{"meta":{"query_hash":"0e97332707a1","filters":{"venue":"Accreditation and Quality Assurance"},"cohort_total":29,"direct_labels_cover":0,"predictions_cover":29,"exported":29,"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/0e97332707a1","api":"https://metacan.xera.ac/api/v1/cohort?venue=Accreditation+and+Quality+Assurance"},"results":[{"id":"W1964146798","doi":"10.1007/s00769-006-0118-8","title":"Quality specifications for evaluation and comparison of performance among external quality assessment schemes in occupational and environmental laboratory medicine","year":2006,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institut National de Santé Publique du Québec","funders":"European Commission","keywords":"Equivalence (formal languages); External quality assessment; Quality assurance; Reliability engineering; Quality (philosophy); Quality assessment; Computer science; Set (abstract data type); Data mining; Risk analysis (engineering); Evaluation methods; Engineering; Mathematics; Operations management; Medicine","score_opus":0.17667360289144882,"score_gpt":0.4736216374677477,"score_spread":0.2969480345762989,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1964146798","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":[0.991031,0.0012921082,0.004593921,0.001842315,0.000086835054,0.000784237,0.00023272127,0.00001381828,0.00012307486],"genre_scores_gemma":[0.9947941,0.0002006985,0.004177886,0.00018534176,0.00018222138,0.00009043403,0.00033482106,0.000008335095,0.000026190612],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99685645,0.0006764178,0.001345296,0.0003729361,0.0006073993,0.00014150521],"domain_scores_gemma":[0.9957528,0.0028352544,0.00080397207,0.0002119712,0.00030406602,0.00009189761],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0061262273,0.00015088511,0.00054639933,0.000084787236,0.00012490465,0.000024250938,0.000044187575,0.000111712376,0.000038758168],"category_scores_gemma":[0.0016352998,0.00014111494,0.000034169374,0.0001433773,0.00038828296,0.00044371493,0.000019625684,0.00018666494,4.0507967e-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":[0.00053968345,0.00032280412,0.9685325,0.00034171963,0.000023243212,1.05646585e-7,0.00033105485,0.000025547602,0.011296047,0.011384329,0.00006811149,0.007134826],"study_design_scores_gemma":[0.005534842,0.00018681784,0.9835331,0.00011139279,0.000065944136,4.9506355e-7,0.0014264222,0.006880901,0.00034538272,0.0005066265,0.0012617781,0.00014628666],"about_ca_topic_score_codex":0.0002631028,"about_ca_topic_score_gemma":0.00039752005,"teacher_disagreement_score":0.015000583,"about_ca_system_score_codex":0.00006732207,"about_ca_system_score_gemma":0.00013428199,"threshold_uncertainty_score":0.57545},"labels":[],"label_agreement":null},{"id":"W1964182244","doi":"10.1007/s00769-009-0561-4","title":"Criteria to define the standard deviation for proficiency assessment for the determination of essential trace elements in serum: comparison of Z-scores based on the Horwitz function or on biological variability","year":2009,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Trace Elements in Health","field":"Nursing","cited_by":5,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institut National de Santé Publique du Québec","funders":"","keywords":"TRACE (psycholinguistics); Standard deviation; Quality (philosophy); Function (biology); Statistics; Mathematics; Reference values; Medicine; Physics","score_opus":0.10825897629631515,"score_gpt":0.46031199485717417,"score_spread":0.352053018560859,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1964182244","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":[0.7079737,0.000018572422,0.27408102,0.014422391,0.00044176957,0.0028505437,0.0001845124,0.000012690692,0.000014758049],"genre_scores_gemma":[0.99018574,0.0000038281546,0.00859725,0.0008141149,0.00006819092,0.00028647055,0.000037119495,0.000005427456,0.0000018633152],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99770063,0.00074467453,0.00075520674,0.00025335222,0.00037671987,0.00016939735],"domain_scores_gemma":[0.9931596,0.0058088177,0.00049658515,0.00024157796,0.00027225754,0.000021174066],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0053335982,0.000117768686,0.00022859813,0.00004118847,0.00025129315,0.00003275012,0.00014578286,0.000062880266,0.000016267415],"category_scores_gemma":[0.002648799,0.00006171067,0.000046404584,0.00023608869,0.00006233343,0.00010100094,0.000007076641,0.00010841947,1.4134632e-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":[0.038795296,0.0027499087,0.20994481,0.0009203007,0.000039002804,8.394355e-8,0.006107187,0.005151004,0.022961397,0.04228661,0.0019303676,0.66911405],"study_design_scores_gemma":[0.0013306402,0.0036397374,0.9377721,0.00008393536,0.000022753724,5.120084e-8,0.0007922083,0.046365805,0.007186783,0.0024182992,0.00030446568,0.00008320253],"about_ca_topic_score_codex":0.000017784208,"about_ca_topic_score_gemma":0.00016670917,"teacher_disagreement_score":0.7278273,"about_ca_system_score_codex":0.0001031193,"about_ca_system_score_gemma":0.00006554986,"threshold_uncertainty_score":0.3171051},"labels":[],"label_agreement":null},{"id":"W1967499314","doi":"10.1007/s00769-008-0421-7","title":"External quality assessment in microbiology: comparison of results from Belgian and Canadian laboratories with regard to their ability to identify Streptococcus pyogenes","year":2008,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Bacterial Identification and Susceptibility Testing","field":"Biochemistry, Genetics and Molecular Biology","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Clinical microbiology; External quality assessment; Streptococcus pyogenes; Viridans streptococci; Streptococcus; Microbiology; Medicine; Throat; Medical physics; Biology; Pathology; Surgery; Staphylococcus aureus; Bacteria","score_opus":0.05474549356872136,"score_gpt":0.3630003531813596,"score_spread":0.3082548596126382,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1967499314","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":[0.99702126,0.000085586544,0.0005118803,0.0008954489,0.000110735404,0.00034405664,0.0009865515,0.00000881246,0.000035690002],"genre_scores_gemma":[0.9958559,0.000016181564,0.0034871828,0.00019579985,0.000057980673,0.000023837369,0.00033575532,0.000006884079,0.000020476675],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.99827063,0.00039230706,0.00059711386,0.00048428276,0.00009004134,0.00016563661],"domain_scores_gemma":[0.99881107,0.00016398674,0.0001962493,0.00033834303,0.00031969196,0.00017067025],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00091181655,0.00013234466,0.00027498894,0.00005693534,0.000114673254,0.00003850783,0.00011269587,0.0001057954,0.0000043800496],"category_scores_gemma":[0.0009301459,0.00012324266,0.00002072097,0.00016931744,0.00012917405,0.000018937148,0.000044716435,0.00008333304,9.349827e-7],"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":[0.00015837127,0.000031516924,0.46647936,0.00001794631,0.000008124331,3.1676365e-7,0.0014247705,0.000018336696,0.5311069,0.000031174568,0.00009138826,0.00063180964],"study_design_scores_gemma":[0.0005563029,0.000111207075,0.94449043,0.000032872016,0.0000031906795,8.135798e-7,0.001381925,0.000005559656,0.05244331,0.000031122378,0.00079491583,0.00014836255],"about_ca_topic_score_codex":0.035697125,"about_ca_topic_score_gemma":0.34374234,"teacher_disagreement_score":0.4786636,"about_ca_system_score_codex":0.000034340894,"about_ca_system_score_gemma":0.00024839857,"threshold_uncertainty_score":0.9707242},"labels":[],"label_agreement":null},{"id":"W1970174238","doi":"10.1007/s00769-014-1082-3","title":"Performance of Canadian clinical laboratories processing throat culture proficiency testing surveys","year":2014,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Bacterial Identification and Susceptibility Testing","field":"Biochemistry, Genetics and Molecular Biology","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of British Columbia; University of British Columbia Hospital","funders":"University of British Columbia","keywords":"Streptococcus pyogenes; Pharyngitis; Throat; Throat culture; Streptococcus; Pathogen; Clinical microbiology; Medicine; Microbiology; Biology; Surgery; Internal medicine; Bacteria; Staphylococcus