{"id":"W2042099895","doi":"10.1016/s0002-9610(00)00470-0","title":"Using operative outcome to assess technical skill","year":2000,"lang":"en","type":"article","venue":"The American Journal of Surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":98,"is_retracted":false,"has_abstract":false,"ca_institutions":"University Health Network; University of Toronto","funders":"","keywords":"Inter-rater reliability; Competence (human resources); Medicine; Reliability (semiconductor); Quality Score; Product (mathematics); Quality (philosophy); Physical therapy; Operations management; Medical physics; Psychology; Statistics; Rating scale; Mathematics; Engineering","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":[0.002517736,0.0005918948,0.0006027303,0.003657392,0.0005359838,0.001174203,0.0005624643,0.0007897553,0.003606274],"category_scores_gemma":[0.01169757,0.000121858,0.0008337803,0.001170574,0.0008046121,0.001704382,0.001109585,0.001079915,0.001066681],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005253274,"about_ca_system_score_gemma":0.0007118057,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001293341,"about_ca_topic_score_gemma":0.002686237,"domain_scores_codex":[0.9984463,0.0003267229,0.0002935156,0.0001137758,0.0006231397,0.0001966643],"domain_scores_gemma":[0.9882638,0.003790579,0.003663399,0.0003994468,0.002173826,0.001708794],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000296558,0.0003065702,0.9867293,0.00002721469,0.00007653336,0.00004332759,0.00009075491,0.0001792225,0.0004292922,0.00008943774,0.0003312631,0.01140054],"study_design_scores_gemma":[0.00004105749,0.001558101,0.9925076,0.00007045533,0.0001481464,0.0004108591,0.0005631529,0.002176563,0.001038909,0.0006420782,0.0008142392,0.0000287012],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9853547,0.0002830762,0.001789136,0.0002391293,0.000113475,0.00007308844,0.0005804852,0.00004675283,0.01152021],"genre_scores_gemma":[0.9979141,0.0001163514,0.0005191636,0.00005030768,0.00002804312,0.00004943246,0.0004283728,0.00001213427,0.0008821181],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003657392,"threshold_uncertainty_score":0.0133152,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.206647817255585,"score_gpt":0.434047351774382,"score_spread":0.227399534518797,"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."}}