{"id":"W4399626204","doi":"10.4300/jgme-d-23-00569.1","title":"Feedback in an Entrustment-Based Objective Structured Clinical Examination: Analysis of Content and Scoring Methods","year":2024,"lang":"en","type":"article","venue":"Journal of Graduate Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval","funders":"","keywords":"Objective structured clinical examination; Computer science; Content analysis; Medical education; Content (measure theory); MEDLINE; Data science; Medical physics; Medicine; Information retrieval","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":[0.006023908,0.000128397,0.000602402,0.001603917,0.00003001018,0.00003072311,0.000116918,0.000178658,0.0002293948],"category_scores_gemma":[0.007873718,0.00009918804,0.0001647386,0.002100116,0.0002379355,0.0002272782,0.00001614697,0.0007629129,7.498646e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003037385,"about_ca_system_score_gemma":0.002678144,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003121042,"about_ca_topic_score_gemma":0.00001582714,"domain_scores_codex":[0.9966475,0.0005512912,0.001480187,0.0002356311,0.0009403412,0.0001450207],"domain_scores_gemma":[0.9977208,0.0004463228,0.0004524879,0.0001921975,0.0009661571,0.0002220224],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0001276153,0.00107042,0.05514629,0.0002281709,0.0005809651,0.0000169828,0.00177745,0.00001993875,0.0004456894,0.0008601581,0.0005148262,0.9392115],"study_design_scores_gemma":[0.001141537,0.0005196767,0.9547514,0.001011174,0.001569838,0.00007155777,0.003837315,0.03552629,0.0006663316,0.0004351109,0.0003702634,0.00009954397],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9774466,0.0008780918,0.01152519,0.007472969,0.002387366,0.0002142742,0.00000138217,0.00001069038,0.00006348424],"genre_scores_gemma":[0.9648727,0.0001782916,0.03335218,0.0009885825,0.0004967775,0.0000103343,0.00004796072,0.0000130421,0.00004018429],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9391119,"threshold_uncertainty_score":0.9426144,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1203724717112272,"score_gpt":0.4970956975898894,"score_spread":0.3767232258786622,"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."}}