{"id":"W2912065376","doi":"10.1097/acm.0000000000002234","title":"National Internal Medicine Milestone Ratings: Validity Evidence From Longitudinal Three-Year Follow-up","year":2018,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":31,"is_retracted":false,"has_abstract":true,"ca_institutions":"123 Certification (Canada)","funders":"","keywords":"Milestone; MEDLINE; Medicine; Medical education; Psychology; Family medicine; Political science; Geography; Law","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":[],"consensus_categories":[],"category_scores_codex":[0.02016418,0.0002653712,0.0004875296,0.001724216,0.0006703413,0.0009816192,0.001105666,0.0005126555,0.00112947],"category_scores_gemma":[0.1004217,0.0002703515,0.00080834,0.00188019,0.0006682851,0.001254843,0.001701337,0.0006416454,0.0003593527],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009974769,"about_ca_system_score_gemma":0.001265468,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009111841,"about_ca_topic_score_gemma":0.01209492,"domain_scores_codex":[0.98905,0.00426363,0.001481311,0.001548932,0.003259955,0.0003961491],"domain_scores_gemma":[0.9000388,0.02542181,0.04167201,0.007855893,0.02275666,0.002254858],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000074335,0.00002436743,0.9947228,0.00005200522,0.00013544,0.00000645366,0.0001836497,0.00007872386,0.00002676831,0.0000318719,0.0002489575,0.004414617],"study_design_scores_gemma":[0.000004833151,0.0001098091,0.9986193,0.00008050482,0.00004807813,0.0000289235,0.0001555238,0.0001927196,0.00009453978,0.00002176116,0.0006381685,0.000005815255],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9899311,0.002478928,0.001506221,0.0003389331,0.00008518712,0.0001054166,0.002281379,0.00003077028,0.003242025],"genre_scores_gemma":[0.995461,0.0004497992,0.000834167,0.00009220532,0.00003699392,0.0001153795,0.002729281,0.00001072582,0.0002702827],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02016418,"threshold_uncertainty_score":0.1066396,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1412742352831648,"score_gpt":0.4221228193773531,"score_spread":0.2808485840941883,"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."}}