aureus","score_opus":0.07274539070457586,"score_gpt":0.3449978617734287,"score_spread":0.27225247106885286,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1970174238","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":[0.99764234,0.00008256265,0.00067380874,0.00014406058,0.00019768604,0.00009919398,0.000019829322,0.000013634186,0.0011268692],"genre_scores_gemma":[0.99646974,0.000044255918,0.0029628982,0.000108094995,0.00020190823,0.000005741014,0.00008682898,0.0000064353717,0.00011407873],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9986794,0.00043212075,0.0004018154,0.00025459388,0.000100317564,0.00013177848],"domain_scores_gemma":[0.99886847,0.00008592821,0.0002124676,0.0001717153,0.00057717896,0.00008425198],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0027353049,0.00008323744,0.00013992144,0.000041917934,0.00016046387,0.000044419987,0.00009315282,0.000104330305,0.000002683136],"category_scores_gemma":[0.004071665,0.000078779485,0.000021313132,0.0003181008,0.00012327253,0.000023630357,0.000017837672,0.000070450566,0.0000017178683],"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":[0.000009826906,0.000021882815,0.85800743,0.00013618343,0.0000046066148,3.2862943e-8,0.00024339829,0.000007469349,0.12268273,0.00016245629,0.0001964018,0.018527586],"study_design_scores_gemma":[0.00019227814,0.00011612989,0.9895389,0.000033253957,0.0000044812386,7.4559745e-7,0.00021202592,0.00048085983,0.006387074,0.000017447082,0.0029034656,0.00011335468],"about_ca_topic_score_codex":0.00198699,"about_ca_topic_score_gemma":0.008140862,"teacher_disagreement_score":0.13153145,"about_ca_system_score_codex":0.0000054467464,"about_ca_system_score_gemma":0.0002057823,"threshold_uncertainty_score":0.48744577},"labels":[],"label_agreement":null},{"id":"W1970248714","doi":"10.1007/s00769-006-0144-6","title":"Estimation of the biological variation of glucose-6-phosphate dehydrogenase in dried blood spots","year":2006,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","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":"Newborn Screening Ontario","funders":"","keywords":"Coefficient of variation; Mathematics; Matrix (chemical analysis); Analytical Chemistry (journal); Chemistry; Systematic error; Combinatorics; Statistics; Chromatography","score_opus":0.043625093662082386,"score_gpt":0.34454190133102996,"score_spread":0.3009168076689476,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1970248714","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":[0.9963509,0.00033840592,0.000867527,0.001839671,0.00009433903,0.00031076846,0.000033945114,0.000012140363,0.00015228929],"genre_scores_gemma":[0.99849665,0.000061749866,0.0010637641,0.00026615598,0.000048964896,0.000009587496,0.000030761497,0.000004062566,0.000018306835],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9983678,0.00037858615,0.0007695847,0.00016620403,0.00022571094,0.00009210484],"domain_scores_gemma":[0.9978154,0.0012349004,0.0005659949,0.00018933207,0.0001630354,0.000031359123],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0015373937,0.000078585836,0.00029841336,0.000034927125,0.000034181143,0.000008015601,0.000052019535,0.00010325191,0.000016151518],"category_scores_gemma":[0.002750385,0.000056789475,0.00006123842,0.00029862783,0.00009879709,0.00015353039,0.000014693582,0.00011615112,0.0000012259297],"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":[0.002280948,0.0023104222,0.5731689,0.0007903937,0.00013736737,0.0000059066483,0.0012307621,0.0022036822,0.29076573,0.1109406,0.00008276716,0.016082538],"study_design_scores_gemma":[0.0024745676,0.00009223333,0.96772116,0.00007596116,0.00005521532,0.0000013602477,0.00006608394,0.002354748,0.024974756,0.0019897788,0.00013153603,0.00006257319],"about_ca_topic_score_codex":0.0004602586,"about_ca_topic_score_gemma":0.00014274119,"teacher_disagreement_score":0.39455232,"about_ca_system_score_codex":0.000010802799,"about_ca_system_score_gemma":0.000077899436,"threshold_uncertainty_score":0.32926667},"labels":[],"label_agreement":null},{"id":"W1981139298","doi":"10.1007/s00769-008-0385-7","title":"The European conferences on quality in the spotlight in medical laboratories, Antwerp, Belgium, 20–21 March 2006 and 22–23 March 2007","year":2008,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","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":"Shared Health","funders":"","keywords":"Commercial law; Quality (philosophy); Engineering; Library science; Political science; Computer science; Law; Physics","score_opus":0.08485616912799505,"score_gpt":0.40242432933400296,"score_spread":0.3175681602060079,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1981139298","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":[0.89483625,0.002214428,0.000066823224,0.09349513,0.00040967957,0.0006712928,0.00016752574,0.000036464357,0.0081024],"genre_scores_gemma":[0.9893627,0.0033632696,0.000059565064,0.0062838164,0.00051036733,0.000030836884,0.00007031522,0.000014363163,0.00030474606],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99132395,0.0055389116,0.0011740898,0.0004816217,0.0011126162,0.00036882964],"domain_scores_gemma":[0.9864432,0.012195578,0.0003460106,0.00046978335,0.00033675047,0.0002086771],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.018150149,0.00021612726,0.0005022543,0.00007335605,0.0003493334,0.00010506073,0.00025844225,0.00016688266,0.00005992105],"category_scores_gemma":[0.010690659,0.00013311964,0.000048892824,0.00059241726,0.0006241173,0.00025389311,0.000055407898,0.0009077606,0.000027071535],"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":[0.0026803506,0.0009763775,0.8884402,0.00034038624,0.00008982923,0.0003339211,0.0041163205,0.000003144323,0.0001890783,0.051508572,0.020955421,0.030366385],"study_design_scores_gemma":[0.0026488062,0.00017746021,0.8625077,0.00011324281,0.000012953104,0.000013982567,0.0019885069,0.00007358673,0.000008481063,0.0003258547,0.13196665,0.00016277375],"about_ca_topic_score_codex":0.001140241,"about_ca_topic_score_gemma":0.006173745,"teacher_disagreement_score":0.11101123,"about_ca_system_score_codex":0.00003078588,"about_ca_system_score_gemma":0.0004852852,"threshold_uncertainty_score":0.9976427},"labels":[],"label_agreement":null},{"id":"W1982922574","doi":"10.1007/s00769-009-0506-y","title":"Instability of mercury in specimens of human urine for external quality assessment","year":2009,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Mercury impact and mitigation studies","field":"Environmental Science","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":"Institut National de Santé Publique du Québec","funders":"","keywords":"Mercury (programming language); Urine; Chemistry; Chromatography; Uric acid; Urine sample; External quality assessment; Environmental chemistry; Biochemistry; Medicine; Computer science","score_opus":0.07372258331384528,"score_gpt":0.39603436428930616,"score_spread":0.32231178097546087,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1982922574","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":[0.99419063,0.00005079058,0.0022880738,0.000572209,0.000053787306,0.00027203606,0.000051157698,0.000008127771,0.002513203],"genre_scores_gemma":[0.99642974,0.000030242085,0.0033326459,0.000106381034,0.000021335269,0.000011490146,0.000011747388,0.0000025303027,0.00005390155],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9985379,0.00022513507,0.000646582,0.00019043978,0.00027476138,0.00012517],"domain_scores_gemma":[0.99901706,0.00041074966,0.00032553822,0.00015988013,0.00004245973,0.00004433287],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0020648895,0.00009072415,0.00028793336,0.000030026935,0.00005792453,0.0000064952246,0.00007753546,0.00003687473,0.00012138558],"category_scores_gemma":[0.00034498863,0.00008556003,0.000050147013,0.00013650575,0.00016858046,0.0002431465,0.000024806232,0.000058903042,0.0000010350838],"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":[0.00012233897,0.0003578068,0.6636305,0.00013642278,0.000011139011,1.6705758e-7,0.00257781,0.000069402275,0.311018,0.01210108,0.00025307268,0.009722292],"study_design_scores_gemma":[0.0007456485,0.0001336686,0.9834551,0.000023485802,0.0000050870326,1.6151968e-7,0.0005033335,0.000027804766,0.008599982,0.006250229,0.00016585857,0.000089607944],"about_ca_topic_score_codex":0.00039345794,"about_ca_topic_score_gemma":0.00043797286,"teacher_disagreement_score":0.31982467,"about_ca_system_score_codex":0.00004763727,"about_ca_system_score_gemma":0.000014739144,"threshold_uncertainty_score":0.34890366},"labels":[],"label_agreement":null},{"id":"W1992242927","doi":"10.1007/s00769-006-0108-x","title":"Occupational and environmental laboratory medicine: A network of EQAS organisers","year":2006,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Chemical Safety and Risk