{"id":"W2970977809","doi":"10.11157/fohpe.v20i2.321","title":"Script concordance test examinations: Student perception and study approaches","year":2019,"lang":"en","type":"article","venue":"Focus on Health Professional Education A Multi-Professional Journal","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Concordance; Construct (python library); Multiple choice; Test (biology); Psychology; Construct validity; Medical education; Perception; Psychometrics; Medicine; Clinical psychology; Computer science; Significant difference; Internal medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01441225,0.0003824605,0.0004514991,0.002504545,0.001005932,0.001857955,0.0007520699,0.0004951721,0.004122312],"category_scores_gemma":[0.03838274,0.000326588,0.0005897583,0.001543986,0.001082444,0.001298578,0.002754049,0.0008319388,0.0008051152],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001608303,"about_ca_system_score_gemma":0.001856671,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001225053,"about_ca_topic_score_gemma":0.001142997,"domain_scores_codex":[0.9890251,0.005763202,0.000985131,0.0009829538,0.002484223,0.0007594229],"domain_scores_gemma":[0.9746626,0.01236895,0.00359491,0.001118079,0.006795655,0.001459861],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005934635,0.003062254,0.8131278,0.0001663181,0.00004797366,0.0003604667,0.1094027,0.0003991099,0.002637847,0.0008884092,0.0009302514,0.06838344],"study_design_scores_gemma":[0.00009935665,0.00597457,0.83827,0.0001689164,0.00005651543,0.001000148,0.1345284,0.005567374,0.006124885,0.001841084,0.00626686,0.0001018449],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9959444,0.00002448738,0.001849654,0.00005123646,0.000009619622,0.0003595053,0.00008306012,0.0000158812,0.001661989],"genre_scores_gemma":[0.9960472,0.00003296316,0.002059577,0.00004730634,0.000008992388,0.0007974625,0.0001069014,0.00001133679,0.0008881544],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01441225,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1071179608798967,"score_gpt":0.4778008864963227,"score_spread":0.370682925616426,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}