{"id":"W2191794480","doi":"10.1002/bjs.10047","title":"Setting pass scores for assessment of technical performance by surgical trainees","year":2015,"lang":"en","type":"article","venue":"British journal of surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":28,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mount Sinai Hospital; University of Toronto; St. Michael's Hospital","funders":"","keywords":"Medicine; Consistency (knowledge bases); Set (abstract data type); Artificial intelligence","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":[0.01679719,0.001019998,0.0008074378,0.004612778,0.0005694776,0.001066329,0.001563331,0.0007916964,0.004854178],"category_scores_gemma":[0.0513416,0.0002636171,0.001407564,0.001955484,0.001061772,0.0009122039,0.002098002,0.0008763098,0.002365569],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001334838,"about_ca_system_score_gemma":0.001710408,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001609503,"about_ca_topic_score_gemma":0.002591412,"domain_scores_codex":[0.9798263,0.00792079,0.003218613,0.0008905276,0.007657086,0.0004865662],"domain_scores_gemma":[0.9647648,0.01291262,0.008934544,0.002413659,0.009565195,0.00140907],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003199565,0.0006546582,0.8324483,0.0009074819,0.000623147,0.0002211224,0.001460402,0.004403829,0.005178437,0.002450395,0.01025756,0.1381951],"study_design_scores_gemma":[0.0003192441,0.007363666,0.9575453,0.0003115459,0.000223425,0.001063651,0.001132788,0.01139936,0.01088538,0.001975869,0.007635977,0.0001438109],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9044505,0.0007356387,0.06158849,0.0003716655,0.0002640238,0.00234494,0.005038753,0.0009289001,0.02427717],"genre_scores_gemma":[0.9243971,0.0002481768,0.06543932,0.0001050484,0.00005008915,0.002872387,0.004013818,0.000140069,0.002733981],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01679719,"threshold_uncertainty_score":0.08883309,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08092526935615033,"score_gpt":0.3448260392858218,"score_spread":0.2639007699296715,"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."}}