{"id":"W2401432848","doi":"10.1016/j.surg.2016.03.030","title":"Simulation-based summative assessments in surgery","year":2016,"lang":"en","type":"article","venue":"Surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Medicine; Summative assessment; Surgery; General surgery; Medical physics; Formative assessment; Mathematics education","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0006753267,0.0001085684,0.0003487648,0.0003130275,0.00002893292,0.000009289098,0.00002052041,0.0000717017,0.001307992],"category_scores_gemma":[0.002154876,0.00007267081,0.0001649074,0.0003438285,0.00004113818,0.0001080254,0.000007827668,0.00007002526,0.0001123565],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008692809,"about_ca_system_score_gemma":0.0002127564,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000011987,"about_ca_topic_score_gemma":0.000005213532,"domain_scores_codex":[0.9988182,0.0001005533,0.0003732847,0.0001994636,0.0002675111,0.0002409912],"domain_scores_gemma":[0.9798149,0.01974938,0.00008425117,0.0001557276,0.00007991656,0.0001158039],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001936689,0.0001021704,0.9202669,0.00001924359,0.00001771018,0.00008732349,0.00001564922,0.001916893,0.00004068628,0.00003794199,0.0001328856,0.07716897],"study_design_scores_gemma":[0.002122631,0.0000177597,0.9600679,0.0007180753,0.00001685521,8.547472e-7,0.00003584282,0.02293327,0.0005206879,0.0002249542,0.01310436,0.0002367722],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9921206,0.00003130234,0.002978249,0.0008316085,0.0002852158,0.0001416873,0.000005661303,0.00009827716,0.003507358],"genre_scores_gemma":[0.9989159,0.000004419523,0.00006610196,0.0006028367,0.00007590661,0.00001265054,0.00002560762,0.00001689646,0.0002797367],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0769322,"threshold_uncertainty_score":0.9996049,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09315326538017576,"score_gpt":0.3724215118978444,"score_spread":0.2792682465176687,"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."}}