{"id":"W4392260739","doi":"10.1016/j.sopen.2024.02.008","title":"“Doing well”: Intraoperative entrustable professional activity assessments provided limited technical feedback","year":2024,"lang":"en","type":"article","venue":"Surgery Open Science","topic":"Innovations in Medical Education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Formative assessment; Subspecialty; Narrative; Narrative review; Medical education; Inclusion (mineral); Medicine; Psychology; Pedagogy; Family medicine; Social psychology","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00443851,0.0001505292,0.0002907478,0.000443999,0.0004754478,0.0005399747,0.0005413979,0.00009359396,0.0009849243],"category_scores_gemma":[0.001838794,0.0001147514,0.00004889424,0.003932457,0.0005461006,0.001771775,0.0005041845,0.0006035374,0.0001948211],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005922674,"about_ca_system_score_gemma":0.006066242,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004910028,"about_ca_topic_score_gemma":0.000003722897,"domain_scores_codex":[0.9973312,0.00008877574,0.0003684164,0.0006786123,0.001072313,0.000460679],"domain_scores_gemma":[0.9985404,0.0003667108,0.00007989125,0.0004289219,0.00043426,0.0001498096],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0003276644,0.001663319,0.01697714,0.0004678597,0.00006847316,0.0001455628,0.001010729,0.000009617319,0.2366669,0.01242618,0.1997749,0.5304617],"study_design_scores_gemma":[0.001487383,0.00052394,0.2633763,0.007748931,0.0001876093,0.0006252569,0.00417124,0.03743703,0.2137285,0.001189822,0.4679931,0.001530964],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9451276,0.00004265441,0.002284325,0.01358614,0.003837989,0.001518392,0.000004494932,0.0001738941,0.03342453],"genre_scores_gemma":[0.9800957,0.00001017123,0.01269864,0.001176153,0.0001602966,0.0001846605,0.00002552483,0.00001665668,0.005632224],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5289307,"threshold_uncertainty_score":0.9999283,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04051796398346803,"score_gpt":0.4240008851567331,"score_spread":0.383482921173265,"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."}}