{"id":"W2902682793","doi":"10.1080/10401334.2018.1503961","title":"Comparison of the Ottawa Surgical Competency Operating Room Evaluation (O-SCORE) to a Single-Item Performance Score","year":2018,"lang":"en","type":"article","venue":"Teaching and Learning in Medicine","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"University of Ottawa","keywords":"Orthopedic surgery; Physical therapy; Medicine; Competence (human resources); Correlation; Psychology; Surgery; Social psychology","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004109918,0.0001319415,0.0004177278,0.0001356285,0.0003736291,0.00001127843,0.00007979207,0.00006620714,0.0001788271],"category_scores_gemma":[0.003568195,0.00008195337,0.0000330521,0.0002886169,0.0001543752,0.00004823791,0.00004566919,0.001002124,0.000004491073],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005408667,"about_ca_system_score_gemma":0.00005058455,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001412326,"about_ca_topic_score_gemma":0.00003067454,"domain_scores_codex":[0.9978725,0.0006316997,0.0005071008,0.0002384629,0.000543456,0.00020682],"domain_scores_gemma":[0.9987461,0.0007159271,0.0001405318,0.0001583296,0.0001296521,0.0001094401],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007607739,0.00007122638,0.8075873,0.000046123,0.0000101946,0.000002338265,0.01197543,0.002383852,0.0009462197,0.0001332625,0.00002940013,0.1767386],"study_design_scores_gemma":[0.003877537,0.001381819,0.6832759,0.002248783,0.00005869276,0.00003199378,0.002402489,0.2984402,0.0001517284,0.000007263077,0.00801439,0.0001091498],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9773681,0.0001965879,0.0001362153,0.001565891,0.0001573714,0.0002930081,1.300944e-7,0.00004728575,0.02023543],"genre_scores_gemma":[0.998362,0.000002118485,0.0006768754,0.000310013,0.0003668189,0.000005959665,0.000008824853,0.00001383045,0.0002535696],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2960564,"threshold_uncertainty_score":0.4353787,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08601178017627288,"score_gpt":0.3859940064628766,"score_spread":0.2999822262866038,"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."}}