{"id":"W4402339813","doi":"10.1080/10872981.2024.2396166","title":"Recommendations to address and research systemic bias in assessment: perspectives from directors of research in medical education","year":2024,"lang":"en","type":"article","venue":"Medical Education Online","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medical education; Medical research; Medicine; Psychology; Engineering ethics; Engineering; Pathology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0102136,0.0001700773,0.0004178676,0.00379251,0.0001041697,0.00005753236,0.0003413332,0.0004072156,0.00290286],"category_scores_gemma":[0.03554601,0.0001548931,0.00004086881,0.006369242,0.0005497946,0.0001811556,0.0001647434,0.002526797,0.00003723227],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001426619,"about_ca_system_score_gemma":0.02369081,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003891804,"about_ca_topic_score_gemma":0.0009545393,"domain_scores_codex":[0.993434,0.001116551,0.00113087,0.0007371844,0.003120464,0.0004609651],"domain_scores_gemma":[0.9953369,0.001899442,0.00007696221,0.0005403671,0.001537938,0.0006083261],"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":[0.00008815471,0.01161921,0.0485125,0.001080802,0.00005133944,0.00002531211,0.01943545,0.000001278099,0.0002882441,0.01786264,0.0807507,0.8202844],"study_design_scores_gemma":[0.002562232,0.0007193987,0.5486953,0.06452395,0.0000650471,0.0002644599,0.2047584,0.0149568,0.0001847617,0.006742239,0.1559073,0.0006200724],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7830446,0.003527549,0.00005585876,0.2090547,0.001699285,0.0009143644,0.00001948728,0.0000461758,0.001637923],"genre_scores_gemma":[0.9894773,0.001548176,0.004040801,0.001434674,0.00127084,0.0005882342,0.000863721,0.00003956258,0.0007366728],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8196643,"threshold_uncertainty_score":0.9997744,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1569873740058067,"score_gpt":0.5822385601706253,"score_spread":0.4252511861648186,"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."}}