{"id":"W2499069400","doi":"10.1097/sla.0000000000001849","title":"Defining Benchmarks for Major Liver Surgery","year":2016,"lang":"en","type":"article","venue":"Annals of Surgery","topic":"Hepatocellular Carcinoma Treatment and Prognosis","field":"Medicine","cited_by":221,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto General Hospital","funders":"","keywords":"Medicine; Complication; Incidence (geometry); Surgery; Hepatectomy; Percentile; Resection","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":["insufficient_payload"],"domain":null,"study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03427722,0.001775796,0.001071448,0.00396705,0.001381231,0.005218152,0.002447933,0.001232938,0.001649072],"category_scores_gemma":[0.08204462,0.0002740711,0.0005842335,0.003091373,0.002088343,0.00379096,0.005998038,0.001860991,0.0006959978],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00442348,"about_ca_system_score_gemma":0.00825467,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002530465,"about_ca_topic_score_gemma":0.002222999,"domain_scores_codex":[0.9796274,0.008527531,0.002465607,0.001125766,0.00655229,0.001701447],"domain_scores_gemma":[0.948279,0.01343525,0.01475473,0.002401045,0.01582715,0.00530283],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000992829,0.000753402,0.412774,0.00253811,0.0002710927,0.0003201847,0.002845445,0.02427329,0.00205756,0.09218519,0.05802812,0.4029608],"study_design_scores_gemma":[0.0002286687,0.004586419,0.528698,0.008411495,0.0002543063,0.001433318,0.0133154,0.0417546,0.008275131,0.2605653,0.1320017,0.000475604],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5209651,0.04546132,0.2525612,0.05961132,0.003641331,0.001488662,0.005589813,0.002031555,0.1086497],"genre_scores_gemma":[0.9258082,0.002751849,0.06181785,0.002725315,0.0007157908,0.0009672489,0.003941102,0.0002032122,0.001069429],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03427722,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4735158460163808,"score_gpt":0.3170297145963276,"score_spread":0.1564861314200532,"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."}}