Management","field":"Chemical Engineering","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":"Institut National de Santé Publique du Québec","funders":"","keywords":"External quality assessment; Environmental health; Medical laboratory; Medicine; Traceability; Computer science; Pathology","score_opus":0.01119142992951947,"score_gpt":0.25189131332869574,"score_spread":0.24069988339917628,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1992242927","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":[0.9904916,0.0011108476,0.0063846614,0.0008155905,0.000095307725,0.000106826505,0.00006723788,0.00004212567,0.0008857784],"genre_scores_gemma":[0.9984591,0.00014238837,0.0008578828,0.00012770148,0.00022720874,0.0000044434987,0.000076844786,0.0000058484015,0.00009857684],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.999317,0.000027518308,0.00025545902,0.0001365768,0.00016933406,0.000094142895],"domain_scores_gemma":[0.99953777,0.000251775,0.00009203324,0.000062120576,0.000018996168,0.000037299204],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00021688854,0.00007693918,0.00013260334,0.000011186528,0.00004404122,0.0000044087624,0.00003834998,0.000047209734,0.00007002974],"category_scores_gemma":[0.0000813434,0.00006978,0.0000139021895,0.000089303496,0.00011435271,0.00008580748,0.000023361818,0.000061671504,0.0000020593463],"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":[0.0004024872,0.0002260585,0.2071967,0.000879593,0.00014758712,0.000003766245,0.001065933,0.0074905306,0.6220125,0.12020531,0.03423732,0.006132243],"study_design_scores_gemma":[0.0013511757,0.000032013097,0.9732308,0.0000761288,0.00003309724,8.710544e-7,0.00036436683,0.0030612,0.007997818,0.002429953,0.011182538,0.00024004208],"about_ca_topic_score_codex":0.000081014005,"about_ca_topic_score_gemma":0.000014909061,"teacher_disagreement_score":0.76603407,"about_ca_system_score_codex":0.00001852057,"about_ca_system_score_gemma":0.000007997683,"threshold_uncertainty_score":0.28455457},"labels":[],"label_agreement":null},{"id":"W1994851363","doi":"10.1007/s00769-005-0045-0","title":"Chemical and environmental sampling: quality through accreditation, certification and industrial standards","year":2005,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Nordic Life Science Pipeline (Canada)","funders":"European Commission","keywords":"Accreditation; Sampling (signal processing); Certification; Quality assurance; Lot quality assurance sampling; Quality (philosophy); Product certification; Computer science; Engineering; Engineering management; Business; Sampling design; Marketing; Service (business); Political science; Telecommunications; Environmental health; Medicine","score_opus":0.3186829572167391,"score_gpt":0.4813249204239368,"score_spread":0.1626419632071977,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1994851363","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":[0.71985894,0.0007471866,0.27348378,0.004305008,0.00041303996,0.0002990935,0.00060406764,0.00006084204,0.0002280666],"genre_scores_gemma":[0.97480553,0.00014970936,0.024120824,0.00020019323,0.00055517745,0.00003358584,0.000054436274,0.000011642297,0.000068925045],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9962567,0.00037196992,0.0010211136,0.0007154275,0.0014024049,0.0002323848],"domain_scores_gemma":[0.9925681,0.006282693,0.0004441667,0.00027913201,0.00025396983,0.00017193846],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0035283498,0.00019283086,0.00034180877,0.00006336138,0.00033498343,0.0003116256,0.00016844027,0.00015712454,0.000046581226],"category_scores_gemma":[0.020328984,0.00017624126,0.000029465313,0.0002251447,0.0004879269,0.0016720252,0.00009195405,0.00024348749,0.000008500393],"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":[0.0005991287,0.00015710483,0.056998268,0.00008623749,0.000036257872,9.952541e-7,0.00726927,0.00018245052,0.02593545,0.03931017,0.0025086103,0.86691606],"study_design_scores_gemma":[0.004531029,0.0001560729,0.6232682,0.00009730261,0.0000580259,0.000016666007,0.014137928,0.0026468309,0.01193961,0.20429118,0.13756439,0.0012927613],"about_ca_topic_score_codex":0.000033031345,"about_ca_topic_score_gemma":0.000017745546,"teacher_disagreement_score":0.8656233,"about_ca_system_score_codex":0.00008992525,"about_ca_system_score_gemma":0.00007525772,"threshold_uncertainty_score":0.9879232},"labels":[],"label_agreement":null},{"id":"W1997603456","doi":"10.1007/s00769-007-0282-5","title":"Analysis of 19-norandrosterone in human urine by gas chromatography–isotope-dilution mass spectrometry: method adopted by LGC for participation in the Comité Consultatif pour la Quantité de Matière (CCQM) Pilot Study P68","year":2007,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Mass Spectrometry Techniques and Applications","field":"Chemistry","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"World Anti-Doping Agency","keywords":"Isotope dilution; Chromatography; Chemistry; Mass spectrometry; Calibration; Analytical Chemistry (journal); Mathematics","score_opus":0.039330071977626696,"score_gpt":0.3834681268113606,"score_spread":0.34413805483373394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1997603456","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":[0.8797297,0.00010018646,0.11803708,0.00049253413,0.00000645499,0.00057230063,0.00030646412,0.00004636696,0.0007088985],"genre_scores_gemma":[0.9875702,0.00004065152,0.011423978,0.000062848354,0.000018643268,0.0003187604,0.00051445747,0.00001319332,0.000037251106],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.99783117,0.00039617225,0.000860916,0.00034815402,0.0002959677,0.00026763664],"domain_scores_gemma":[0.9970847,0.0019653577,0.00046546702,0.00031007035,0.000115749914,0.00005866221],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0038665005,0.00017833292,0.00045637594,0.00035158652,0.00011633219,0.000055251425,0.00019974077,0.00008910287,0.0000721101],"category_scores_gemma":[0.0003541721,0.00016677173,0.0000718521,0.0017589973,0.00007404494,0.00012683964,0.000016677905,0.0001950744,2.976498e-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":[0.00029431024,0.0022780932,0.44970474,0.00025156522,0.00023923289,0.0000014073105,0.0015530601,0.000052535743,0.5337505,0.010441893,0.0006945912,0.0007380625],"study_design_scores_gemma":[0.00464965,0.0004952569,0.91273445,0.000056029232,0.00042977624,0.0000019742868,0.0046598082,0.00453998,0.06826601,0.0031180254,0.0006374008,0.00041162377],"about_ca_topic_score_codex":0.001990742,"about_ca_topic_score_gemma":0.0027402127,"teacher_disagreement_score":0.4654845,"about_ca_system_score_codex":0.000073208845,"about_ca_system_score_gemma":0.000022696508,"threshold_uncertainty_score":0.68007535},"labels":[],"label_agreement":null},{"id":"W2009412568","doi":"10.1007/s00769-012-0906-2","title":"Determination of the standard deviation for proficiency assessment from past participant’s performances","year":2012,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Pesticide Residue Analysis and Safety","field":"Agricultural and Biological Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institut National de Santé Publique du Québec","funders":"","keywords":"Standard deviation; Reproducibility; Statistics; Relative standard deviation; Mathematics; Mercury (programming language); Analytical Chemistry (journal); Computer science; Chromatography; Chemistry; Detection limit","score_opus":0.08245211673099916,"score_gpt":0.34273756272306555,"score_spread":0.2602854459920664,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2009412568","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":[0.99496174,0.00016238012,0.0030694194,0.0012963562,0.00011767506,0.00018189062,0.00011394111,0.000008190988,0.00008838678],"genre_scores_gemma":[0.99853176,0.000041285148,0.0010615232,0.000074276744,0.00018414044,0.00003210993,0.00005884348,3.9300377e-7,0.000015647352],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9991605,0.00013010121,0.00027096088,0.00010140665,0.00021151485,0.00012551466],"domain_scores_gemma":[0.99898964,0.0005692909,0.00024172268,0.000040286843,0.0001259491,0.000033124183],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00080429466,0.0000585509,0.00012923707,0.0000051240277,0.00018512984,0.000022220001,0.00007540702,0.00003481148,0.00002183587],"category_scores_gemma":[0.00018809405,0.000021352318,0.00005308851,0.00012894011,0.000039812272,0.0002769303,0.000015056562,0.000033091943,5.5983566e-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":[0.000034933753,0.000055614262,0.8739732,0.00002762609,0.000011831789,9.4984385e-9,0.00051524397,0.000011224322,0.03255307,0.002668523,0.000079654914,0.09006907],"study_design_scores_gemma":[0.000105649044,0.000046836467,0.99399924,0.00001594462,0.00002783391,7.6398365e-8,0.00050156255,0.00093040406,0.0026530998,0.00048462843,0.0011766627,0.000058087244],"about_ca_topic_score_codex":0.00010056947,"about_ca_topic_score_gemma":0.0003267625,"teacher_disagreement_score":0.120026015,"about_ca_system_score_codex":0.000011788701,"about_ca_system_score_gemma":0.000011222503,"threshold_uncertainty_score":0.14238878},"labels":[],"label_agreement":null},{"id":"W2010598206","doi":"10.1007/s00769-010-0669-6","title":"Conceptual