{"id":"W4294351348","doi":"10.36834/cmej.72369","title":"Assessment of laparoscopic skills: comparing the reliability of global rating and entrustability tools","year":2022,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Network for Innovation in Education; Memorial University of Newfoundland; University of Ottawa","funders":"","keywords":"Reliability (semiconductor); Computer science; Reliability engineering; Engineering; Physics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02514024,0.0004421101,0.0005375888,0.001835991,0.0002751393,0.0009383028,0.0005818076,0.0005007813,0.0007972581],"category_scores_gemma":[0.0655451,0.0002958809,0.0007941353,0.0008284583,0.001136382,0.001145698,0.001560588,0.0005278763,0.0003015338],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004576035,"about_ca_system_score_gemma":0.0004677777,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001105637,"about_ca_topic_score_gemma":0.001368876,"domain_scores_codex":[0.9798615,0.01014494,0.001953396,0.0017846,0.005733365,0.000522305],"domain_scores_gemma":[0.9283593,0.04682771,0.008172062,0.003545683,0.01211651,0.0009788432],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001783961,0.0002405189,0.903331,0.0004530861,0.001335867,0.00008520748,0.004886076,0.001025277,0.003337464,0.0003783577,0.0007792918,0.08236391],"study_design_scores_gemma":[0.0001318773,0.002240038,0.98355,0.0002272222,0.0003532501,0.000372915,0.002503216,0.005870991,0.002454173,0.000530827,0.001700634,0.00006486147],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9887628,0.0009752517,0.006728342,0.00009041832,0.00009833067,0.0001915213,0.000170233,0.00004039934,0.00294273],"genre_scores_gemma":[0.9970933,0.000159403,0.002279801,0.00002632843,0.00002039889,0.00008357221,0.0001501728,0.00001545981,0.0001713925],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02514024,"threshold_uncertainty_score":0.1329559,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02064250361716217,"score_gpt":0.349355624035823,"score_spread":0.3287131204186609,"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."}}