{"id":"W1981905415","doi":"10.1007/s00464-003-8157-2","title":"Toward reliable operative assessment: the reliability and feasibility of videotaped assessment of laparoscopic technical skills","year":2004,"lang":"en","type":"article","venue":"Surgical Endoscopy","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":145,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta; University of Toronto; McMaster University","funders":"","keywords":"Medicine; Reliability (semiconductor); Abdominal surgery; Hepatology; Medical physics; Laparoscopy; Endoscopy; Laparoscopic surgery; General surgery; Surgery","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0601557,0.0004855361,0.0003813826,0.001702478,0.0005210646,0.001944711,0.001612412,0.001525815,0.0008936406],"category_scores_gemma":[0.2964376,0.0005172666,0.0004204576,0.0007288015,0.001902896,0.002611265,0.00154328,0.001563732,0.0004888684],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007451281,"about_ca_system_score_gemma":0.001844704,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002571137,"about_ca_topic_score_gemma":0.003343188,"domain_scores_codex":[0.9567972,0.02679625,0.003251581,0.001713095,0.01094737,0.0004943624],"domain_scores_gemma":[0.6282803,0.2616571,0.02239709,0.01651307,0.06929107,0.001861385],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.007036971,0.001389797,0.6379892,0.001111742,0.0004997989,0.0003611005,0.01478371,0.004892917,0.02823541,0.003091033,0.00371262,0.2968957],"study_design_scores_gemma":[0.0007516075,0.009226469,0.8483889,0.001144484,0.0005442075,0.002478639,0.007823632,0.08042099,0.03284436,0.005612965,0.01044392,0.0003197761],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9364913,0.001356067,0.05169919,0.0005963718,0.0002993895,0.0009372037,0.0002580502,0.000160327,0.008202114],"genre_scores_gemma":[0.9641325,0.000494551,0.03395338,0.0002213356,0.00009204027,0.0003929908,0.0001689465,0.00004146563,0.0005028205],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9398443,"threshold_uncertainty_score":0.3181376,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03598934891247512,"score_gpt":0.3939235914010661,"score_spread":0.357934242488591,"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."}}