thinking and metrology concepts","year":2010,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Statistics Education and Methodologies","field":"Mathematics","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":"Toronto Rehabilitation Institute; University of Toronto","funders":"","keywords":"Competence (human resources); Computer science; Enabling; Operationalization; Metrology; Management science; Engineering; Psychology; Mathematics","score_opus":0.20620552319380506,"score_gpt":0.47691645462167703,"score_spread":0.27071093142787195,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2010598206","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":[0.94413555,0.00015723777,0.04953495,0.0021763595,0.0017092887,0.00015543694,0.00004470236,0.00010477509,0.0019816747],"genre_scores_gemma":[0.798658,0.00004836297,0.20012915,0.0006334391,0.00015172707,0.000015780048,0.000011552591,0.0000076129986,0.00034434843],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9989405,0.0003166531,0.00026414165,0.00020290616,0.00014181154,0.00013395661],"domain_scores_gemma":[0.9926462,0.006846784,0.00016660307,0.00014402349,0.00012939959,0.00006695316],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0015544079,0.00009697694,0.00020620656,0.000041748972,0.00014034713,0.00004547636,0.00007414394,0.000100449615,0.00009940308],"category_scores_gemma":[0.010916857,0.000088648434,0.000018194478,0.00006606181,0.00039153927,0.000116448,0.000028196906,0.00020379711,0.000005716737],"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":[0.000014429052,0.000016839975,0.0061386838,0.000031284664,0.000013119985,4.001448e-7,0.00475218,1.35104e-7,0.002431893,0.97890204,0.0041667637,0.0035322492],"study_design_scores_gemma":[0.0010686814,0.00006470037,0.20463179,0.000014856662,0.000034774148,0.000017866796,0.009260206,0.00004845793,0.001403693,0.7539504,0.029217022,0.00028755656],"about_ca_topic_score_codex":0.00002514519,"about_ca_topic_score_gemma":0.00011914427,"teacher_disagreement_score":0.22495162,"about_ca_system_score_codex":0.000004711832,"about_ca_system_score_gemma":0.000043715223,"threshold_uncertainty_score":0.9974146},"labels":[],"label_agreement":null},{"id":"W2021927114","doi":"10.1007/s00769-008-0450-2","title":"Uncertainties when using dual-wavelength or single-wavelength spectrophotometry to monitor the course of hexavalent chromium reduction mediated by zero-valent iron","year":2008,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Environmental remediation with nanomaterials","field":"Engineering","cited_by":6,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Hexavalent chromium; Spectrophotometry; Chromium; Zerovalent iron; Chemistry; Wavelength; Analytical Chemistry (journal); Contamination; Nuclear chemistry; Environmental chemistry; Chromatography; Optics; Physics; Organic chemistry","score_opus":0.03931189639586145,"score_gpt":0.2735875994995712,"score_spread":0.23427570310370974,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2021927114","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":[0.9958811,0.0004725664,0.0015577928,0.0003014435,0.0009319173,0.0004801278,0.00023508197,0.00010775798,0.000032226002],"genre_scores_gemma":[0.99679184,0.0010185237,0.0016009546,0.000060699746,0.00024050925,0.00003243186,0.00007387714,0.00003679163,0.00014437268],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9981547,0.00019033972,0.0006517232,0.0002647158,0.00047916864,0.0002593542],"domain_scores_gemma":[0.9989599,0.0003485979,0.0002379865,0.00026468243,0.00007510862,0.0001137377],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004939831,0.00023568465,0.0003359659,0.00009741923,0.00014845139,0.00003522765,0.00011267186,0.00010505377,0.000053369666],"category_scores_gemma":[0.00026168794,0.00018669498,0.000049299924,0.00025064446,0.00017867732,0.00029181864,0.000031575168,0.00011279222,0.000013887808],"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":[0.00011937218,0.0000969931,0.00059575983,0.00017074274,0.000059592167,0.0000017338145,0.0032478767,0.0026087416,0.98749775,0.000009297026,0.0041997395,0.0013924073],"study_design_scores_gemma":[0.0014971958,0.00029476947,0.10783948,0.00013947922,0.00006700769,0.000044563116,0.0022639476,0.004295992,0.8807402,0.000055088927,0.0021398233,0.00062240544],"about_ca_topic_score_codex":0.00008136936,"about_ca_topic_score_gemma":0.000008345845,"teacher_disagreement_score":0.107243724,"about_ca_system_score_codex":0.00022045054,"about_ca_system_score_gemma":0.000034683613,"threshold_uncertainty_score":0.76132},"labels":[],"label_agreement":null},{"id":"W2033558742","doi":"10.1007/s00769-009-0522-y","title":"Report of the TrainMiC® Convention, held in Celje, Slovenia (10–14 January 2009)","year":2009,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Technology Assessment and Management","field":"Engineering","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":"123 Certification (Canada)","funders":"Joint Research Centre","keywords":"License; Convention; Interactivity; European commission; Commission; Position (finance); Political science; Medical education; Public relations; Focus (optics); Engineering; Business; Computer science; Law; Multimedia; European union; Medicine; Finance","score_opus":0.016889053888468248,"score_gpt":0.28005497479656005,"score_spread":0.2631659209080918,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2033558742","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":[0.97026324,0.00055647746,0.0074596615,0.004429349,0.00064766064,0.00037385046,0.000016735705,0.00019297567,0.01606002],"genre_scores_gemma":[0.9984769,0.00006761966,0.00050504605,0.00013816249,0.000022804441,0.000010384153,0.0000138701425,0.0000040333757,0.00076113455],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9992476,0.000043961147,0.00035429763,0.000120204546,0.00012748702,0.00010643387],"domain_scores_gemma":[0.9995986,0.000052451167,0.00009971,0.00019751825,0.000036512705,0.000015203848],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00046708164,0.00007893884,0.00013680433,0.000060493232,0.000028451228,0.000009382494,0.000097742966,0.00006325973,0.000016045438],"category_scores_gemma":[0.000088637426,0.00007063944,0.00003549412,0.000204478,0.00004166953,0.00015005641,0.000011412383,0.00010469191,0.0000028529537],"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":[0.00020906648,0.0008478621,0.15451989,0.0015991019,0.00035076047,0.00010890112,0.0051015406,0.008253723,0.057811867,0.21039169,0.41989943,0.14090618],"study_design_scores_gemma":[0.00058128784,0.000026377147,0.9745208,0.000053252923,0.000009717053,0.0000034862874,0.00022779558,0.00045538563,0.0021885394,0.006534764,0.0152738,0.00012480255],"about_ca_topic_score_codex":0.0000122531865,"about_ca_topic_score_gemma":0.00008111242,"teacher_disagreement_score":0.8200009,"about_ca_system_score_codex":0.000016437212,"about_ca_system_score_gemma":0.000011188294,"threshold_uncertainty_score":0.2880593},"labels":[],"label_agreement":null},{"id":"W2036990291","doi":"10.1007/s00769-002-0492-9","title":"Addressing issues associated with the development and management of PT programs and with their optimum use","year":2002,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Quality and Safety in Healthcare","field":"Health Professions","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":"University of British Columbia","funders":"National Institutes of Health","keywords":"Atlanta; Quality (philosophy); Library science; Political science; Engineering management; Engineering; Geography; Computer science; Metropolitan