{"id":"W2006326318","doi":"10.4300/jgme-d-11-00123.1","title":"Prospective Comparison of Live Evaluation and Video Review in the Evaluation of Operator Performance in a Pediatric Emergency Airway Simulation","year":2012,"lang":"en","type":"article","venue":"Journal of Graduate Medical Education","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":33,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa Skills and Simulation Centre","funders":"","keywords":"Inter-rater reliability; Concordance; Medicine; Intra-rater reliability; Intubation; Reliability (semiconductor); Confidence interval; Medical physics; Rating scale; Psychology; 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":[0.01261877,0.0001121412,0.000408067,0.0004215715,0.00003635018,0.000003449299,0.00009850618,0.00009611751,0.00030069],"category_scores_gemma":[0.00769105,0.00008106932,0.00005803757,0.0009061378,0.00004604799,0.0003472574,0.000009160367,0.0004021075,0.000002371668],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004224652,"about_ca_system_score_gemma":0.003595014,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005752996,"about_ca_topic_score_gemma":0.0000158127,"domain_scores_codex":[0.9947555,0.000926736,0.001613403,0.0001221215,0.002432971,0.0001492676],"domain_scores_gemma":[0.9942582,0.000424037,0.001155569,0.0001829427,0.003827382,0.0001518781],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007205026,0.001039542,0.9504039,0.00130254,0.000009983661,5.495882e-8,0.01502679,0.001376287,0.000006580957,0.0000209509,0.0002879717,0.03045339],"study_design_scores_gemma":[0.001118781,0.0002847735,0.9399748,0.001691807,0.000395659,0.00001237038,0.003527048,0.0527541,0.00004827338,0.0001195328,0.00001252704,0.00006027607],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9644839,0.02897333,0.00002458245,0.004126939,0.0007556138,0.001559194,6.752225e-7,0.000001923981,0.00007380998],"genre_scores_gemma":[0.9944689,0.004578583,0.0001418924,0.0001867648,0.0005000324,0.00009203173,0.00002007518,0.000009137585,0.000002555291],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05137782,"threshold_uncertainty_score":0.9207461,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1591514316052036,"score_gpt":0.5048788781775292,"score_spread":0.3457274465723256,"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."}}