{"id":"W2911784036","doi":"10.1016/j.jsurg.2019.01.001","title":"The Inter-Rater Reliability of Technical Skills Assessment and Retention of Rater Training","year":2019,"lang":"en","type":"article","venue":"Journal of surgical education","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Ottawa; Ottawa Hospital","funders":"","keywords":"Intraclass correlation; Session (web analytics); Inter-rater reliability; Context (archaeology); Reliability (semiconductor); Laparoscopic cholecystectomy; Medicine; Physical therapy; Psychology; General surgery; Computer science; Rating scale; Psychometrics; Clinical psychology","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":[0.08314121,0.0008359895,0.001609981,0.002275586,0.00149343,0.002067542,0.001186434,0.001452408,0.001590754],"category_scores_gemma":[0.1896547,0.0006855463,0.002779871,0.001633223,0.002131331,0.002434322,0.002237054,0.001601414,0.001365275],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001015764,"about_ca_system_score_gemma":0.001043486,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001914231,"about_ca_topic_score_gemma":0.003176117,"domain_scores_codex":[0.8969544,0.05503219,0.01504793,0.009293224,0.02203227,0.001639992],"domain_scores_gemma":[0.6929781,0.2047427,0.0164836,0.02284284,0.06148703,0.001465702],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.007663637,0.001041968,0.8204229,0.001261124,0.005866849,0.0002647254,0.01982274,0.004452261,0.01626644,0.002800081,0.00380174,0.1163355],"study_design_scores_gemma":[0.0004644583,0.002976727,0.9490347,0.0002885018,0.001564685,0.0008898595,0.002106072,0.02226419,0.01239066,0.001679815,0.006075208,0.0002651479],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9199071,0.002216574,0.06114656,0.0002250835,0.0008645398,0.002027848,0.001212069,0.0003474114,0.01205282],"genre_scores_gemma":[0.974661,0.0003354295,0.02017012,0.0001319449,0.0001759495,0.001407093,0.001078837,0.0002077579,0.001831986],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9168588,"threshold_uncertainty_score":0.439698,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02521866518024988,"score_gpt":0.3640945645040595,"score_spread":0.3388758993238096,"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."}}