{"id":"W4320894722","doi":"10.4300/jgme-d-22-00222.1","title":"Evaluating the Quality of Written Feedback Within Entrustable Professional Activities in an Internal Medicine Cohort","year":2023,"lang":"en","type":"article","venue":"Journal of Graduate Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University; Centre for Addiction and Mental Health; University of Toronto; Queen's University; University of Ottawa","funders":"","keywords":"Task (project management); Test (biology); Quality (philosophy); Orientation (vector space); Inter-rater reliability; Medicine; Medical education; Psychology; Family medicine; Mathematics; Developmental psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.01591138,0.000146652,0.0005282212,0.0006010644,0.00009953646,0.00001338639,0.0003384339,0.0001486612,0.000575166],"category_scores_gemma":[0.02155779,0.0000881925,0.0000786911,0.001347086,0.0004660852,0.0003006872,0.00005069409,0.001203436,0.00000696042],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002955924,"about_ca_system_score_gemma":0.004391233,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000249362,"about_ca_topic_score_gemma":0.00002777045,"domain_scores_codex":[0.9940363,0.0006126635,0.001897595,0.00018509,0.003025048,0.0002433003],"domain_scores_gemma":[0.9965563,0.0005134898,0.001263098,0.0003042999,0.001182498,0.0001803506],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001751304,0.003861728,0.5118421,0.001027326,0.0002728254,0.0000236517,0.03008121,0.0001044315,0.01030392,0.004055926,0.1091833,0.3274922],"study_design_scores_gemma":[0.002419801,0.001158909,0.9429947,0.003412557,0.0001356791,0.0002283095,0.036722,0.007751594,0.0005879754,0.00395492,0.0004935077,0.0001400141],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9377411,0.00009180413,0.00009933221,0.05759091,0.00378584,0.000347632,9.174817e-7,0.00001487287,0.0003275646],"genre_scores_gemma":[0.9938377,0.00009095229,0.001193925,0.002257561,0.001473083,0.00003661247,0.00005883828,0.00001914,0.001032262],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4311526,"threshold_uncertainty_score":0.986684,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1376363597821011,"score_gpt":0.507944071433075,"score_spread":0.3703077116509738,"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."}}