{"id":"W2088823622","doi":"10.1097/acm.0b013e3181d73fb6","title":"Toward Authentic Clinical Evaluation: Pitfalls in the Pursuit of Competency","year":2010,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":222,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mount Sinai Hospital; University Health Network; University of Toronto; The Wilson Centre","funders":"","keywords":"Psychology; Medical education; MEDLINE; Representation (politics); Standardized test; Medicine; Mathematics education; Political science","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2865293,0.000921629,0.001380937,0.004023833,0.007812702,0.01326287,0.004693992,0.003622789,0.002595888],"category_scores_gemma":[0.4028822,0.0009075164,0.0006729233,0.002304313,0.04682999,0.01012201,0.01350404,0.008341069,0.001082441],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01186148,"about_ca_system_score_gemma":0.02396601,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008001185,"about_ca_topic_score_gemma":0.01022449,"domain_scores_codex":[0.554543,0.378454,0.0160451,0.007994333,0.04076562,0.002198031],"domain_scores_gemma":[0.4841181,0.3872444,0.03106475,0.02979463,0.05699167,0.01078652],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0005527057,0.0004080014,0.02310062,0.005150202,0.0001323104,0.00108784,0.3028372,0.001277562,0.001699154,0.1767243,0.02915098,0.4578792],"study_design_scores_gemma":[0.0003218511,0.001459953,0.04133037,0.02281168,0.0001146809,0.006814231,0.3433001,0.009694519,0.004529313,0.3813893,0.1874956,0.0007384612],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.2315881,0.03420243,0.3062378,0.3126387,0.003483362,0.002537486,0.0002373277,0.0009498899,0.1081248],"genre_scores_gemma":[0.8974474,0.004043784,0.07951836,0.01430155,0.0006481626,0.001897894,0.00006809458,0.0001602293,0.001914611],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.2865293,"threshold_uncertainty_score":0.8798364,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1178613056261369,"score_gpt":0.4775186042921444,"score_spread":0.3596572986660075,"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."}}