{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05612329,0.0008621447,0.001100744,0.00477375,0.001505785,0.002811085,0.001129679,0.001096017,0.002464308],"category_scores_gemma":[0.1289067,0.0006702537,0.0007148648,0.00170793,0.001170793,0.001893731,0.003243109,0.001152378,0.002252644],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002568638,"about_ca_system_score_gemma":0.005790373,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001004859,"about_ca_topic_score_gemma":0.002405322,"domain_scores_codex":[0.9358565,0.04235395,0.005829122,0.002013615,0.01285764,0.001089112],"domain_scores_gemma":[0.8389039,0.07125461,0.01740581,0.01594065,0.05324881,0.003246182],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0009069587,0.001800238,0.1481064,0.0006586835,0.000060868,0.0003156097,0.00382024,0.004722317,0.01265528,0.00195103,0.007793943,0.8172085],"study_design_scores_gemma":[0.000843938,0.01202893,0.7282439,0.001813626,0.0001527448,0.001444618,0.01290314,0.05877529,0.06426036,0.005487084,0.1133782,0.0006681712],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6574275,0.0008021924,0.299353,0.002304961,0.0006129174,0.01979578,0.001086672,0.002259456,0.01635752],"genre_scores_gemma":[0.4346892,0.0004220991,0.5416422,0.0007147961,0.0002103203,0.0151796,0.001334281,0.0003511537,0.005456358],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9438767,"threshold_uncertainty_score":0.2968119,"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."}}