{"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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0007408609,0.0001492681,0.0005435714,0.000238093,0.00004530017,0.000006661598,0.00003593356,0.0001050949,0.001225253],"category_scores_gemma":[0.0005305507,0.00009857098,0.0006536867,0.0001404383,0.00006326104,0.0001151595,0.00001887369,0.00003539994,0.0001247526],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000009706334,"about_ca_system_score_gemma":0.0001504888,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003478895,"about_ca_topic_score_gemma":0.000001454902,"domain_scores_codex":[0.9988081,0.00004027629,0.0003977267,0.0002215028,0.000179429,0.0003529361],"domain_scores_gemma":[0.9961211,0.003133379,0.0001471575,0.000237186,0.0002131635,0.0001480183],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002494296,0.0001063077,0.785772,0.0001589577,0.0001592484,0.00005886897,0.00002307218,6.098369e-9,0.0007086933,0.0001467895,0.137575,0.07504168],"study_design_scores_gemma":[0.001189357,0.0002871912,0.4453985,0.002128614,0.0003810106,0.00007374458,0.00004513078,0.00003782145,0.4258677,0.0004976904,0.1234712,0.0006220619],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9889014,0.004824137,0.0001155407,0.003705211,0.0002203999,0.0002470547,0.00006509368,0.00002779587,0.001893386],"genre_scores_gemma":[0.9965277,0.001300237,0.0002476527,0.0005219879,0.0001281575,0.00006868647,0.0001066493,0.00002612853,0.001072746],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.425159,"threshold_uncertainty_score":0.9996878,"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."}}