{"id":"W2067223870","doi":"10.1007/bf03019750","title":"Validity and reliability of undergraduate performance assessments in an anesthesia simulator","year":2001,"lang":"en","type":"article","venue":"Canadian Journal of Anesthesia/Journal canadien d anesthésie","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":60,"is_retracted":false,"has_abstract":false,"ca_institutions":"Health Sciences Centre; University of Toronto; Sunnybrook Health Science Centre","funders":"University of Toronto; Canadian Anesthesiologists' Society","keywords":"Checklist; Reliability (semiconductor); Repeatability; Simulation; Psychology; Protocol (science); Consistency (knowledge bases); Computer science; Medical physics; Medicine; Statistics; Mathematics; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":true,"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":[],"consensus_categories":[],"category_scores_codex":[0.01101124,0.0003932455,0.0004627311,0.001316339,0.0005903733,0.001046275,0.0006890043,0.0007806631,0.0006098339],"category_scores_gemma":[0.06196672,0.0004067293,0.0007396286,0.0007130163,0.0009130908,0.0009155989,0.0009942334,0.0006895718,0.0004926757],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001458723,"about_ca_system_score_gemma":0.001593664,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005226902,"about_ca_topic_score_gemma":0.008550768,"domain_scores_codex":[0.9906171,0.003133974,0.001044467,0.0008742501,0.003918757,0.0004114897],"domain_scores_gemma":[0.9434358,0.02667005,0.006667455,0.003216131,0.01735844,0.00265218],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0007437101,0.0007753218,0.9695255,0.00005530303,0.0002293115,0.00003658673,0.0020079,0.0009003296,0.002609462,0.0001757238,0.0004584855,0.02248231],"study_design_scores_gemma":[0.00005549813,0.001320801,0.9900098,0.00004565992,0.0000470305,0.0001102734,0.0007294737,0.005359051,0.00135456,0.000144962,0.0007985434,0.00002437784],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9976978,0.00009406286,0.0007226015,0.00004818875,0.0000561577,0.00006633333,0.00005250722,0.00001895833,0.001243417],"genre_scores_gemma":[0.9985606,0.00006456629,0.0008393183,0.00002635365,0.00001563772,0.00005055972,0.0001242212,0.000007958344,0.0003109298],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01101124,"threshold_uncertainty_score":0.05823374,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03342188791124928,"score_gpt":0.3141108664119323,"score_spread":0.280688978500683,"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."}}