{"id":"W4395454380","doi":"10.1097/sla.0000000000006308","title":"Evaluating the Quality of Narrative Feedback for Entrustable Professional Activities in a Surgery Residency Program","year":2024,"lang":"en","type":"article","venue":"Annals of Surgery","topic":"Innovations in Medical Education","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Medicine; Quality (philosophy); Narrative; Medical education; Residency training; MEDLINE; Medical physics; Linguistics","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":[0.02828768,0.000333006,0.0006172292,0.002155487,0.0004775968,0.001610131,0.0009146698,0.0005548666,0.0009971856],"category_scores_gemma":[0.1788936,0.0003377712,0.0006563556,0.001108862,0.0004990232,0.001220398,0.001809054,0.0006963382,0.0002633227],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001824336,"about_ca_system_score_gemma":0.003026282,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002979862,"about_ca_topic_score_gemma":0.005074774,"domain_scores_codex":[0.9808931,0.007752579,0.002961971,0.0008548176,0.006781429,0.0007559666],"domain_scores_gemma":[0.8233784,0.05240319,0.07622297,0.003495482,0.03693363,0.00756634],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.0008925548,0.0005865225,0.932474,0.0004464377,0.0001884839,0.00009940418,0.005015261,0.0003785685,0.0008468829,0.00002592772,0.0005401443,0.05850576],"study_design_scores_gemma":[0.0000357479,0.003583102,0.9884926,0.000336177,0.00008398103,0.0002189313,0.0031702,0.001238573,0.00151846,0.00003251784,0.00124592,0.00004387873],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.997229,0.0004294371,0.0008642313,0.0001919362,0.00002925343,0.0002273394,0.0002390134,0.00003194861,0.0007579015],"genre_scores_gemma":[0.9966089,0.0002424152,0.002422551,0.00005727738,0.00001674067,0.0001413704,0.000214389,0.000007423785,0.0002888875],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9717123,"threshold_uncertainty_score":0.1496013,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4991166000162388,"score_gpt":0.5674987551780075,"score_spread":0.06838215516176871,"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."}}