{"id":"W2073773377","doi":"10.1007/s10459-008-9120-8","title":"Script concordance testing: more cases or more questions?","year":2008,"lang":"en","type":"article","venue":"Advances in Health Sciences Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":66,"is_retracted":false,"has_abstract":false,"ca_institutions":"Université de Montréal","funders":"","keywords":"Concordance; Reliability (semiconductor); Variance (accounting); Nested set model; Statistics; Variance components; Test (biology); Computer science; Mathematics; Medicine; Data mining; 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":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09128761,0.001497085,0.003159797,0.0048402,0.001930547,0.006045871,0.004273367,0.005098892,0.0149932],"category_scores_gemma":[0.3499948,0.001235277,0.001923007,0.004138329,0.004484294,0.01753588,0.004595222,0.005064367,0.001572045],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002634256,"about_ca_system_score_gemma":0.003226308,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00239863,"about_ca_topic_score_gemma":0.004476503,"domain_scores_codex":[0.8896981,0.05681746,0.02132609,0.008682345,0.0205672,0.00290873],"domain_scores_gemma":[0.5571458,0.3408896,0.03055017,0.03163378,0.03185929,0.007921323],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00292981,0.001153631,0.3684811,0.002552041,0.0006603405,0.00422516,0.005576663,0.00186833,0.004953936,0.02140443,0.04453908,0.5416554],"study_design_scores_gemma":[0.001175026,0.002015701,0.3976166,0.00787784,0.002504647,0.08261196,0.02161886,0.06940091,0.02482216,0.2966457,0.0927874,0.0009231056],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6484504,0.0108287,0.1561898,0.1331176,0.005432718,0.001299833,0.003759746,0.002535977,0.0383852],"genre_scores_gemma":[0.8717662,0.001372959,0.1045532,0.0151599,0.002573225,0.0005817219,0.001551337,0.0005772508,0.00186414],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09128761,"threshold_uncertainty_score":0.4827808,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0905657237337662,"score_gpt":0.4756656871024329,"score_spread":0.3850999633686667,"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."}}