{"id":"W2072489006","doi":"10.1097/00001888-200510001-00017","title":"A Comparison of Physician Examiners and Trained Assessors in a High-Stakes OSCE Setting","year":2005,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":42,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medical education; Educational measurement; Psychology; MEDLINE; Medicine; Family medicine; Curriculum; Pedagogy; Political science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01820093,0.0004527051,0.0005360543,0.000970619,0.0006132714,0.00146697,0.0006603338,0.0006725009,0.001588617],"category_scores_gemma":[0.0717126,0.0004364463,0.0005459205,0.0004204732,0.0008089696,0.001136117,0.001427921,0.0005857883,0.0005869371],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001075262,"about_ca_system_score_gemma":0.001367227,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002677741,"about_ca_topic_score_gemma":0.006412252,"domain_scores_codex":[0.9800445,0.01325025,0.001106221,0.001261968,0.003340691,0.0009963179],"domain_scores_gemma":[0.9447014,0.02987753,0.008270605,0.0030759,0.008427007,0.005647539],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01830022,0.003992666,0.888653,0.0003639541,0.0005975941,0.0002832295,0.009295889,0.0005057197,0.00329539,0.0002009817,0.001214596,0.07329684],"study_design_scores_gemma":[0.001437389,0.02568436,0.9611794,0.0000995754,0.0002653428,0.0003177985,0.005849104,0.001966953,0.001710242,0.0001482945,0.001273443,0.00006806941],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9986418,0.0001186758,0.000334005,0.00004178685,0.0000272344,0.0001174721,0.00004434891,0.00001017246,0.0006644867],"genre_scores_gemma":[0.9987207,0.00007535826,0.0005540624,0.00005177155,0.00002362727,0.0001058495,0.00008295238,0.000003492878,0.0003821483],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01820093,"threshold_uncertainty_score":0.09625691,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04484784662818801,"score_gpt":0.4129022770139917,"score_spread":0.3680544303858037,"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."}}