{"id":"W2114700362","doi":"10.1002/bjs.9168","title":"Error rating tool to identify and analyse technical errors and events in laparoscopic surgery","year":2013,"lang":"en","type":"article","venue":"British journal of surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":84,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Medicine; Intraclass correlation; Interquartile range; Reliability (semiconductor); Rating scale; Surgery; Statistics; Psychometrics","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":[0.001368334,0.00009701391,0.0006004746,0.0003622533,0.00005146097,0.00005594641,0.00003037985,0.00008738244,0.0001796732],"category_scores_gemma":[0.001920572,0.00009925039,0.0001178559,0.0003255956,0.0000407841,0.0002890994,0.00003216559,0.0003040259,0.000005769856],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000322392,"about_ca_system_score_gemma":0.0000760511,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001245875,"about_ca_topic_score_gemma":0.00003213261,"domain_scores_codex":[0.9984311,0.00009374639,0.0008043328,0.0001529056,0.0002929207,0.000224957],"domain_scores_gemma":[0.9983537,0.000992354,0.000194004,0.00006399566,0.0001230643,0.0002729071],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005448458,0.0000688695,0.9106075,0.00004005457,0.00002920641,0.0007026137,0.00003851366,0.00001108875,0.0005846524,9.744358e-7,0.0009275321,0.08693454],"study_design_scores_gemma":[0.0008120657,0.00001946316,0.9952003,0.001203352,0.00002695399,0.002068714,0.0001431365,0.00007023052,0.000050645,0.00004429009,0.0002575331,0.0001032573],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9982898,0.0008220142,0.00002144711,0.0004947003,0.0001506289,0.0001288706,0.000001333258,0.00001243794,0.00007872886],"genre_scores_gemma":[0.998848,0.0002120007,0.0003027389,0.0004768135,0.00009639589,0.000005012977,0.000003125165,0.00001353528,0.00004240366],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08683129,"threshold_uncertainty_score":0.4047314,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05528367630974348,"score_gpt":0.338047804865489,"score_spread":0.2827641285557455,"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."}}