{"id":"W4319082716","doi":"10.1007/s40670-023-01729-8","title":"Considering the Structured Oral Examinations Beyond Its Psychometrics Properties","year":2023,"lang":"en","type":"article","venue":"Medical Science Educator","topic":"Innovations in Medical Education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":false,"ca_institutions":"Centre Hospitalier Universitaire de Sherbrooke; Université de Sherbrooke","funders":"","keywords":"Psychometrics; Psychology; Educational measurement; Medical education; Medicine; Clinical psychology; Pedagogy; Curriculum","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003106286,0.0001246853,0.0001718183,0.0007462581,0.0006935224,0.00008857998,0.0005585199,0.0001054888,0.001285768],"category_scores_gemma":[0.03218042,0.00007290849,0.00003730834,0.01039873,0.001742406,0.0002007363,0.000124428,0.0005064009,0.0003379071],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002089653,"about_ca_system_score_gemma":0.00317585,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002143099,"about_ca_topic_score_gemma":0.000006956361,"domain_scores_codex":[0.9963565,0.00004321773,0.0003533763,0.0003650904,0.002362607,0.0005191971],"domain_scores_gemma":[0.9984815,0.0001617875,0.00007469866,0.0004337422,0.0005007328,0.000347517],"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":[0.00002227591,0.0006351052,0.02896568,0.0002294457,0.00005381987,0.00003296192,0.01046036,0.000005997449,0.01501858,0.02233493,0.3226326,0.5996083],"study_design_scores_gemma":[0.001627061,0.0003066269,0.635519,0.0004403984,0.0001183666,0.0004705234,0.02391024,0.01310039,0.01580484,0.002740478,0.3053186,0.0006435643],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9202388,0.0001276957,0.00008694258,0.07230918,0.004542072,0.0004058295,0.00000125585,0.0001556334,0.002132573],"genre_scores_gemma":[0.9914062,0.00003690643,0.0009037389,0.005227575,0.0006921897,0.0001050751,0.00001084672,0.00001413795,0.001603339],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6065533,"threshold_uncertainty_score":0.9996272,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08053409116000604,"score_gpt":0.3870177223022442,"score_spread":0.3064836311422382,"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."}}