{"id":"W2022664256","doi":"10.1097/acm.0b013e3181b37f37","title":"Reliability and Acceptance of the Mini-Clinical Evaluation Exercise as a Performance Assessment of Practicing Physicians","year":2009,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":20,"is_retracted":false,"has_abstract":true,"ca_institutions":"St. Paul's Hospital","funders":"","keywords":"Generalizability theory; Reliability (semiconductor); Medicine; Inter-rater reliability; Psychology; Physical therapy; Family medicine; Rating scale","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":[0.02667372,0.000318079,0.0004150063,0.0008937053,0.0002729818,0.0008293343,0.0004534916,0.0005361062,0.0006769116],"category_scores_gemma":[0.09649204,0.000223123,0.0004935511,0.0004300245,0.0006849435,0.0006543346,0.0008506168,0.0004754254,0.0003355164],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000331766,"about_ca_system_score_gemma":0.0004206847,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000786548,"about_ca_topic_score_gemma":0.0007863175,"domain_scores_codex":[0.9797643,0.0102671,0.001758347,0.001239673,0.006524782,0.0004457603],"domain_scores_gemma":[0.9066535,0.0561549,0.01254739,0.004464326,0.01868119,0.001498633],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001009919,0.000286273,0.9446255,0.0001469734,0.0001883232,0.00006665794,0.00360438,0.0004740425,0.002785039,0.0001113701,0.0005962662,0.04610524],"study_design_scores_gemma":[0.00005530578,0.001604964,0.9903746,0.0000917291,0.00005319608,0.0003473935,0.001083268,0.003755266,0.001561951,0.0000804514,0.0009676895,0.00002428286],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9954855,0.0002721187,0.002018409,0.00009541386,0.00004422025,0.0001157057,0.00006632936,0.00003331864,0.001868952],"genre_scores_gemma":[0.9979548,0.00009240959,0.001548964,0.00003516281,0.00002265962,0.00007173808,0.00008627548,0.000006357457,0.0001816143],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02667372,"threshold_uncertainty_score":0.1410658,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05387971244302858,"score_gpt":0.4793478086345647,"score_spread":0.4254680961915361,"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."}}