{"id":"W1969509691","doi":"10.1007/s10459-008-9104-8","title":"The reliability of workplace-based assessment in postgraduate medical education and training: a national evaluation in general practice in the United Kingdom","year":2008,"lang":"en","type":"article","venue":"Advances in Health Sciences Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":76,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University","funders":"Royal College of General Practitioners","keywords":"Reliability (semiconductor); Audit; Referral; Medical education; Medicine; Descriptive statistics; Psychology; Nursing; Statistics","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02100407,0.0001066713,0.0001881626,0.0008427367,0.0002268453,0.0000185759,0.0002255848,0.00007897484,0.00001174251],"category_scores_gemma":[0.02355443,0.00007458097,0.00001504753,0.004394761,0.0006505231,0.0005086413,0.00001407634,0.0005480103,3.970032e-7],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001091784,"about_ca_system_score_gemma":0.02688253,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002615629,"about_ca_topic_score_gemma":0.001435371,"domain_scores_codex":[0.9954169,0.001145606,0.001037811,0.0003496687,0.001768438,0.0002815373],"domain_scores_gemma":[0.9967794,0.001772396,0.0004794659,0.0002205949,0.0006932991,0.00005488117],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001547205,0.003671581,0.6557892,0.0002824666,0.000002020563,0.000001018047,0.01749705,0.009129935,0.00000633995,0.02675917,0.0006399543,0.2860665],"study_design_scores_gemma":[0.001126193,0.0002367768,0.8572255,0.0008713631,0.000004731416,0.0000517149,0.01423267,0.1101908,0.000003676348,0.007229386,0.008727306,0.00009986162],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8749157,0.001337396,0.00008155547,0.1207845,0.0006801107,0.001223452,4.480518e-7,0.000005090877,0.0009717977],"genre_scores_gemma":[0.9781699,0.001203785,0.0122735,0.007618096,0.0001080547,0.0005486062,0.000063908,0.000004794017,0.00000939002],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2859667,"threshold_uncertainty_score":0.9846706,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07716651093129817,"score_gpt":0.5095382937674905,"score_spread":0.4323717828361923,"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."}}