area","score_opus":0.376334339508065,"score_gpt":0.4494525600640289,"score_spread":0.0731182205559639,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2036990291","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":[0.9875043,0.0010805669,0.00056971353,0.009371102,0.00005057118,0.00089312263,0.000019675772,0.000050264454,0.000460701],"genre_scores_gemma":[0.99355894,0.00072930724,0.0043219463,0.00050647574,0.000029231305,0.00012172578,0.000033238626,0.0000119182705,0.0006872287],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99798894,0.00081439374,0.00044004983,0.00023547476,0.0002742649,0.0002468615],"domain_scores_gemma":[0.9980127,0.0011609107,0.00035865488,0.0001668652,0.00022248308,0.00007839369],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0014730971,0.00014193114,0.0002562603,0.00002854649,0.000779995,0.000035937865,0.00006286672,0.00009561067,0.000011258148],"category_scores_gemma":[0.00009468844,0.000083325176,0.000010334051,0.0001762104,0.00022111017,0.00026851974,0.000043167827,0.00023076311,0.0000010922557],"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":[0.0005460993,0.00027761233,0.61120397,0.0054055844,0.00051409827,0.0000071617956,0.24075341,0.000029148221,0.000022616732,0.015406342,0.0014380643,0.124395914],"study_design_scores_gemma":[0.0013215992,0.00013684096,0.94106317,0.0019297644,0.000029097704,0.000001144267,0.046560526,0.00027276055,0.00001855476,0.00010421549,0.008371311,0.00019099003],"about_ca_topic_score_codex":0.0001540628,"about_ca_topic_score_gemma":0.0016734889,"teacher_disagreement_score":0.32985926,"about_ca_system_score_codex":0.000026498912,"about_ca_system_score_gemma":0.000035935092,"threshold_uncertainty_score":0.599917},"labels":[],"label_agreement":null},{"id":"W2061797984","doi":"10.1007/s00769-002-0500-0","title":"Global Odyssey 2002: International conference on proficiency testing for medical laboratories","year":2002,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Quality and Safety in Healthcare","field":"Health Professions","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 British Columbia","funders":"","keywords":"Commercial law; Engineering; Medical laboratory; Engineering ethics; Library science; Medical physics; Management science; Political science; Computer science; Medicine; Pathology; Law","score_opus":0.30388766781366894,"score_gpt":0.5129245037421812,"score_spread":0.2090368359285123,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2061797984","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":[0.61421156,0.0008284766,0.015950523,0.23148488,0.010628458,0.0032754338,0.0028888274,0.000655821,0.12007605],"genre_scores_gemma":[0.9908347,0.00014495284,0.0020731823,0.0049688695,0.0008499783,0.00022981912,0.0000651197,0.000009396607,0.00082394615],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.99719644,0.00069661654,0.0007757387,0.00034262505,0.00063267647,0.00035592433],"domain_scores_gemma":[0.99522114,0.0030010531,0.00032103562,0.00018785121,0.0010722304,0.00019672063],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002128118,0.00014189733,0.00024366943,0.00003549124,0.0007692342,0.000027701628,0.00021123308,0.0002589408,0.00080999074],"category_scores_gemma":[0.012246939,0.00013124481,0.000031446925,0.00030816323,0.00012769792,0.00025259727,0.0000392686,0.00031901654,0.00009986273],"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":[0.00019359322,0.00020468752,0.22787611,0.000975509,0.000028005194,0.0000028203044,0.0064153103,0.00000950309,0.000035028785,0.711725,0.031759426,0.02077503],"study_design_scores_gemma":[0.0045055565,0.0006975058,0.866141,0.0014459492,0.000021605347,0.0000050769813,0.022096805,0.017506463,0.000017983602,0.022156438,0.06463735,0.00076823944],"about_ca_topic_score_codex":0.0001640524,"about_ca_topic_score_gemma":0.00048522372,"teacher_disagreement_score":0.6895685,"about_ca_system_score_codex":0.00009568594,"about_ca_system_score_gemma":0.00031987843,"threshold_uncertainty_score":0.9960733},"labels":[],"label_agreement":null},{"id":"W2074327302","doi":"10.1007/s00769-012-0899-x","title":"Using target measurement uncertainty to determine fitness for purpose","year":2012,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Pesticide Residue Analysis and Safety","field":"Agricultural and Biological Sciences","cited_by":16,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dynamic Systems Analysis (Canada)","funders":"","keywords":"Measurement uncertainty; Process (computing); Uncertainty analysis; Guard (computer science); Computer science; Range (aeronautics); Propagation of uncertainty; Data mining; Risk analysis (engineering); Engineering; Statistics; Simulation; Mathematics; Algorithm","score_opus":0.19003194181979322,"score_gpt":0.34748597376685303,"score_spread":0.1574540319470598,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2074327302","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":[0.99040926,0.00033338057,0.0069980877,0.0016877885,0.00013860532,0.00025583204,0.000072175535,0.000021918724,0.000082932835],"genre_scores_gemma":[0.9960089,0.00001665473,0.0030158483,0.00048004175,0.00037751827,0.00003092192,0.000040876286,7.2469624e-7,0.000028540077],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.99899393,0.00011287545,0.0002443609,0.00016869251,0.00024612754,0.00023400482],"domain_scores_gemma":[0.9991802,0.00032686634,0.00009349173,0.000043514774,0.0002119793,0.00014394429],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0013226034,0.00009349519,0.00017716658,0.000009846893,0.00019300649,0.00003770071,0.00007710895,0.00004438011,0.00004015831],"category_scores_gemma":[0.00055745506,0.000041318315,0.00006132123,0.00016935203,0.000019545243,0.00019684483,0.000019600377,0.000032378728,0.0000039653237],"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":[0.00027693008,0.00026752555,0.11777312,0.00011583843,0.00008028534,5.763438e-7,0.0008710322,0.0014057963,0.6401719,0.0032443076,0.0016845374,0.23410814],"study_design_scores_gemma":[0.00024249475,0.00008546912,0.9610606,0.00003098041,0.00003202907,0.0000014152942,0.0003913102,0.001688996,0.003338087,0.0007032159,0.032164026,0.000261367],"about_ca_topic_score_codex":0.00027367205,"about_ca_topic_score_gemma":0.0005844532,"teacher_disagreement_score":0.84328747,"about_ca_system_score_codex":0.000029194793,"about_ca_system_score_gemma":0.00000931073,"threshold_uncertainty_score":0.1684912},"labels":[],"label_agreement":null},{"id":"W2080442435","doi":"10.1007/s007690050442","title":"Reference materials in environmental trace organic analysis","year":2000,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Water Quality Monitoring and Analysis","field":"Environmental Science","cited_by":13,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Research Council Canada; Bundesministerium für Bildung und Forschung","keywords":"TRACE (psycholinguistics); Quality assurance; Environmental quality; Pollutant; Environmental science; Environmental analysis; Environmental chemistry; Engineering; Chemistry; Political science; Law","score_opus":0.025781274176230863,"score_gpt":0.2754081212659894,"score_spread":0.24962684708975855,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2080442435","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":[0.99866873,0.000041677267,0.00010285577,0.00037834214,0.000034412056,0.000062446874,0.000039589442,0.000027071166,0.0006449026],"genre_scores_gemma":[0.9977793,0.00014569481,0.00036484862,0.00007781993,0.000034321747,0.0000106684665,0.000059547743,0.000006172974,0.0015216152],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9984537,0.0003153932,0.00039430903,0.00036114093,0.00028334468,0.00019214164],"domain_scores_gemma":[0.9995158,0.00011226316,0.00008585542,0.00021325127,0.0000025378258,0.0000703088],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0007673095,0.00012332737,0.00025435927,0.000057898167,0.000091703136,0.000055099383,0.00012397085,0.00006682309,0.0080288965],"category_scores_gemma":[0.00003431988,0.000119047414,0.000042836473,0.00046798878,0.00009881164,0.00031766243,0.000026233345,0.00008238129,0.00039825318],"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":[0.00009603248,0.00026460824,0.5976948,0.000028204995,0.00015723481,0.000006881075,0.0036997737,0.0016885118,0.3744702,0.000097362,0.00027652222,0.02151986],"study_design_scores_gemma":[0.00025538943,0.000014985743,0.9856359,0.0000049848936,0.000058751193,6.2397214e-7,0.00026848493,0.00008919811,0.0116997715,0.00015726112,0.001634057,0.00018060001],"about_ca_topic_score_codex":0.0013208289,"about_ca_topic_score_gemma":0.000258763,"teacher_disagreement_score":0.38794106,"about_ca_system_score_codex":0.00006440371,"about_ca_system_score_gemma":0.0000038266007,"threshold_uncertainty_score":0.9928779},"labels":[],"label_agreement":null},{"id":"W2083486886","doi":"10.1007/s00769-007-0283-4","title":"The WADA