{"id":"W4200505601","doi":"10.1001/jamanetworkopen.2021.37179","title":"Longitudinal Reliability of Milestones-Based Learning Trajectories in Family Medicine Residents","year":2021,"lang":"en","type":"article","venue":"JAMA Network Open","topic":"Innovations in Medical Education","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Formative assessment; Graduate medical education; Developmental Milestone; Longitudinal study; Accreditation; Medical education; Psychology; Learning curve; Consistency (knowledge bases); Medicine; Family medicine; Developmental psychology; Mathematics education; Artificial intelligence; Computer science","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":[],"consensus_categories":[],"category_scores_codex":[0.002529294,0.0001298902,0.000492033,0.0001024405,0.00008454938,0.00002330452,0.000235243,0.000150197,0.0004933565],"category_scores_gemma":[0.005600864,0.0001150374,0.00004138847,0.001699713,0.0002067435,0.000149179,0.0001156239,0.000594485,0.000008586404],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001793969,"about_ca_system_score_gemma":0.0006468376,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009984424,"about_ca_topic_score_gemma":0.0001041679,"domain_scores_codex":[0.9979447,0.0002583315,0.0006992898,0.0003499242,0.0004852223,0.0002625778],"domain_scores_gemma":[0.9983951,0.0002766276,0.000199079,0.0004637458,0.000604224,0.00006123108],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004031409,0.0002145894,0.9686567,0.0001684497,0.00001983724,0.00005863611,0.0004085757,0.002099531,0.0003096896,0.0003734753,0.02036099,0.006926449],"study_design_scores_gemma":[0.0023103,0.0002045166,0.987313,0.001623023,0.00004571601,0.00001144803,0.000648428,0.001135451,0.0002843309,0.0004815042,0.005845443,0.00009680245],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9825535,0.0005178585,0.0004320215,0.01015677,0.0006035667,0.0004403136,8.847986e-7,0.00002171776,0.005273319],"genre_scores_gemma":[0.987658,0.00005901921,0.009299811,0.001306666,0.0006163493,0.00004981237,0.0001596301,0.00001950987,0.0008312651],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0186564,"threshold_uncertainty_score":0.6705162,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0447865523951364,"score_gpt":0.3701014806087468,"score_spread":0.3253149282136104,"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."}}