{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01995914,0.000336444,0.0008121325,0.002906178,0.000803686,0.00230455,0.001205606,0.001977813,0.002366043],"category_scores_gemma":[0.1797314,0.000216644,0.0005761558,0.0030924,0.002534489,0.002689434,0.001701757,0.001596561,0.0007904423],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002282846,"about_ca_system_score_gemma":0.004624465,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01026213,"about_ca_topic_score_gemma":0.02039337,"domain_scores_codex":[0.9813997,0.006714817,0.001495717,0.0009061296,0.008835667,0.0006479158],"domain_scores_gemma":[0.897864,0.05054735,0.02375782,0.00312953,0.02045376,0.004247523],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001797975,0.0002678709,0.8679858,0.0007338437,0.0002086821,0.0000559628,0.001406814,0.0001805907,0.0002211945,0.00252499,0.01401189,0.1122226],"study_design_scores_gemma":[0.00005631479,0.0003842478,0.9739665,0.001817164,0.0002292741,0.0003963327,0.002145885,0.0005791098,0.0002980957,0.002675175,0.01740338,0.00004859752],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6920927,0.04178804,0.007819368,0.1405109,0.004522386,0.0005810857,0.002442085,0.0001743495,0.1100691],"genre_scores_gemma":[0.9803807,0.00597902,0.004373468,0.005603009,0.0009167357,0.0001381592,0.0007295355,0.0000425446,0.001836707],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9977171,"threshold_uncertainty_score":0.1055553,"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."}}