Proficiency Testing Program as an integral part of the fight against doping in sport","year":2007,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Hormonal and reproductive studies","field":"Medicine","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":"World Anti-Doping Agency","funders":"","keywords":"Accreditation; Context (archaeology); Wada test; Agency (philosophy); Psychology; Medicine; Medical education; Epistemology; History; Philosophy","score_opus":0.060638034808893605,"score_gpt":0.37427991111969716,"score_spread":0.31364187631080354,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2083486886","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":[0.99264044,0.000428833,0.000014994674,0.0021189821,0.00021031799,0.00056898687,0.0000018221747,0.000023637553,0.0039920057],"genre_scores_gemma":[0.99882805,0.000054834058,0.0005392448,0.00019048172,0.00018184974,0.000022204005,0.00000485733,0.000004260464,0.00017422583],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9989787,0.000057084428,0.00036210494,0.00018758979,0.00025370982,0.00016084485],"domain_scores_gemma":[0.9991349,0.00025958283,0.0001565823,0.00016960147,0.0002450617,0.000034268116],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0017652521,0.000078547804,0.0001611819,0.000019375018,0.00019707665,0.000011602426,0.00006948043,0.000031052852,6.8238336e-7],"category_scores_gemma":[0.0013984425,0.000039046732,0.00003199192,0.00033330108,0.00021658625,0.0000966589,0.000024813146,0.00015499123,7.112854e-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":[0.00022106832,0.00021691706,0.8474601,0.00012991916,0.000022201124,0.000004108021,0.0029201962,0.000008761191,0.0033953602,0.0032273566,0.00009214014,0.14230186],"study_design_scores_gemma":[0.00028043747,0.00013223317,0.9882433,0.00014187631,0.0000070382102,0.0000033661975,0.0048385486,0.00005298655,0.0022572093,0.0002910461,0.003693998,0.00005798304],"about_ca_topic_score_codex":0.00012583355,"about_ca_topic_score_gemma":0.0003383315,"teacher_disagreement_score":0.14224388,"about_ca_system_score_codex":0.000017992641,"about_ca_system_score_gemma":0.00007420125,"threshold_uncertainty_score":0.16741674},"labels":[],"label_agreement":null},{"id":"W2086650703","doi":"10.1007/s00769-013-1031-6","title":"TDRM/TDLM Workshop at the AOAC INTERNATIONAL Annual Meeting 2013","year":2014,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Meteorological Phenomena and Simulations","field":"Earth and Planetary 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":"National Research Council Canada","funders":"","keywords":"Commercial law; Library science; Engineering; Analytical Chemistry (journal); Political science; Computer science; Chemistry; Environmental chemistry; Law","score_opus":0.03974803267383914,"score_gpt":0.28924179587303483,"score_spread":0.2494937631991957,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2086650703","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":[0.96373963,0.0003249733,0.0018036603,0.011409904,0.00074832386,0.0001222535,0.00011971242,0.000044599245,0.021686964],"genre_scores_gemma":[0.99622697,0.00006695387,0.0006473568,0.0013098583,0.0003292066,0.0000022058043,0.0001695854,0.0000017164302,0.0012461797],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99879915,0.0003196289,0.00025563355,0.00020730535,0.00026514646,0.00015316659],"domain_scores_gemma":[0.99742305,0.0021641355,0.00012371018,0.00013264471,0.00008470241,0.000071733666],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0012367918,0.00008671306,0.000110746456,0.000024917119,0.00034793385,0.00007038695,0.00016051866,0.000051560808,0.0011956533],"category_scores_gemma":[0.00092352636,0.00005637593,0.00003451998,0.00010607385,0.0001129069,0.00026506028,0.000018093122,0.00010057298,0.00017005845],"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":[0.00019496086,0.000035176712,0.7739771,0.00002828244,0.00006268036,9.447369e-7,0.0033056538,0.05525273,0.0001602382,0.0063374653,0.027752334,0.13289244],"study_design_scores_gemma":[0.00026067728,0.000024427538,0.8963652,0.0000067435135,0.0000053201593,0.0000010163172,0.0004700035,0.019717703,0.000009672641,0.0026614254,0.0803696,0.000108166685],"about_ca_topic_score_codex":0.00035616173,"about_ca_topic_score_gemma":0.0025164334,"teacher_disagreement_score":0.13278428,"about_ca_system_score_codex":0.0000035680441,"about_ca_system_score_gemma":0.000008870588,"threshold_uncertainty_score":0.9997174},"labels":[],"label_agreement":null},{"id":"W2092953325","doi":"10.1007/s00769-014-1049-4","title":"Harmonization and transferability of performance assessment: experience from four serum aluminum proficiency testing schemes","year":2014,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Pesticide Residue Analysis and Safety","field":"Agricultural and Biological 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":"Institut National de Santé Publique du Québec","funders":"","keywords":"Harmonization; Standard deviation; Transferability; Computer science; Statistics; Reliability engineering; Mathematics; Engineering","score_opus":0.06021817124645472,"score_gpt":0.288018029644113,"score_spread":0.22779985839765826,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2092953325","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":[0.9956561,0.000079990874,0.0032645704,0.00053874095,0.000031772088,0.000099580575,0.00003432825,0.000023459954,0.00027148647],"genre_scores_gemma":[0.99788755,0.00008409515,0.0018409138,0.00007088882,0.00006324115,0.000010143716,0.000031746335,6.5839544e-7,0.000010752218],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9988434,0.00018885662,0.00034339575,0.00027626564,0.00022151532,0.00012662284],"domain_scores_gemma":[0.9987531,0.0008132897,0.00014808553,0.000058470523,0.00017175797,0.000055296838],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00068402145,0.000096380514,0.00020743157,0.000008775947,0.00019722634,0.000041391122,0.000088602384,0.000052342428,0.000029579505],"category_scores_gemma":[0.00049826555,0.000047511658,0.000026643667,0.0002525834,0.000107553264,0.00032476065,0.000020033533,0.0000658288,5.833051e-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":[0.000009538206,0.000027243033,0.7874461,0.000033498436,0.00000533129,6.1965295e-8,0.0002695072,0.000018115526,0.1111872,0.0005179799,0.00000187962,0.10048358],"study_design_scores_gemma":[0.0001269924,0.00008164254,0.9830018,0.000031849984,0.000010343283,4.1789943e-7,0.00031413592,0.010062861,0.0057502147,0.00033378764,0.00018034092,0.000105606465],"about_ca_topic_score_codex":0.0002895623,"about_ca_topic_score_gemma":0.00015716643,"teacher_disagreement_score":0.19555575,"about_ca_system_score_codex":0.000006815407,"about_ca_system_score_gemma":0.000010820281,"threshold_uncertainty_score":0.19374692},"labels":[],"label_agreement":null},{"id":"W2897236559","doi":"10.1007/s00769-018-1349-1","title":"Cooperation in publicly funded reference material production","year":2018,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Scientific Measurement and Uncertainty Evaluation","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":true,"ca_institutions":"National Research Council Canada","funders":"International Atomic Energy Agency; Gobierno del Principado de Asturias","keywords":"Certification; Production (economics); European commission; Commercial law; Accounting; Business; Commission; Library science; Engineering; Political science; Computer science; European union; Finance; Economics; Law","score_opus":0.47412113293856345,"score_gpt":0.48009324608865606,"score_spread":0.005972113150092606,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2897236559","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":[0.9872082,0.00001857243,0.00237847,0.0037145808,0.0029870523,0.00030076696,0.000013803271,0.000037518425,0.0033410692],"genre_scores_gemma":[0.99674946,0.000008084464,0.0006953608,0.00015122624,0.00044985558,0.000024725525,0.00005727189,0.0000037710315,0.0018602433],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9964805,0.00080550014,0.0007115392,0.0005736738,0.0012578331,0.00017095226],"domain_scores_gemma":[0.9975343,0.00037457322,0.00027871793,0.0003470403,0.0014085679,0.000056797526],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0140173985,0.000096941476,0.00017182135,0.00022785577,0.00022687449,0.0005554028,0.00020955884,0.0000710196,0.00060035154],"category_scores_gemma":[0.011079132,0.00007915795,0.000019126384,0.000917503,0.00015745572,0.0013106464,0.000027947659,0.000070165785,0.00018488363],"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":[0.0012579786,0.0003040952,0.31251508,0.000051084946,0.000024112363,6.782307e-7,0.008946513,0.00017888764,0.2531482,0.07917584,0.10096176,0.24343579],"study_design_scores_gemma":[0.00066634663,0.00011866787,0.9333077,0.00003134918,0.0000054931684,8.5529257e-7,0.0016364087,0.0026657956,0.019307418,0.01907002,0.02293152,0.00025842615],"about_ca_topic_score_codex":0.00029003058,"about_ca_topic_score_gemma":0.005882592,"teacher_disagreement_score":0.6207926,"about_ca_system_score_codex":0.00004256524,"about_ca_system_score_gemma":0.00007674599,"threshold_uncertainty_score":0.997251},"labels":[],"label_agreement":null},{"id":"W2900128974","doi":"10.1007/s00769-018-1357-1","title":"A