000046390887,0.000060746446,0.8762095],"genre_scores_gemma":[0.9519466,0.035594594,0.0018585216,0.006798548,0.00036426092,0.00012181306,0.000051921877,0.000007535082,0.0032561868],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.99746984,0.0015066338,0.00017440158,0.00012776787,0.00063742965,0.00008395069],"domain_scores_gemma":[0.99905145,0.00033536577,0.000089584544,0.000117265,0.0003442645,0.00006209367],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0082707405,0.000042948068,0.00009430989,0.000016830594,0.0003561829,0.000014297974,0.000068451016,0.000038709015,0.0024801],"category_scores_gemma":[0.0049058613,0.000037967184,0.000025209358,0.00019489677,0.00009720088,0.00014678533,0.00001796394,0.00006493048,0.00015990308],"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":[2.9344497e-7,0.000009436711,0.0022189997,0.00008734815,0.000002857627,2.0189056e-8,0.00021881581,0.0000010244801,0.0000016651686,0.93143857,0.011888773,0.05413218],"study_design_scores_gemma":[0.000098245415,0.0000055263454,0.04649836,0.0007094136,0.000054927485,2.939949e-7,0.00003495615,0.00014319511,6.957817e-7,0.46949086,0.48289248,0.00007105372],"about_ca_topic_score_codex":0.000045053755,"about_ca_topic_score_gemma":0.000097625845,"teacher_disagreement_score":0.9474563,"about_ca_system_score_codex":0.00005024173,"about_ca_system_score_gemma":0.00015397333,"threshold_uncertainty_score":0.99843174},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2168096943","doi":"10.1177/0193841x12439704","title":"Homelessness Outcome Reporting Normative Framework","year":2012,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Homelessness and Social Issues","field":"Health Professions","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 Victoria","funders":"","keywords":"Normative; Report card; Outcome (game theory); Psychology; Political science; Economics","score_opus":0.3399599014597608,"score_gpt":0.588645813129939,"score_spread":0.24868591167017823,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2168096943","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":[0.53568244,0.36702973,0.004101579,0.0025649855,0.0050667073,0.0064414735,0.000014251324,0.00034024977,0.07875858],"genre_scores_gemma":[0.9860975,0.009819901,0.0006589536,0.001487945,0.00077835756,0.0008717691,0.00004097363,0.000024243673,0.00022032685],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99440515,0.0019888247,0.0022076236,0.00017116283,0.00072488445,0.00050235295],"domain_scores_gemma":[0.99566996,0.00058342575,0.0025092126,0.00038036576,0.00069715653,0.0001598744],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.01589316,0.00015990081,0.0009033907,0.000046186946,0.00062623504,0.000008572465,0.00013412467,0.00017840549,0.0070087695],"category_scores_gemma":[0.010921997,0.00012593101,0.00013583116,0.0004232129,0.000026250947,0.00035396407,0.00007331522,0.00054857944,0.002294539],"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":[0.0000031850675,0.0000762141,0.80417943,0.0076352768,0.000039953746,9.128202e-7,0.043205984,0.000002055662,0.0000023324537,0.019135173,0.0005535409,0.12516597],"study_design_scores_gemma":[0.0007777304,0.000033278215,0.75800043,0.02830599,0.00083230966,8.557414e-7,0.073816605,0.00021105856,0.000014044869,0.010989695,0.12607713,0.00094087474],"about_ca_topic_score_codex":0.000053809577,"about_ca_topic_score_gemma":0.000049541355,"teacher_disagreement_score":0.4504151,"about_ca_system_score_codex":0.00021369607,"about_ca_system_score_gemma":0.00017565578,"threshold_uncertainty_score":0.9984823},"labels":[],"label_agreement":null},{"id":"W2302098269","doi":"10.1177/0193841x16637950","title":"How Do Evaluators Differentiate Successful From Less-Than-Successful Experiences With Collaborative Approaches to Evaluation?","year":2016,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University; Queen's University; University of Ottawa","funders":"University of Ottawa","keywords":"Stakeholder; Antecedent (behavioral psychology); Psychology; Control (management); Computer-assisted web interviewing; Applied psychology; Knowledge management; Medical