{"id":"W2611560891","doi":"10.1097/sla.0000000000002284","title":"How Do Thresholds of Principle and Preference Influence Surgeon Assessments of Learner Performance?","year":2017,"lang":"en","type":"article","venue":"Annals of Surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Institutes of Health Research; Western University","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Competence (human resources); Subjectivity; Variation (astronomy); Medical education; Preference; Competency assessment; Social psychology; Psychology","routes":{"ca_aff":true,"ca_fund":true,"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":[0.01830291,0.0001760188,0.0003169529,0.0009071114,0.0005578402,0.002898626,0.0005435271,0.0005631571,0.001981277],"category_scores_gemma":[0.1331302,0.0001740871,0.0003871056,0.0006201519,0.00153852,0.001387876,0.001435978,0.0006090321,0.0003641973],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001052321,"about_ca_system_score_gemma":0.00149103,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002016429,"about_ca_topic_score_gemma":0.002887194,"domain_scores_codex":[0.9773278,0.013643,0.001423787,0.001456646,0.005272177,0.0008765401],"domain_scores_gemma":[0.8689805,0.09864807,0.01753225,0.003309838,0.007847948,0.003681493],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000482893,0.0002801376,0.8972201,0.0002167345,0.000196729,0.0001207325,0.02517254,0.0005751752,0.00187103,0.0006493672,0.0005701003,0.07264449],"study_design_scores_gemma":[0.00002685875,0.0008161225,0.9612401,0.0002219267,0.00008585772,0.0002791109,0.0294496,0.002393536,0.001789741,0.001651078,0.001970956,0.00007509906],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9929125,0.0002153354,0.002170603,0.0003396997,0.0000229305,0.0000329763,0.00003982726,0.00001457691,0.0042515],"genre_scores_gemma":[0.9991019,0.00004098358,0.0006390678,0.0000440461,0.000004787773,0.00001662638,0.00001569518,0.000004435726,0.0001326175],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01830291,"threshold_uncertainty_score":0.09679615,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4047410585289761,"score_gpt":0.4208013525918248,"score_spread":0.0160602940628487,"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."}}