{"id":"W3129454103","doi":"10.4300/jgme-d-20-00619.1","title":"Comparing 2 Approaches for the File Review of Residency Applications","year":2021,"lang":"en","type":"article","venue":"Journal of Graduate Medical Education","topic":"Diversity and Career in Medicine","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Inter-rater reliability; Intraclass correlation; Reliability (semiconductor); Computer science; Trait; Psychology; Medicine; Statistics; Rating scale; Clinical psychology; Mathematics; Psychometrics","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":[0.1604317,0.002031354,0.00191289,0.009184408,0.002284965,0.004604598,0.00340266,0.001689037,0.004519526],"category_scores_gemma":[0.3667007,0.001594821,0.003351685,0.004985966,0.002166228,0.004817231,0.006304941,0.002043525,0.001614489],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00512978,"about_ca_system_score_gemma":0.008700689,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004772024,"about_ca_topic_score_gemma":0.0134748,"domain_scores_codex":[0.6566128,0.2448352,0.02855937,0.0161185,0.05123167,0.002642533],"domain_scores_gemma":[0.4650603,0.3631759,0.05921608,0.03278907,0.07444684,0.005311967],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01536522,0.003144724,0.09084954,0.008150059,0.003280863,0.000278928,0.02298931,0.005085581,0.006989848,0.002450465,0.007402256,0.8340132],"study_design_scores_gemma":[0.01121032,0.05862318,0.711455,0.006976861,0.006860773,0.003880594,0.02778237,0.07653786,0.02494514,0.01122283,0.05785585,0.002649136],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6549224,0.005954986,0.2670126,0.001944807,0.00121239,0.04704967,0.001863443,0.002271489,0.01776823],"genre_scores_gemma":[0.5150215,0.002397303,0.4352386,0.0007646041,0.0004608864,0.0422583,0.0009939541,0.0004004904,0.002464388],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8395683,"threshold_uncertainty_score":0.8484542,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2185515798955294,"score_gpt":0.3966764882167559,"score_spread":0.1781249083212265,"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."}}