{"id":"W2039358902","doi":"10.1080/01421590802334309","title":"Clerkship evaluation–what are we measuring?","year":2009,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Innovations in Medical Education","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Objective structured clinical examination; Principal (computer security); Educational measurement; Medical education; Multiple choice; Computer science; Psychology; Medicine; Curriculum; Pedagogy; Internal medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003234918,0.0001551498,0.0002671689,0.0001589615,0.00009589287,0.00004043884,0.0001859484,0.0003195127,0.01571848],"category_scores_gemma":[0.007681239,0.000125563,0.00008059041,0.0005329864,0.0001383559,0.000194426,0.00002114079,0.0008311434,0.000326066],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002905545,"about_ca_system_score_gemma":0.0004478147,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000003706285,"about_ca_topic_score_gemma":0.000001873397,"domain_scores_codex":[0.9961613,0.00015362,0.0004344413,0.0003184619,0.002613295,0.000318865],"domain_scores_gemma":[0.9987447,0.00002234491,0.0001059642,0.0004576376,0.0003908177,0.0002785139],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003486621,0.0007685476,0.01291266,0.00003730432,0.00003655237,0.00002511579,0.0007812434,0.000001284044,0.00002458794,0.0005390272,0.1317054,0.8531334],"study_design_scores_gemma":[0.01001817,0.0007844504,0.3800941,0.004980241,0.0007042989,0.0004892612,0.008519489,0.08690634,0.0003078968,0.01530134,0.4908458,0.001048552],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6763838,0.002409548,0.0007599872,0.2939183,0.00141765,0.0006625921,1.859253e-7,0.0002387737,0.02420918],"genre_scores_gemma":[0.9818481,0.0002085551,0.0006265293,0.01188701,0.00170857,0.00005637476,0.00004017386,0.00002109982,0.003603542],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8520849,"threshold_uncertainty_score":0.9851813,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1147198611655236,"score_gpt":0.4022657078693233,"score_spread":0.2875458467037997,"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."}}