education; Social psychology; Public relations; Computer science; Political science; Business; Marketing; Medicine","score_opus":0.4512052275898277,"score_gpt":0.4632224952179479,"score_spread":0.01201726762812022,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2302098269","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":[0.89155376,0.035332173,0.012257058,0.044442862,0.001119952,0.008307938,0.00012322952,0.00010871651,0.0067542857],"genre_scores_gemma":[0.9875416,0.0024545793,0.0014131113,0.0011465873,0.0002551433,0.006001501,0.00008972424,0.00004012029,0.0010576345],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.97721094,0.00516171,0.0016038335,0.00178119,0.013663849,0.0005784689],"domain_scores_gemma":[0.9890987,0.0016659063,0.0014054758,0.0018203985,0.005571951,0.0004375296],"candidate_categories":["metaepi_narrow","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.025448257,0.0006103596,0.0010893012,0.00047631975,0.00049192767,0.0012676689,0.0015179805,0.00013760045,0.023181902],"category_scores_gemma":[0.0068408926,0.00031649723,0.00021978641,0.0032830173,0.0002480286,0.002438165,0.00021629638,0.00016821362,0.0011803685],"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":[0.00013037425,0.00020803815,0.037016418,0.00009240625,0.00018196343,0.0000013437661,0.007926441,0.000252876,0.00011702149,0.0010088573,0.010144334,0.9429199],"study_design_scores_gemma":[0.025731193,0.003795957,0.33307865,0.02531322,0.0066817235,0.000032172054,0.23301339,0.07742056,0.0108191,0.06644071,0.2094437,0.0082296245],"about_ca_topic_score_codex":0.000024381407,"about_ca_topic_score_gemma":0.0003772673,"teacher_disagreement_score":0.9346903,"about_ca_system_score_codex":0.0004785463,"about_ca_system_score_gemma":0.0020789178,"threshold_uncertainty_score":0.9999287},"labels":[],"label_agreement":null},{"id":"W2889669442","doi":"10.1177/0193841x18799093","title":"The Sensitivity of Impact Estimates to Data Sources Used: Analysis From an Access to Postsecondary Education Experiment","year":2018,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":4,"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":"Canada Millennium Scholarship Foundation","keywords":"Context (archaeology); Psychological intervention; Survey data collection; Randomized experiment; Psychology; Estimation; Political science; Demographic economics; Statistics; Economics; Geography","score_opus":0.3855258045495976,"score_gpt":0.6285228913714911,"score_spread":0.24299708682189347,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2889669442","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":[0.96836394,0.010469194,0.0032133237,0.014922366,0.0006599692,0.0017577882,0.00017940773,0.000012341641,0.00042169174],"genre_scores_gemma":[0.9927411,0.00028191312,0.0022995365,0.003601598,0.00030429594,0.00013184232,0.0005462805,0.0000071494383,0.00008626941],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9956521,0.0009699126,0.00076504867,0.0005642118,0.001904584,0.00014416073],"domain_scores_gemma":[0.9934931,0.0014185212,0.0004630261,0.0018705338,0.0025806287,0.00017423935],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.014602446,0.00013525896,0.00035262448,0.00021511577,0.00022633382,0.00046814428,0.0011744407,0.000020195652,0.0044800113],"category_scores_gemma":[0.007822216,0.000078592806,0.00010304115,0.0016587419,0.000047931517,0.0009241071,0.00039264833,0.000040060004,0.00022433036],"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":[0.000023011287,0.00034858997,0.055525243,0.000020840955,0.0003100172,4.01626e-8,0.0012743974,0.00015035202,0.0025333846,0.00007632827,0.13282828,0.8069095],"study_design_scores_gemma":[0.000103283804,0.00025137968,0.9411926,0.0002996905,0.00084556785,2.8756378e-7,0.0011097601,0.016578251,0.0044832015,0.0030742034,0.03183115,0.00023062342],"about_ca_topic_score_codex":0.0007242469,"about_ca_topic_score_gemma":0.0010232893,"teacher_disagreement_score":0.8856674,"about_ca_system_score_codex":0.00011374465,"about_ca_system_score_gemma":0.0009041511,"threshold_uncertainty_score":0.99643004},"labels":[],"label_agreement":null},{"id":"W4226126191","doi":"10.1177/0193841x221085355","title":"Analyzing