{"id":"W4410888806","doi":"10.1016/j.jsurg.2025.103553","title":"Variability in Testing Scenario Complexity Reduces the Validity of the Advanced Trauma Life Support Certification Assessment","year":2025,"lang":"en","type":"article","venue":"Journal of surgical education","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Certification; Advanced trauma life support; Psychology; Computer science; Medicine; Reliability engineering; Medical emergency; Engineering; Political science","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":[0.0673308,0.0006501225,0.001003368,0.001853911,0.001011409,0.004058096,0.002074703,0.002096262,0.002721327],"category_scores_gemma":[0.5409568,0.000642759,0.001625221,0.001400783,0.001616561,0.00393468,0.003090002,0.002791786,0.0007971388],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002001317,"about_ca_system_score_gemma":0.002767937,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005963957,"about_ca_topic_score_gemma":0.005476485,"domain_scores_codex":[0.8887573,0.05992258,0.009453421,0.01425542,0.02589782,0.001713367],"domain_scores_gemma":[0.2379464,0.6763499,0.02886082,0.03784097,0.01686602,0.002135926],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002899941,0.0008873974,0.8720137,0.0002401817,0.0008599746,0.0002218806,0.002738782,0.01785572,0.004077373,0.001487975,0.00237728,0.09433974],"study_design_scores_gemma":[0.0002145224,0.001945163,0.8233777,0.0004959147,0.0005134492,0.001432301,0.001683321,0.1471055,0.007756923,0.01026672,0.00498454,0.0002238211],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9281921,0.0004661671,0.05897548,0.0009249552,0.0003606802,0.0004320227,0.0007665289,0.0005537017,0.009328234],"genre_scores_gemma":[0.991284,0.00004396187,0.007539704,0.0002344983,0.00003986233,0.0001028755,0.0003761755,0.0001065204,0.000272448],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9326692,"threshold_uncertainty_score":0.3560836,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1450224810570286,"score_gpt":0.4404700779937134,"score_spread":0.2954475969366848,"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."}}