{"id":"W2155284013","doi":"10.3109/0142159x.2013.760036","title":"Script concordance testing: From theory to practice: AMEE Guide No. 75","year":2013,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":190,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; McGill University","funders":"","keywords":"Concordance; Competence (human resources); Medical education; Psychology; Test (biology); Variety (cybernetics); Medicine; Applied psychology; Management science; Computer science; Social psychology; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03957295,0.002776596,0.002437146,0.009943039,0.001781932,0.003227139,0.005697398,0.003019031,0.03781322],"category_scores_gemma":[0.0903068,0.003253404,0.001967192,0.006380985,0.003269189,0.004228826,0.005250779,0.005437382,0.03715592],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002923307,"about_ca_system_score_gemma":0.009711466,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005212881,"about_ca_topic_score_gemma":0.01202006,"domain_scores_codex":[0.9632095,0.01981326,0.005968344,0.001028601,0.009316521,0.0006637641],"domain_scores_gemma":[0.9339103,0.03835218,0.003319648,0.004338433,0.01849259,0.001586881],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002391266,0.0008179986,0.001481989,0.002213133,0.00004091003,0.0004308468,0.001829409,0.001472335,0.001855327,0.005864332,0.3683816,0.615373],"study_design_scores_gemma":[0.0004225899,0.0007115481,0.01178774,0.008371766,0.00006786917,0.00375016,0.002445717,0.009130801,0.005201018,0.03506812,0.9226785,0.0003642424],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.007499217,0.009140234,0.7671439,0.01977149,0.002824833,0.05171328,0.01050515,0.02974019,0.1016617],"genre_scores_gemma":[0.004891022,0.00528662,0.9368182,0.002106956,0.000239391,0.02720049,0.003752785,0.001963471,0.01774117],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.03957295,"threshold_uncertainty_score":0.2092843,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03335734072977935,"score_gpt":0.3691587632072861,"score_spread":0.3358014224775068,"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."}}