{"id":"W4322757277","doi":"10.1111/medu.15057","title":"Does allowing access to electronic differential diagnosis support threaten the reliability of a licensing exam?","year":2023,"lang":"en","type":"article","venue":"Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Medical Council of Canada; University of Ottawa; McMaster University","funders":"Medical Council of Canada","keywords":"Cronbach's alpha; Test (biology); Internal consistency; Reliability (semiconductor); Medicine; Family medicine; Psychology; Medical education; Clinical psychology; Psychometrics","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"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":[0.01818617,0.0002781026,0.0003420049,0.0009283397,0.0006813224,0.00239169,0.001009836,0.0008892785,0.004665782],"category_scores_gemma":[0.1572825,0.0004299632,0.0005704775,0.0009526014,0.001906789,0.001678326,0.001853264,0.0009065787,0.0006707592],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001958578,"about_ca_system_score_gemma":0.002965841,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01157567,"about_ca_topic_score_gemma":0.01660703,"domain_scores_codex":[0.9837559,0.00608349,0.001906525,0.001065027,0.005824726,0.00136432],"domain_scores_gemma":[0.8147224,0.1075921,0.05168221,0.007474593,0.01249929,0.006029405],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002218441,0.00011008,0.9763957,0.00006922882,0.00002578316,0.00007051699,0.001409122,0.00002967253,0.0003259553,0.00004090258,0.000261695,0.02103959],"study_design_scores_gemma":[0.00001672942,0.0006939935,0.9948823,0.0001203265,0.00004230221,0.0003467056,0.001995812,0.0002007334,0.0006819571,0.00006200057,0.0009436136,0.00001351823],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9972872,0.0002309019,0.0002532667,0.0008997342,0.0000188234,0.00001940803,0.00004933281,0.00001453689,0.001226911],"genre_scores_gemma":[0.999302,0.00009790572,0.0002546565,0.0001405119,0.00001758157,0.000006493555,0.00002750448,0.000003039529,0.0001502244],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01818617,"threshold_uncertainty_score":0.09617877,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01611369693475713,"score_gpt":0.3713945278280026,"score_spread":0.3552808308932455,"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."}}