{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001104961,0.0001935357,0.000724368,0.00004975241,0.00008804814,0.00001534658,0.0001234141,0.0001475775,0.0003674046],"category_scores_gemma":[0.0003865106,0.0001182452,0.0001729755,0.00036039,0.0005978827,0.00009089052,0.0001070525,0.0004471535,0.000001899528],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001547038,"about_ca_system_score_gemma":0.0003498365,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008608196,"about_ca_topic_score_gemma":0.000005038953,"domain_scores_codex":[0.9978747,0.0001704873,0.0007145975,0.0004213269,0.0005551694,0.0002637252],"domain_scores_gemma":[0.9980678,0.0008716087,0.0001722094,0.0004769187,0.0002226119,0.0001888277],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002225724,0.005647814,0.8793858,0.000879581,0.00026618,0.0001465583,0.0008425726,0.004433693,0.006405177,0.09764232,0.00002625281,0.002098306],"study_design_scores_gemma":[0.07786515,0.002231696,0.8905645,0.0006704366,0.000311427,0.00006938742,0.0004310683,0.002125324,0.01413686,0.008586005,0.002637492,0.0003706806],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9708276,0.0001060499,0.0002415273,0.001102914,0.00008836439,0.0007923071,0.00001635762,0.00005227378,0.02677258],"genre_scores_gemma":[0.9939966,0.00007736028,0.005645835,0.0001235665,0.00004173377,0.00003160251,0.0000159385,0.00001352102,0.00005388971],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08905631,"threshold_uncertainty_score":0.4821897,"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."}}