{"id":"W2330673540","doi":"10.1097/acm.0000000000000643","title":"Comparability of the National Board of Medical Examiners Comprehensive Clinical Science Examination and a Set of Five Clinical Science Subject Examinations","year":2015,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Health Sciences Centre","funders":"","keywords":"Subject (documents); Summative assessment; United States Medical Licensing Examination; Formative assessment; Medical education; Comparability; Medicine; Test (biology); Clinical clerkship; Educational measurement; Family medicine; Psychology; Medical school; Mathematics education; Curriculum; Computer science; Library science; Pedagogy; Mathematics","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":[0.01520601,0.000322074,0.000342543,0.001943094,0.0004088006,0.00126435,0.0006912005,0.0005555911,0.001635121],"category_scores_gemma":[0.07490251,0.0002103531,0.0006057153,0.001549231,0.0007193413,0.0007303603,0.001803711,0.0008384719,0.0004184086],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001095293,"about_ca_system_score_gemma":0.001426595,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004145279,"about_ca_topic_score_gemma":0.005282624,"domain_scores_codex":[0.9836869,0.007232863,0.001790909,0.001778382,0.00487621,0.0006348697],"domain_scores_gemma":[0.9567247,0.01773438,0.01203235,0.004380678,0.006092288,0.003035746],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002567886,0.0000782327,0.9934492,0.00001662836,0.0001507242,0.00001522154,0.0001812905,0.0001026679,0.0003080799,0.00007112478,0.0001342517,0.005235828],"study_design_scores_gemma":[0.00001170876,0.0002716703,0.9988285,0.000008335226,0.0000160088,0.00003382189,0.0000839074,0.0002485267,0.0002006027,0.00004070379,0.000253001,0.00000318245],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9973685,0.0001711314,0.0006475276,0.00007309034,0.00004261375,0.0000710136,0.0003349474,0.00001063594,0.001280651],"genre_scores_gemma":[0.9984316,0.00003452633,0.0003613764,0.00002724883,0.00001655173,0.0000406462,0.0008620107,0.000005641602,0.0002203451],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01520601,"threshold_uncertainty_score":0.08041805,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2497812939107412,"score_gpt":0.5220355605604755,"score_spread":0.2722542666497343,"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."}}