proficiency testing scheme to evaluate the effectiveness of laboratory sample reduction of a soil sample","year":2018,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Pesticide Residue Analysis and Safety","field":"Agricultural and Biological Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Environment and Climate Change Canada","keywords":"Sample (material); Accreditation; Homogeneity (statistics); Computer science; Engineering; Reliability engineering; Statistics; Mathematics; Chemistry; Chromatography; Medical education; Medicine","score_opus":0.060860853854059795,"score_gpt":0.3278827809757692,"score_spread":0.26702192712170936,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2900128974","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":[0.9983333,0.00006591973,0.0008135709,0.00036278507,0.000052773765,0.00017330964,0.000098186916,0.000011907085,0.00008824382],"genre_scores_gemma":[0.99851084,0.000008514598,0.0012954674,0.000037947782,0.00011148609,0.00001649496,0.000016192964,5.2834963e-7,0.0000025078723],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.99863327,0.0005762006,0.00027845515,0.00017585771,0.00022970932,0.000106521635],"domain_scores_gemma":[0.99551225,0.003581404,0.00014719625,0.00006716314,0.0006559132,0.000036095556],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0029134925,0.000067777204,0.00017266505,0.000011848198,0.00015934599,0.000014985137,0.00010304509,0.000033463573,0.00001919722],"category_scores_gemma":[0.005505436,0.000026442603,0.000034959627,0.0007074044,0.00013486108,0.000099127785,0.000026611106,0.000040616702,0.0000017685234],"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":[0.000116666306,0.000044087556,0.06707725,0.000057540452,0.000015279229,3.1396553e-8,0.00025374946,0.000042681964,0.9199462,0.0016578231,0.000025175656,0.01076355],"study_design_scores_gemma":[0.00009283109,0.00021571743,0.953791,0.00005749803,0.000013118691,2.929173e-7,0.00037249044,0.00033902944,0.043378633,0.0015787305,0.000095557196,0.000065084845],"about_ca_topic_score_codex":0.0024140112,"about_ca_topic_score_gemma":0.000499143,"teacher_disagreement_score":0.88671374,"about_ca_system_score_codex":0.000008628593,"about_ca_system_score_gemma":0.000026922839,"threshold_uncertainty_score":0.6590919},"labels":[],"label_agreement":null},{"id":"W2964970392","doi":"10.1007/s00769-019-01397-5","title":"Non-carcinogenic health risk assessment of nitrate in bottled drinking waters sold in Iranian markets: A Monte Carlo simulation","year":2019,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Water Treatment and Disinfection","field":"Environmental Science","cited_by":65,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Hospital for Sick Children","keywords":"Hazard quotient; Nitrate; Bottled water; Health risk assessment; Health risk; Toxicology; Environmental health; Risk assessment; Environmental science; Human health; Environmental chemistry; Chemistry; Medicine; Environmental engineering; Biology","score_opus":0.0177469213794428,"score_gpt":0.3059776679947836,"score_spread":0.2882307466153408,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2964970392","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":[0.9982958,0.000020966187,0.00039404875,0.00019145878,0.00010877749,0.0004444179,0.000007819524,0.000012387722,0.0005243482],"genre_scores_gemma":[0.9993323,0.000051552073,0.00044967938,0.00006681315,0.000010233817,0.00001611934,0.000014644861,0.0000069017437,0.000051773655],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99863595,0.00027110113,0.00042072282,0.00026556983,0.00022306602,0.0001835722],"domain_scores_gemma":[0.99942696,0.00014652232,0.00023670579,0.00013608432,0.000010739188,0.000043003893],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010384886,0.00011098543,0.00021750592,0.00007541169,0.000047250574,0.00002368395,0.00005174744,0.00004981536,0.000050641404],"category_scores_gemma":[0.000024964793,0.00010389145,0.00003578184,0.00022067782,0.000031954678,0.00047817014,0.000024873532,0.00009589535,0.000013341738],"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":[0.00005677047,0.000070920774,0.94010967,0.00003930833,0.0000057961997,4.6715894e-7,0.0019204373,0.055503733,0.0011193659,0.000008075387,0.000008702969,0.0011567471],"study_design_scores_gemma":[0.0014837282,0.000068772104,0.94425553,0.000050870385,0.000004237735,1.8395575e-7,0.00028948218,0.053454753,0.0001511718,0.00012386264,0.000017289123,0.000100126694],"about_ca_topic_score_codex":0.0062802443,"about_ca_topic_score_gemma":0.004273937,"teacher_disagreement_score":0.0041458476,"about_ca_system_score_codex":0.00022623742,"about_ca_system_score_gemma":0.00002058583,"threshold_uncertainty_score":0.9493893},"labels":[],"label_agreement":null},{"id":"W3028443178","doi":"10.1007/s00769-020-01433-9","title":"Development of low and elevated level multivitamin and mineral supplement certified reference materials: VITA-1 and VITB-1","year":2020,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Vitamin D Research Studies","field":"Medicine","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":"National Research Council Canada","funders":"","keywords":"Certified reference materials; Multivitamin; Analyte; Chemistry; Mathematics; Chromatography; Vitamin; Detection limit","score_opus":0.18094198931547836,"score_gpt":0.3795725873741518,"score_spread":0.19863059805867342,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3028443178","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":[0.993608,0.0003338871,0.00048465192,0.004942837,0.000036121317,0.00037902335,0.00017390214,0.000020678779,0.000020889094],"genre_scores_gemma":[0.9899907,0.00033640937,0.00917436,0.00023446488,0.000042196953,0.00003578522,0.00009799947,0.000009193489,0.00007889784],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.9985936,0.00014060286,0.00045398183,0.00033129437,0.00029585906,0.00018462785],"domain_scores_gemma":[0.99911916,0.0003128668,0.0001339354,0.00009212176,0.00018076402,0.0001611412],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006484495,0.0001536815,0.00039390652,0.000050149036,0.000108273445,0.000033192424,0.000034610268,0.000060244638,0.000020616895],"category_scores_gemma":[0.0009398512,0.00012890238,0.0000101415,0.00010029519,0.00019838284,0.00014468592,0.000105924635,0.000091885966,0.0000012484812],"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":[0.00040793265,0.00007876123,0.021488143,0.0025838576,0.00012897662,0.000005338581,0.010504879,3.6749657e-7,0.95694315,0.00019969868,0.00086944725,0.0067894696],"study_design_scores_gemma":[0.0024424314,0.00016081857,0.897956,0.00017369576,0.000021974633,0.0000039290744,0.0023294252,0.00012074114,0.09577358,0.000020067508,0.00085072895,0.000146596],"about_ca_topic_score_codex":0.00006185814,"about_ca_topic_score_gemma":0.00010339402,"teacher_disagreement_score":0.8764679,"about_ca_system_score_codex":0.00001584479,"about_ca_system_score_gemma":0.000061435945,"threshold_uncertainty_score":0.52564865},"labels":[],"label_agreement":null},{"id":"W4284689332","doi":"10.1007/s00769-022-01505-y","title":"Interlaboratory comparisons of chemical measurements: Quo Vadis?","year":2022,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Scientific Measurement and Uncertainty Evaluation","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":"National Research Council Canada","funders":"National Research Council Canada","keywords":"Statistician; Criticism; The arts; Status quo; Value (mathematics); Management science; Engineering; Statistics; Political