the Nexus Between Geopolitical Risk, Policy Uncertainty, and Tourist Arrivals: Evidence From the United States","year":2022,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Market Dynamics and Volatility","field":"Economics, Econometrics and Finance","cited_by":30,"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":"Nexus (standard); Geopolitics; Tourism; Political science; Economics; Regional science; Geography; Computer science","score_opus":0.10686984929016416,"score_gpt":0.3450513100099599,"score_spread":0.23818146071979576,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4226126191","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":[0.53550303,0.40282667,0.0021849836,0.053431407,0.00020561836,0.001788322,0.0020705417,0.000033000357,0.0019564144],"genre_scores_gemma":[0.9288128,0.06804453,0.00005235277,0.0022280093,0.00021548798,0.00016646257,0.00038833075,0.000013173997,0.0000788382],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9976046,0.0010342465,0.0006685035,0.00034034086,0.00014326717,0.00020900516],"domain_scores_gemma":[0.99678135,0.0020343647,0.0004692925,0.0005634421,0.00008658192,0.00006495147],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01135692,0.00012758611,0.0003465902,0.00007576479,0.00065219385,0.00010399612,0.00037577585,0.000024607809,0.0022192134],"category_scores_gemma":[0.004544131,0.00009089265,0.00010510872,0.0006642952,0.00009293205,0.00008955182,0.00023922804,0.00032729455,0.00002617636],"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":[0.000013058665,0.000036007157,0.91433185,0.0002308404,0.0002082903,5.5734876e-7,0.0009995161,0.0014860713,1.0641888e-7,0.034339156,0.008102235,0.040252287],"study_design_scores_gemma":[0.00016826323,0.000030455622,0.23884977,0.00018961146,0.00015390638,8.003263e-7,0.00012421522,0.4164659,5.33807e-8,0.24257393,0.101287045,0.00015604634],"about_ca_topic_score_codex":0.014692764,"about_ca_topic_score_gemma":0.00012576004,"teacher_disagreement_score":0.6754821,"about_ca_system_score_codex":0.00021284894,"about_ca_system_score_gemma":0.00008012902,"threshold_uncertainty_score":0.99869287},"labels":[],"label_agreement":null},{"id":"W4281261681","doi":"10.1177/0193841x221098211","title":"Post-Graduation Work Visas and Loopholes: Insights into Support Provision for International Graduates from the Perspectives of Migration Agents, Universities, and International Graduates","year":2022,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Impulse Buying and Technology Impacts","field":"Economics, Econometrics and Finance","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Impact","funders":"","keywords":"Graduation (instrument); Nexus (standard); Work (physics); Scope (computer science); International education; Public relations; Political science; Career Pathways; Higher education; Medical education; Medicine; Engineering; Law","score_opus":0.05548297578315207,"score_gpt":0.3011743031318248,"score_spread":0.24569132734867274,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4281261681","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":[0.89132893,0.07159904,0.00078400614,0.033872176,0.00062216696,0.0012111436,0.00021704129,0.00003171685,0.00033378875],"genre_scores_gemma":[0.97678113,0.020970639,0.0005525587,0.00048208234,0.00005192899,0.00011781428,0.000799568,0.000010742162,0.00023352042],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9989838,0.00007923468,0.0004178098,0.000297913,0.00013646415,0.00008477529],"domain_scores_gemma":[0.9989049,0.0001524791,0.00047909698,0.000158541,0.00028063377,0.00002436436],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010296578,0.00011172968,0.00021761146,0.0001793887,0.00022982169,0.00004296363,0.00019781129,0.000036325437,0.00030263438],"category_scores_gemma":[0.00058667606,0.00010363232,0.00006349192,0.000186167,0.00008730563,0.00026083575,0.000115100665,0.0001019155,0.000007964999],"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":[0.0004966412,0.00095488416,0.13239726,0.00093928835,0.0017