{"id":"W2530440400","doi":"10.1016/j.amjsurg.2016.07.033","title":"Multicenter proficiency benchmarks for advanced laparoscopic suturing tasks","year":2016,"lang":"en","type":"article","venue":"The American Journal of Surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":30,"is_retracted":false,"has_abstract":false,"ca_institutions":"McGill University Health Centre","funders":"","keywords":"Multicenter study; Computer science; Medicine; Medical physics; Surgery; Randomized controlled trial","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":[0.0006470236,0.00008250402,0.0003760846,0.00009446419,0.00005996646,0.00000838744,0.00007120435,0.00001301844,0.0001334855],"category_scores_gemma":[0.0005124669,0.00003705302,0.0002220671,0.0001474532,0.000167663,0.00008896266,0.00001163717,0.00009455142,0.000003490289],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003498296,"about_ca_system_score_gemma":0.00007441352,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000001693975,"about_ca_topic_score_gemma":1.394606e-7,"domain_scores_codex":[0.9990716,0.00005124495,0.0003815807,0.0000838193,0.0002022351,0.000209529],"domain_scores_gemma":[0.9973925,0.001776602,0.0004456372,0.0001214901,0.0001497256,0.0001140243],"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":[0.003363369,0.0001538152,0.3086229,0.00002902748,0.0001419431,0.00004513508,0.000334297,0.00004678015,0.009032416,0.00006408922,0.0003579024,0.6778083],"study_design_scores_gemma":[0.01240018,0.001095826,0.9289706,0.001936693,0.0002871797,0.0003004755,0.002352898,0.0002733159,0.01698744,0.0002618942,0.03471801,0.0004155248],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9954579,0.0001056426,0.0008709219,0.00271061,0.0004038509,0.0001103957,0.00000195092,0.000009135703,0.0003295478],"genre_scores_gemma":[0.9984796,0.00006276704,0.000415177,0.0006095969,0.0003172065,0.000004339806,8.485899e-7,0.00001134757,0.00009911412],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6773928,"threshold_uncertainty_score":0.1510978,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03599969755057482,"score_gpt":0.3177345069750498,"score_spread":0.281734809424475,"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."}}