{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01951566,0.0004039482,0.001182412,0.001743083,0.001035893,0.00163394,0.001207415,0.001619031,0.0009570341],"category_scores_gemma":[0.07704753,0.0005692187,0.001171396,0.001425012,0.001175991,0.001332102,0.00271262,0.001155648,0.0004384554],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002131086,"about_ca_system_score_gemma":0.001620459,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01321959,"about_ca_topic_score_gemma":0.0208316,"domain_scores_codex":[0.9725348,0.0136928,0.00322731,0.001684401,0.008010976,0.0008496734],"domain_scores_gemma":[0.9136332,0.04682091,0.01163003,0.003924759,0.02023708,0.003754047],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001971726,0.0008233571,0.9526213,0.0003595679,0.0002616828,0.00009001414,0.004611351,0.0003556182,0.0007188894,0.00004684024,0.0006045843,0.03753508],"study_design_scores_gemma":[0.0001072859,0.002242447,0.9944714,0.00007741649,0.00006882341,0.00009729641,0.001637099,0.0007703073,0.0001479853,0.00002245843,0.0003309374,0.00002647676],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9978854,0.0004789219,0.0003777218,0.0001755725,0.00005325517,0.00009338697,0.0001006712,0.00001691051,0.0008180243],"genre_scores_gemma":[0.9982534,0.0002202249,0.0009039571,0.00008036963,0.00002431541,0.00007835402,0.0001495098,0.000007961558,0.0002818847],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9804844,"threshold_uncertainty_score":0.1032099,"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."}}