161667,0.000003825046,0.10375638,0.00042147425,0.00037196817,0.27299914,0.057502072,0.4284409],"study_design_scores_gemma":[0.006248095,0.0012030683,0.19387862,0.0015008897,0.00061616104,0.000020319545,0.043131933,0.040609866,0.00036526102,0.39902872,0.31221545,0.0011816119],"about_ca_topic_score_codex":0.0003170347,"about_ca_topic_score_gemma":0.000030117344,"teacher_disagreement_score":0.42725927,"about_ca_system_score_codex":0.00020583784,"about_ca_system_score_gemma":0.00006274191,"threshold_uncertainty_score":0.42260036},"labels":[],"label_agreement":null},{"id":"W4315752656","doi":"10.1177/0193841x221149809","title":"The Dynamic Nexus Between International Tourism and Environmental Degradation in Top Twenty Tourist Destinations: New Insights From Quantile-on-Quantile Approach","year":2023,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Energy, Environment, Economic Growth","field":"Economics, Econometrics and Finance","cited_by":28,"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":"Quantile; Tourism; Nexus (standard); Quantile regression; Granger causality; Destinations; Econometrics; China; Geography; Economic geography; Economics; Computer science","score_opus":0.05660189426987145,"score_gpt":0.2808662194890804,"score_spread":0.22426432521920897,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4315752656","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":[0.9012896,0.075780734,0.0033566887,0.0062113376,0.0008003011,0.0019528386,0.00031068822,0.000076387696,0.010221425],"genre_scores_gemma":[0.94880927,0.046813164,0.000639759,0.00031733207,0.00013979363,0.00022838886,0.0019566908,0.00003065975,0.0010649218],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9981271,0.00012069489,0.0008730513,0.00053754734,0.00015422626,0.00018736425],"domain_scores_gemma":[0.99872535,0.00028934187,0.00052112225,0.00038313714,0.000008430039,0.00007262297],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0013347296,0.00018284538,0.00033605742,0.00020860879,0.00015579471,0.00009555586,0.0002874946,0.000076141696,0.00030089845],"category_scores_gemma":[0.00036480383,0.00018020828,0.000075176446,0.00023524265,0.00005596747,0.00030094152,0.0000939025,0.00015753388,0.0005990287],"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":[0.000040452156,0.00056715374,0.34513626,0.00035480803,0.0005648482,0.000008811171,0.00160967,0.012063397,0.00007421556,0.16959254,0.02417519,0.44581267],"study_design_scores_gemma":[0.0011586264,0.0000433043,0.6869699,0.00028070185,0.000050473987,0.0000015219989,0.0001580647,0.19254054,0.000015166712,0.06834165,0.050045066,0.00039501206],"about_ca_topic_score_codex":0.00014064266,"about_ca_topic_score_gemma":0.000049530914,"teacher_disagreement_score":0.44541767,"about_ca_system_score_codex":0.00039773626,"about_ca_system_score_gemma":0.000026950966,"threshold_uncertainty_score":0.7699502},"labels":[],"label_agreement":null},{"id":"W4366256521","doi":"10.1177/0193841x231166973","title":"Examining the Effects of Renewable Energy and Economic Growth on Carbon Emission in Canada: Evidence from the Nonlinear ARDL Approaches","year":2023,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Energy, Environment, Economic Growth","field":"Economics, Econometrics and Finance","cited_by":26,"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":"Renewable energy; Per capita; Economics; Natural resource economics; Greenhouse gas; Energy consumption; Fossil fuel; Shock (circulatory); Energy intensity; Environmental science; Consumption (sociology); Unit root; Econometrics; Ecology; Engineering; Waste