{"id":"W2136442190","doi":"10.1080/01421590500136113","title":"Creating a reliable and valid blueprint for the internal medicine clerkship evaluation","year":2005,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Patient Satisfaction in Healthcare","field":"Health Professions","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Blueprint; Clinical clerkship; Medicine; Objective structured clinical examination; Cohen's kappa; Presentation (obstetrics); Medical education; Medical physics; Family medicine; Psychology; Curriculum; Statistics; Surgery; Mathematics; Pedagogy","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.006858321,0.0001105342,0.0001922154,0.00004671441,0.0007108602,0.000004373851,0.0001509649,0.0002267793,0.0122164],"category_scores_gemma":[0.008386382,0.00006611327,0.00003155425,0.00008244523,0.0001366506,0.00007086959,0.00008603459,0.0009106963,0.00009027795],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002838858,"about_ca_system_score_gemma":0.0002484173,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001320902,"about_ca_topic_score_gemma":0.0003797818,"domain_scores_codex":[0.9973975,0.0005442231,0.0005981867,0.0002481343,0.0008618464,0.0003500982],"domain_scores_gemma":[0.9976245,0.00146763,0.0001942166,0.0002502587,0.0002158445,0.0002475374],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006442905,0.00001875481,0.8243728,0.0001386915,0.00002968537,3.732325e-7,0.009912625,0.00002054082,0.000006111725,0.001302544,0.03420362,0.1299298],"study_design_scores_gemma":[0.004628919,0.0002184702,0.3963178,0.001613441,0.0001324337,0.000008200148,0.008966975,0.385925,0.000008994267,0.002959423,0.1990118,0.0002086096],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8372411,0.001676017,0.00893219,0.1171844,0.001900294,0.003066633,0.000003525293,0.0001608109,0.02983505],"genre_scores_gemma":[0.9886279,0.000101036,0.0005511505,0.003310576,0.002372039,0.0009226588,0.00001082743,0.00002200361,0.004081808],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4280551,"threshold_uncertainty_score":0.9999664,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2377277645213538,"score_gpt":0.5272230170736342,"score_spread":0.2894952525522804,"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."}}