science; Mathematics; Law","score_opus":0.42892369295191496,"score_gpt":0.4588551741003888,"score_spread":0.029931481148473826,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284689332","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":[0.98867506,0.00035989002,0.0049849553,0.0018302987,0.0016670222,0.00026414584,0.000096918186,0.000030137942,0.0020915614],"genre_scores_gemma":[0.99849033,0.0000020120117,0.0008579201,0.00021297572,0.000051253388,0.000036521673,0.000036300396,0.0000043461873,0.0003083133],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.99415654,0.00123888,0.0009073043,0.00041136556,0.0031366975,0.0001492001],"domain_scores_gemma":[0.9969607,0.0010874673,0.00057529245,0.0003750339,0.0009252079,0.00007630932],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016087262,0.00009654352,0.00027284608,0.00016791916,0.00022414954,0.00009429925,0.00042917437,0.000029212659,0.0008217727],"category_scores_gemma":[0.005410426,0.00008695487,0.00006952009,0.0008851249,0.0001337962,0.0003316128,0.00010061581,0.00012917152,0.000021820479],"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":[0.0004690162,0.00051462,0.25994653,0.00005110784,0.00009068228,4.4397393e-7,0.008466104,0.0007389776,0.37556103,0.013946839,0.3037987,0.036415942],"study_design_scores_gemma":[0.0050633275,0.00042592848,0.65835035,0.00007301855,0.000107427426,0.0000026398495,0.046828773,0.012675615,0.0777082,0.025111184,0.17252666,0.0011268661],"about_ca_topic_score_codex":0.000038595965,"about_ca_topic_score_gemma":0.000030516923,"teacher_disagreement_score":0.39840382,"about_ca_system_score_codex":0.00006231952,"about_ca_system_score_gemma":0.00015524315,"threshold_uncertainty_score":0.89978343},"labels":[],"label_agreement":null},{"id":"W4304614705","doi":"10.1007/s00769-022-01515-w","title":"The units of rate constants in chemical kinetics","year":2022,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Spectroscopy and Quantum Chemical Studies","field":"Physics and Astronomy","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":"Open Text (Canada)","funders":"","keywords":"Reaction rate constant; Arrhenius equation; Logarithm; Thermodynamics; Constant (computer programming); Base (topology); Rate equation; Kinetics; Chemistry; Mathematics; Physical chemistry; Computer science; Physics; Activation energy; Mathematical analysis","score_opus":0.03169094056643566,"score_gpt":0.3058511319264604,"score_spread":0.27416019136002473,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4304614705","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":[0.99681544,0.00019738032,0.00018061846,0.00070662145,0.000092471775,0.00007110101,0.00008936548,0.0000053074887,0.0018416999],"genre_scores_gemma":[0.9997587,0.000015998572,0.000056848476,0.000045469427,0.000036714937,0.000025734176,0.000020096815,0.0000027141543,0.000037678594],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.99935746,0.00011811775,0.00021827391,0.00009496888,0.0001089586,0.000102218655],"domain_scores_gemma":[0.9990731,0.0006902684,0.0001012119,0.00007206101,0.000045860113,0.000017529823],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00034763402,0.00005318406,0.00011053863,0.000010198036,0.00012305441,0.000009463163,0.00007349286,0.000008564742,0.00003514756],"category_scores_gemma":[0.00008448793,0.000043298176,0.000016012924,0.0002024265,0.000107197775,0.000032547872,0.000050992658,0.000109952336,4.6567007e-7],"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":[0.0003435947,0.0002606099,0.16590476,0.000046892106,0.00008348414,7.5107204e-7,0.002453912,0.00009264569,0.3661874,0.45042536,0.0023889127,0.011811697],"study_design_scores_gemma":[0.0038177473,0.00017974063,0.27752766,0.00004003526,0.00002912968,8.715547e-7,0.021352947,0.00063140586,0.5905339,0.0661073,0.03923461,0.0005446459],"about_ca_topic_score_codex":0.000090046226,"about_ca_topic_score_gemma":0.000008409865,"teacher_disagreement_score":0.38431805,"about_ca_system_score_codex":0.000010334212,"about_ca_system_score_gemma":0.000021581662,"threshold_uncertainty_score":0.17656484},"labels":[],"label_agreement":null},{"id":"W4396736053","doi":"10.1007/s00769-024-01595-w","title":"New statistical framework for interlaboratory evaluation of anti-doping testing results by WADA","year":2024,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","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":"World Anti-Doping Agency; National Research Council Canada","funders":"National Research Council Canada","keywords":"Commercial law; Statistical analysis; Analytical Chemistry (journal); Statistics; Engineering; Accounting; Mathematics; Chemistry; Business; Chromatography; Political science; Law","score_opus":0.6683357042775011,"score_gpt":0.6183950536970836,"score_spread":0.04994065058041752,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4396736053","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":[0.014699875,0.0003583006,0.9808011,0.00083028094,0.0010868971,0.0005583538,0.0013321026,0.00008637159,0.00024670057],"genre_scores_gemma":[0.20809717,0.000014735003,0.79140276,0.00007077685,0.00031506686,0.0000329947,0.00002355491,0.000017875715,0.000025055253],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9964415,0.001159846,0.0012327273,0.00039640407,0.0005981366,0.0001714226],"domain_scores_gemma":[0.7899643,0.20857729,0.00037617018,0.00024737575,0.0007231969,0.000111621586],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.012614057,0.00013616536,0.00043033445,0.000047002148,0.000060847655,0.00008185132,0.00010107673,0.00014874042,0.000028441358],"category_scores_gemma":[0.4549669,0.00012626823,0.000046378325,0.00029556692,0.00009493798,0.00015792531,0.000024494804,0.00020661566,0.000002962785],"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":[0.00016512605,0.000050864048,0.0001614411,0.0012827307,0.000081153434,6.177023e-7,0.0006361542,0.00001071952,0.0017176615,0.7095304,0.03091426,0.25544888],"study_design_scores_gemma":[0.00091609353,0.0001138222,0.003184166,0.0007781668,0.00013684318,5.083043e-7,0.00018944386,0.009015632,0.0014416766,0.9834502,0.0006137119,0.0001597688],"about_ca_topic_score_codex":0.000029972536,"about_ca_topic_score_gemma":0.000006137993,"teacher_disagreement_score":0.44235286,"about_ca_system_score_codex":0.000040514828,"about_ca_system_score_gemma":0.00028424297,"threshold_uncertainty_score":0.54962415},"labels":[],"label_agreement":null},{"id":"W4403718788","doi":"10.1007/s00769-024-01621-x","title":"Production and certification of a freeze-dried reference material for human urinary steroids to ensure quality of longitudinal profiling data","year":2024,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Hormonal and reproductive studies","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"World Anti-Doping Agency","keywords":"Certification; Profiling (computer programming); Certified reference materials; Commercial law; Quality assurance; Chromatography; Chemistry; Engineering; Process engineering; External quality assessment; Computer science; Operations management; Management; Political science; Economics","score_opus":0.267195335066961,"score_gpt":0.4401583683803643,"score_spread":0.1729630333134033,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403718788","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":[0.9944534,0.0006139497,0.0013274138,0.0020807195,0.00026794247,0.0006295088,0.0005277552,0.000029960516,0.00006940675],"genre_scores_gemma":[0.996323,0.00006809863,0.002706737,0.000009660175,0.00031030475,0.000043049276,0.0004087194,0.000007623447,0.00012282838],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.99862075,0.00011767994,0.00048573926,0.00047858886,0.0002083641,0.00008889408],"domain_scores_gemma":[0.9989468,0.00016085184,0.00014459822,0.0003246051,0.0003845372,0.000038630675],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0015863352,0.00009488074,0.00030661136,0.00006494361,0.000086751745,0.000015949323,0.000056921,0.00004711173,0.0000030379451],"category_scores_gemma":[0.0011735174,0.000078188605,0.000021939684,0.00013783893,0.00014452913,0.00020751255,0.000060916635,0.00006494892,4.4907642e-7],"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":[0.0011345026,0.00013824011,0.025841076,0.0047246264,0.00012224357,4.945513e-7,0.0017345282,0.0000018109773,0.95368683,0.0059873243,0.00033384006,0.006294491],"study_design_scores_gemma":[0.00028556434,0.00036687052,0.9060807,0.00027134054,0.00006246256,0.000004402674,0.001011213,0.00002952818,0.09072853,0.0007703922,0.00028861663,0.00010035982],"about_ca_topic_score_codex":0.00012827618,"about_ca_topic_score_gemma":0.000063850704,"teacher_disagreement_score":0.88023967,"about_ca_system_score_codex":0.000012019648,"about_ca_system_score_gemma":0.00004295056,"threshold_uncertainty_score":0.3188439},"labels":[],"label_agreement":null}]}