management","score_opus":0.10403570464279333,"score_gpt":0.24364218887594705,"score_spread":0.13960648423315372,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4366256521","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":[0.8575931,0.1354706,0.000018116125,0.0026134457,0.0003151658,0.0006022375,0.000025950067,0.000009942486,0.0033513934],"genre_scores_gemma":[0.8935496,0.10544698,0.000040773255,0.00055482,0.00006939352,0.00018151043,0.000028062648,0.000018853914,0.0001099917],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99849963,0.00020015209,0.00064717897,0.00040667906,0.00007871123,0.00016762133],"domain_scores_gemma":[0.9978076,0.0011969779,0.000494067,0.00045563097,0.000007439799,0.0000382859],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0026486104,0.00014170897,0.00040659172,0.00006087702,0.00005500268,0.00001583961,0.00025465759,0.000040852483,0.00006846304],"category_scores_gemma":[0.0009872736,0.0001092127,0.000040895255,0.0001695304,0.000037689075,0.00008576559,0.00008630059,0.00008214864,0.000014225087],"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":[0.00003649517,0.00007809598,0.8805548,0.002300233,0.0002698874,0.0000054976626,0.00090586074,0.037953734,0.0002040077,0.009532426,0.015597549,0.052561413],"study_design_scores_gemma":[0.0010157418,0.00012894088,0.6306985,0.0056079635,0.000118096825,0.0000015711171,0.00016765668,0.3383105,0.0016316043,0.01087478,0.010890217,0.00055442424],"about_ca_topic_score_codex":0.45318675,"about_ca_topic_score_gemma":0.2022767,"teacher_disagreement_score":0.30035678,"about_ca_system_score_codex":0.00053530524,"about_ca_system_score_gemma":0.0001747965,"threshold_uncertainty_score":0.81227964},"labels":[],"label_agreement":null},{"id":"W4386533916","doi":"10.1177/0193841x231198706","title":"Building a Logic Model to Foster Engagement and Learning Using the Case of a Province-Wide Multispecies Antimicrobial Use Monitoring System","year":2023,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Centre Hospitalier de l’Université de Montréal; Université de Montréal","funders":"Ministère de l'Agriculture, des Pêcheries et de l'Alimentation","keywords":"Logic model; Computer science; Process (computing); Framing (construction); Corporate governance; Reflexivity; Management science; Knowledge management; Process management; Engineering; Sociology; Business; Social science","score_opus":0.8525169825921586,"score_gpt":0.6806434946012924,"score_spread":0.1718734879908662,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386533916","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":[0.9720006,0.00067648373,0.013677627,0.0072218208,0.00028754282,0.005967696,0.000017585911,0.0000856723,0.0000649872],"genre_scores_gemma":[0.96949416,0.0012245285,0.021943036,0.0064862603,0.00011798201,0.00057881995,0.0000027379335,0.000022207581,0.0001302552],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9937076,0.0042838855,0.00093548413,0.00026896017,0.0004219258,0.00038217098],"domain_scores_gemma":[0.996274,0.002380078,0.0005016767,0.00025482482,0.0004856292,0.000103809594],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017904028,0.0001167501,0.00028216516,0.00015265419,0.0012961681,0.000028626877,0.00011847982,0.000037201033,0.000032259257],"category_scores_gemma":[0.008019641,0.00008502791,0.00003599218,0.0006721594,0.000031834094,0.00024307522,0.00026220188,0.00030574965,0.00004197228],"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":[0.0000871209,0.00007588263,0.061120458,0.13657022,0.000176202,0.0001818309,0.20002986,0.2355048,0.05656067,0.0075649517,0.04507449,0.2570535],"study_design_scores_gemma":[0.0014131566,0.00009112113,0.0072216163,0.034461033,0.00036925456,0.00007881175,0.019408708,0.9206579,0.00033519353,0.00007437605,0.015459031,0.0004298376],"about_ca_topic_score_codex":0.00022762308,"about_ca_topic_score_gemma":0.000056365283,"teacher_disagreement_score":0.68515307,"about_ca_system_score_codex":0.0003717534,"about_ca_system_score_gemma":0.00055224064,"threshold_uncertainty_score":0.9969208},"labels":[],"label_agreement":null},{"id":"W4389722681","doi":"10.1177/0193841x231221303","title":"When Who Matters: Interviewer Effects and Survey Modality","year":2023,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Survey Methodology and Nonresponse","field":"Social Sciences","cited_by":2,"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":"International Development Research Centre","keywords":"Interview; Phone; Modality (human–computer interaction); Psychology; Leverage (statistics); Modalities; Social psychology; Applied psychology; Survey data