{"id":"W2007544525","doi":"10.1097/sla.0b013e318160b371","title":"Toward Feasible, Valid, and Reliable Video-Based Assessments of Technical Surgical Skills in the Operating Room","year":2008,"lang":"en","type":"article","venue":"Annals of Surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":177,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Medicine; Checklist; Rating scale; Reliability (semiconductor); Summative assessment; Medical physics; Scale (ratio); Physical therapy; Statistics; Psychology; Formative assessment","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":[0.001466885,0.0001046658,0.0004733383,0.0001504142,0.00004527886,0.000007464551,0.00005733725,0.00009396494,0.0001000125],"category_scores_gemma":[0.0004920859,0.00007113534,0.0001470179,0.0004032722,0.0001085291,0.00007843185,0.00002198149,0.0001756308,0.000001608053],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000006675691,"about_ca_system_score_gemma":0.0001258185,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006221592,"about_ca_topic_score_gemma":0.000001354759,"domain_scores_codex":[0.9985605,0.0001138458,0.000543975,0.0001730424,0.0004032867,0.000205364],"domain_scores_gemma":[0.9974359,0.002053492,0.0001199932,0.0001783049,0.0001306904,0.00008156564],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002487793,0.0006935757,0.9921286,0.0001160116,0.00003044792,0.0004581287,0.0001400653,0.0003918922,0.0002379594,0.0001821292,0.001007101,0.004365375],"study_design_scores_gemma":[0.002567315,0.0001774356,0.9836734,0.0006601426,0.00002449831,0.0001594305,0.0001556928,0.001186661,0.005064178,0.00009370062,0.006066429,0.0001711304],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9913749,0.0003003437,0.00002739214,0.001466981,0.00003460394,0.0001570201,0.000003978762,0.00001856685,0.006616247],"genre_scores_gemma":[0.9984546,0.0002465367,0.0003024602,0.0008721928,0.0000347739,0.00000863174,0.00001530456,0.000009930533,0.00005561188],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008455149,"threshold_uncertainty_score":0.2900815,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3246438670497029,"score_gpt":0.4228627877583974,"score_spread":0.09821892070869448,"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."}}