{"id":"W4390743131","doi":"10.1007/s00192-023-05717-9","title":"Summative Evaluation of Vaginal Surgery Skills: Setting A Pass–Fail Score","year":2024,"lang":"en","type":"article","venue":"International Urogynecology Journal","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Centre for Advancing Health Outcomes; St. Paul's Hospital; University of British Columbia","funders":"Royal College of Physicians and Surgeons of Canada; Faculty of Medicine, University of British Columbia","keywords":"Medicine; Summative assessment; Checklist; Competence (human resources); Randomized controlled trial; Medical physics; Surgery; Formative assessment","routes":{"ca_aff":true,"ca_fund":true,"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":[0.01069165,0.001300251,0.002091967,0.004089653,0.002136023,0.001885835,0.001526951,0.001731199,0.005627036],"category_scores_gemma":[0.0314377,0.0004902895,0.003003171,0.001125326,0.001350966,0.001684832,0.004443745,0.001638319,0.002706941],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001771621,"about_ca_system_score_gemma":0.004192745,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002585457,"about_ca_topic_score_gemma":0.01028995,"domain_scores_codex":[0.9889439,0.002444217,0.001882365,0.0005865403,0.004477116,0.001665763],"domain_scores_gemma":[0.9721937,0.005348574,0.00402353,0.0008637955,0.0136023,0.003968167],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003546157,0.004218399,0.7536206,0.001082212,0.0007842166,0.0004752118,0.004943561,0.002883134,0.006287476,0.001698311,0.02462225,0.1958384],"study_design_scores_gemma":[0.0003605872,0.006613852,0.9415717,0.0007389073,0.0004937761,0.0005060451,0.006919796,0.01187909,0.01404707,0.002365093,0.0141714,0.0003327067],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9416695,0.0005520455,0.0123605,0.00142675,0.0007507442,0.003689167,0.001869316,0.0006278439,0.03705421],"genre_scores_gemma":[0.9527749,0.0004810386,0.0298547,0.0003771324,0.0002189206,0.007169149,0.002602139,0.0001314113,0.006390579],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01069165,"threshold_uncertainty_score":0.05654359,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05272038689652565,"score_gpt":0.3733568012228358,"score_spread":0.3206364143263102,"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."}}