collection; Statistics; Computer science; Sociology","score_opus":0.4999020040566802,"score_gpt":0.557242604081877,"score_spread":0.05734060002519675,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389722681","genre_codex":"empirical","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":[0.5357648,0.30439055,0.0015426149,0.1414883,0.0030861576,0.006006572,0.000040528816,0.00044885225,0.007231596],"genre_scores_gemma":[0.2702304,0.61472595,0.0027979838,0.09205358,0.0007550142,0.0016701553,0.00050385966,0.00009050169,0.017172553],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9034745,0.095364906,0.00025472886,0.00024776123,0.00046130846,0.00019675928],"domain_scores_gemma":[0.9877156,0.011720172,0.0000882924,0.00018833787,0.0002198718,0.00006772403],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21268375,0.00007557413,0.0002565381,0.000051210238,0.00018518367,0.000024249903,0.0001246422,0.00006337833,0.001266322],"category_scores_gemma":[0.05532127,0.00006647151,0.000048429192,0.00042069203,0.000118098134,0.00012370756,0.000046132234,0.00007832929,0.00050649705],"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":[0.0001920975,0.000039692226,0.046091057,0.0027598767,0.000086848726,0.0000050659883,0.008722382,0.0000016760004,0.00002768082,0.0010735285,0.37247115,0.56852895],"study_design_scores_gemma":[0.0001699737,0.000017269169,0.91782427,0.0010666461,0.00006585312,4.4416467e-7,0.000053377054,0.00007351626,0.0000070331776,0.0050916476,0.0755233,0.00010669727],"about_ca_topic_score_codex":0.0005643748,"about_ca_topic_score_gemma":0.0009725565,"teacher_disagreement_score":0.8717332,"about_ca_system_score_codex":0.000035821537,"about_ca_system_score_gemma":0.00012423744,"threshold_uncertainty_score":0.99964666},"labels":[],"label_agreement":null},{"id":"W4409153149","doi":"10.1177/0193841x251331723","title":"A Critical Reflection of Generalization in Mixed Methods Research","year":2025,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Evaluation and Performance Assessment","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 de Montréal; Centre for Interdisciplinary Research in Rehabilitation","funders":"","keywords":"Generalization; Constructive; Multimethodology; Relevance (law); Management science; Computer science; Reflection (computer programming); Psychology; Epistemology; Mathematics education; Process (computing); Political science","score_opus":0.7561451234814716,"score_gpt":0.7902642498582428,"score_spread":0.03411912637677128,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409153149","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":[0.026428724,0.21613093,0.3893906,0.07450326,0.004043583,0.006459859,0.0000072512757,0.000067734465,0.28296807],"genre_scores_gemma":[0.8900823,0.034898482,0.06545601,0.004164876,0.00011424271,0.0011226669,0.00005305173,0.000016846914,0.0040915124],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9825403,0.012153848,0.001455338,0.00039832492,0.003265872,0.00018629609],"domain_scores_gemma":[0.99180037,0.0029146376,0.00018206134,0.0005307549,0.0045365277,0.000035674573],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1237473,0.00007601793,0.00032607545,0.0008254117,0.000101129444,0.000061049366,0.00029476857,0.00006238579,0.0039748973],"category_scores_gemma":[0.05347765,0.000058857175,0.00007578693,0.004820862,0.00007870713,0.00032815008,0.00007010957,0.00016433741,0.000119396296],"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":[0.000009694826,0.000078893645,0.0017583248,0.00033733924,0.000005509396,9.647043e-8,0.000118122145,0.0004529586,0.0012317554,0.04759291,0.018398045,0.93001634],"study_design_scores_gemma":[0.0009645883,0.00014862852,0.09582551,0.0044455063,0.00011555706,0.0000019497822,0.00044821654,0.2964904,0.0042091673,0.3791482,0.21799298,0.00020931433],"about_ca_topic_score_codex":0.000019546564,"about_ca_topic_score_gemma":0.000080947175,"teacher_disagreement_score":0.92980707,"about_ca_system_score_codex":0.00019420388,"about_ca_system_score_gemma":0.00078321475,"threshold_uncertainty_score":0.9969356},"labels":[],"label_agreement":null}]}