{"id":"W1121875846","doi":"10.1016/j.jsurg.2015.04.029","title":"The Generic Error Rating Tool: A Novel Approach to Assessment of Performance and Surgical Education in Gynecologic Laparoscopy","year":2015,"lang":"en","type":"article","venue":"Journal of surgical education","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":44,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Inter-rater reliability; Intraclass correlation; Laparoscopy; Medicine; Interquartile range; Intra-rater reliability; Reliability (semiconductor); Rating scale; Surgery; Medical physics; Psychometrics; Statistics; Mathematics; Confidence interval; Internal medicine","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":[0.001625642,0.000101873,0.0003093594,0.0001553391,0.00006137957,0.00003258447,0.00007767865,0.00007366487,0.00001762581],"category_scores_gemma":[0.0003436529,0.00006386255,0.00006293216,0.000395497,0.00005609349,0.0001465593,0.00002403131,0.000269201,0.000001138192],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001757174,"about_ca_system_score_gemma":0.001846203,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000009187584,"about_ca_topic_score_gemma":0.000001234391,"domain_scores_codex":[0.998549,0.000094313,0.0006678601,0.0001312303,0.0003999312,0.0001576652],"domain_scores_gemma":[0.9985331,0.0003730535,0.000324867,0.0001066153,0.0004252859,0.0002370564],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001910414,0.0074598,0.4850421,0.0002226472,0.00009139526,0.00001562898,0.002593349,0.008994801,0.0005749519,0.009984505,0.0009365444,0.4821739],"study_design_scores_gemma":[0.008609168,0.001464755,0.7744979,0.0004414742,0.0001048591,0.001222448,0.00787562,0.0345001,0.0001813397,0.0001840384,0.1706168,0.000301446],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9843457,0.0003423303,0.0001449815,0.00135816,0.000446489,0.0002478391,3.683567e-7,0.000005138087,0.01310894],"genre_scores_gemma":[0.9918488,0.0001173842,0.007325522,0.0001429076,0.0002851698,0.00001662796,0.000005694803,0.000007864253,0.000250049],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4818724,"threshold_uncertainty_score":0.3275084,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.070803474136236,"score_gpt":0.3849640659900897,"score_spread":0.3141605918538537,"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."}}