{"id":"W3159472089","doi":"10.4300/jgme-d-20-01039.1","title":"A Validity Framework for Effective Analysis and Interpretation of Milestones Data","year":2021,"lang":"en","type":"article","venue":"Journal of Graduate Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":36,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Milestone; Graduate medical education; Accreditation; Competence (human resources); Medical education; Specialty; Subspecialty; Psychology; Population; Medicine; Family medicine","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":[0.002065449,0.00009108239,0.0004505763,0.0004053814,0.00004567122,0.00001437154,0.0001534434,0.0001429522,0.00009202765],"category_scores_gemma":[0.04332216,0.00007479516,0.0001188895,0.001134361,0.0001463327,0.0001710495,0.00004532469,0.0003918318,6.919154e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008489702,"about_ca_system_score_gemma":0.002037993,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001400636,"about_ca_topic_score_gemma":0.000006072323,"domain_scores_codex":[0.9980905,0.0001441443,0.0007349339,0.0001874171,0.0007395856,0.0001033916],"domain_scores_gemma":[0.9961613,0.0006667682,0.0006250171,0.0003719137,0.002037212,0.0001378283],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0003715948,0.001849017,0.02560754,0.0007608466,0.001057525,0.000007866446,0.002385363,0.000004406353,0.0003465075,0.002276039,0.01418599,0.9511473],"study_design_scores_gemma":[0.004337133,0.002144126,0.804644,0.006202904,0.01608642,0.001524364,0.01123636,0.07622507,0.008884242,0.05867848,0.009513765,0.0005231342],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6959641,0.0004110131,0.2838908,0.01834901,0.001140301,0.0002050631,0.000006571342,0.000004432272,0.00002870387],"genre_scores_gemma":[0.9068565,0.0002789668,0.09068464,0.001305352,0.0005709021,0.00001569769,0.0002487089,0.00000945532,0.00002981464],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9506242,"threshold_uncertainty_score":0.9647363,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05969414396954906,"score_gpt":0.438137147614899,"score_spread":0.3784430036453499,"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."}}