{"id":"W2405662287","doi":"10.19144/1911-1606.11.1.7","title":"The General Internal Medicine Certification Examination: Does it Measure What it Should?","year":2016,"lang":"en","type":"article","venue":"Canadian Journal of General Internal Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medicine; Certification; Measure (data warehouse); Management; Data mining; Computer science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00357522,0.0003608138,0.0006299792,0.0008833811,0.000346086,0.0001007971,0.0007936556,0.0002216845,0.003138711],"category_scores_gemma":[0.006026592,0.000159432,0.0001504664,0.0005842164,0.00135689,0.000627486,0.00002926732,0.000824196,0.00004409682],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001228725,"about_ca_system_score_gemma":0.00122193,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003619272,"about_ca_topic_score_gemma":0.007888796,"domain_scores_codex":[0.995569,0.0002640652,0.001819669,0.0003504471,0.001385348,0.0006114694],"domain_scores_gemma":[0.9945041,0.0003124757,0.0009047795,0.0005857123,0.002714023,0.0009789303],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001993516,0.00003538763,0.00301999,0.00003667201,0.0002731703,0.0002196884,0.00250352,0.000003250522,0.01524474,0.002101904,0.5686013,0.407761],"study_design_scores_gemma":[0.005011265,0.001951394,0.03201651,0.009223014,0.0004629924,0.003686246,0.007265875,0.0005361309,0.004008605,0.001809687,0.9336263,0.0004020013],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.1950328,0.003666569,0.008319498,0.7675794,0.01640167,0.0003983451,0.000006088095,0.00002079131,0.008574812],"genre_scores_gemma":[0.9142296,0.001721398,0.0007732963,0.02144369,0.01573381,0.00002307674,0.00002032731,0.00006234895,0.04599244],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7461357,"threshold_uncertainty_score":0.9977726,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05931714230951838,"score_gpt":0.3431535519842393,"score_spread":0.2838364096747209,"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."}}