{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.6320429,0.003615898,0.004611841,0.02477069,0.009518295,0.01631244,0.008438738,0.005863637,0.003993854],"category_scores_gemma":[0.743597,0.002594211,0.004893477,0.01415956,0.03519268,0.01841804,0.0190291,0.009913894,0.001575588],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01879448,"about_ca_system_score_gemma":0.05433825,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01646731,"about_ca_topic_score_gemma":0.007000331,"domain_scores_codex":[0.2589111,0.6045171,0.0493402,0.01524558,0.06810085,0.003885177],"domain_scores_gemma":[0.1327606,0.6939979,0.02848305,0.05173777,0.09096837,0.00205239],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002988271,0.0002800267,0.02348379,0.002280253,0.0006974904,0.0004590089,0.01864921,0.003892154,0.0006745619,0.8111804,0.007606361,0.1304979],"study_design_scores_gemma":[0.0003269805,0.0005882867,0.01205941,0.008858007,0.0003054682,0.0006705829,0.01058565,0.02928885,0.001840742,0.8734357,0.06171758,0.0003228294],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00870535,0.002334958,0.9219357,0.01604596,0.0006895078,0.01084178,0.000733572,0.0004544959,0.03825868],"genre_scores_gemma":[0.1620514,0.001032914,0.8011559,0.003791523,0.0008934995,0.0287046,0.0008489714,0.0002848066,0.001236433],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6320429,"threshold_uncertainty_score":0.4537566,"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."}}