{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001394316,0.00006322053,0.0002769212,0.00006852696,0.00002443882,0.00001196398,0.00004705737,0.00006778774,0.0002248681],"category_scores_gemma":[0.0001582299,0.00003524907,0.0001278631,0.0001123722,0.00009202366,0.0001026055,0.00001542693,0.0002260989,0.000001235775],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005633126,"about_ca_system_score_gemma":0.0002276286,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000002940108,"about_ca_topic_score_gemma":3.311734e-7,"domain_scores_codex":[0.9987906,0.0001045796,0.0006655404,0.00008621284,0.0002680286,0.00008501832],"domain_scores_gemma":[0.9985407,0.0004622357,0.0004030595,0.000122459,0.0003895065,0.00008206486],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001979928,0.001976051,0.7000508,0.0003055013,0.0001520018,0.00001133851,0.001763611,0.0001022217,0.01500898,0.003765281,0.0002762052,0.2746081],"study_design_scores_gemma":[0.003697147,0.0008557948,0.9676872,0.0006337131,0.0001035216,0.0002335258,0.001625933,0.0004299184,0.001134506,0.001084953,0.02242572,0.0000880301],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9934329,0.00009885077,0.00007093837,0.001742155,0.0003428489,0.000141903,5.061319e-7,0.000003355122,0.004166546],"genre_scores_gemma":[0.998912,0.00009249912,0.0006293547,0.00003899034,0.00008986465,0.000001547747,0.000002766325,0.000005150208,0.000227843],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.27452,"threshold_uncertainty_score":0.2462148,"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."}}