{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002099138,0.00008289785,0.0005184065,0.00009926318,0.00003902918,0.00002380578,0.00005545725,0.00008062526,0.00005912621],"category_scores_gemma":[0.0006951143,0.00007834407,0.0002769065,0.0001442958,0.00007476167,0.0001407349,0.00001340179,0.000238675,3.800991e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005299546,"about_ca_system_score_gemma":0.0002774664,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000005007403,"about_ca_topic_score_gemma":6.570776e-7,"domain_scores_codex":[0.9984044,0.00005075418,0.0007693794,0.0000971804,0.0004814909,0.0001967729],"domain_scores_gemma":[0.9979093,0.001014163,0.000340935,0.00005795344,0.0004225776,0.0002550543],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008691653,0.0007019314,0.5441869,0.0002811437,0.0001753094,0.001243534,0.00006910657,0.0002701653,0.0006760033,0.00005903862,0.02664854,0.4248192],"study_design_scores_gemma":[0.0158889,0.0009789794,0.8604302,0.004423865,0.0002322806,0.02086873,0.0005424522,0.004506746,0.001313857,0.0001255813,0.09023337,0.0004550016],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9961025,0.001014656,0.0005041559,0.0003751853,0.0001988204,0.00010612,0.00001516417,0.00001688834,0.001666564],"genre_scores_gemma":[0.9976557,0.0002092672,0.001677127,0.00008962795,0.000274301,0.00000357294,0.0000177299,0.00001500544,0.00005759277],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4243642,"threshold_uncertainty_score":0.3194779,"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."}}