{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009216536,0.00008936055,0.000521532,0.0001201736,0.00006431553,0.00001926069,0.00008080481,0.00001619818,0.0007986435],"category_scores_gemma":[0.0002829243,0.00005032818,0.0002169658,0.0004844599,0.0001456145,0.00006772635,0.00001264016,0.0002587768,0.00002835388],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005464735,"about_ca_system_score_gemma":0.00009674008,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001493734,"about_ca_topic_score_gemma":2.52712e-7,"domain_scores_codex":[0.9988239,0.0001283455,0.0004892154,0.00007797394,0.0003080817,0.0001725195],"domain_scores_gemma":[0.998403,0.0009333329,0.0002017997,0.0001324986,0.0001238344,0.000205599],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00224803,0.0002586995,0.3603251,0.000006757859,0.0002189113,0.0006476679,0.0006521553,0.005589376,0.001980479,0.0000664402,0.0003316383,0.6276748],"study_design_scores_gemma":[0.001096812,0.0003614183,0.9762871,0.0002829611,0.0001781177,0.00247432,0.001256234,0.0005186982,0.0003422753,0.00006743986,0.01688446,0.0002501816],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.994367,0.00002779354,0.0007919124,0.002657297,0.00008811307,0.00005421133,8.29567e-7,0.00001387945,0.001998949],"genre_scores_gemma":[0.9930493,0.00001696756,0.001779113,0.00476837,0.0002286807,6.298629e-7,3.543306e-7,0.00001306851,0.0001435139],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6274245,"threshold_uncertainty_score":0.8744586,"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."}}