{"id":"W2079886100","doi":"10.1111/j.1365-2929.2006.02566.x","title":"Assessing the mini‐Clinical Evaluation Exercise in comparison to a national specialty examination","year":2006,"lang":"en","type":"article","venue":"Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":82,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary; University of British Columbia","funders":"","keywords":"Medicine; Specialty; Oral examination; Competence (human resources); Objective structured clinical examination; Physical examination; Family medicine; Clinical clerkship; Medical education; Internal medicine; Curriculum; Psychology","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":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01010505,0.0001608632,0.0003084274,0.0004688599,0.0001594164,0.0000889482,0.0002126235,0.0002760979,0.001099531],"category_scores_gemma":[0.01812901,0.0001281021,0.00006774034,0.001602713,0.0001729896,0.0002671963,0.00003885027,0.0006217678,0.000163794],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009175325,"about_ca_system_score_gemma":0.005209362,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002105879,"about_ca_topic_score_gemma":0.0001529508,"domain_scores_codex":[0.9946714,0.0004239191,0.00125869,0.0004266789,0.00294574,0.0002735888],"domain_scores_gemma":[0.997672,0.0002911519,0.000257453,0.0003338668,0.001340041,0.0001054414],"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":[0.00002144772,0.002463094,0.05689779,0.00004649131,0.000007057668,8.662369e-7,0.0005981264,0.00003973757,0.00002567231,0.001141084,0.2110619,0.7276968],"study_design_scores_gemma":[0.0009409411,0.00004953788,0.9273151,0.0006171697,0.00006824717,0.00001523981,0.001425509,0.03126975,0.00005381067,0.0009904328,0.03711648,0.0001377849],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9359448,0.0002276684,0.001602982,0.04445027,0.002757621,0.001428555,7.204133e-7,0.00004870433,0.01353871],"genre_scores_gemma":[0.9839967,0.00001362932,0.004141405,0.007804848,0.002487027,0.0005033998,0.0004591231,0.00002120112,0.0005726972],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8704173,"threshold_uncertainty_score":0.9998136,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07693715332789765,"score_gpt":0.5110928111106555,"score_spread":0.4341556577827579,"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."}}