{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02076901,0.0002364226,0.0003262575,0.001532898,0.0004532151,0.0007285741,0.0006515926,0.000397527,0.000729738],"category_scores_gemma":[0.07275411,0.0002754902,0.0005318903,0.001298165,0.000630536,0.0009705112,0.001378941,0.0005512344,0.0002564586],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006597596,"about_ca_system_score_gemma":0.001026897,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008737835,"about_ca_topic_score_gemma":0.008395623,"domain_scores_codex":[0.9929773,0.003139777,0.0009078506,0.0009846458,0.001591279,0.0003991963],"domain_scores_gemma":[0.9551904,0.01681193,0.01253887,0.004677108,0.009911391,0.0008703166],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006034495,0.00001638216,0.9936193,0.00001925662,0.0000713452,0.000006885549,0.000631558,0.0003107399,0.00008142727,0.00004870784,0.0001506448,0.004983468],"study_design_scores_gemma":[0.000003606329,0.0001128922,0.9972112,0.00003176941,0.00002232399,0.0000378604,0.0003638942,0.001518353,0.0002407813,0.00008921116,0.0003579869,0.00001017775],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9961833,0.0001896612,0.002160776,0.00009597895,0.00001372132,0.00003632102,0.0006782869,0.00002974392,0.0006121033],"genre_scores_gemma":[0.9983867,0.00004338578,0.0007215479,0.000009725584,0.000004656043,0.0000364808,0.0006748828,0.00000795538,0.0001145596],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.979231,"threshold_uncertainty_score":0.1098384,"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."}}