{"id":"W2066392492","doi":"10.1111/medu.12490","title":"In‐training evaluations: developing an automated screening tool to measure report quality","year":2014,"lang":"en","type":"article","venue":"Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Measure (data warehouse); Quality (philosophy); Medical education; Computer science; Medical physics; Training (meteorology); MEDLINE; Medicine; Psychology; Data mining","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":[0.03549797,0.0009446541,0.001055397,0.00585855,0.0003850858,0.002326825,0.001830654,0.000732317,0.002671772],"category_scores_gemma":[0.1111914,0.0006246378,0.001185269,0.002894241,0.0003813052,0.001979312,0.001898702,0.001203035,0.001800201],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002048551,"about_ca_system_score_gemma":0.00333683,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006389095,"about_ca_topic_score_gemma":0.007262726,"domain_scores_codex":[0.9748805,0.01222761,0.003717263,0.002493174,0.006142134,0.0005393089],"domain_scores_gemma":[0.8566111,0.07757387,0.02666617,0.007156583,0.03023904,0.001753318],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0009367243,0.001105225,0.4403277,0.001092864,0.0004957888,0.000117864,0.001010687,0.008324446,0.001904049,0.001102413,0.02615773,0.5174246],"study_design_scores_gemma":[0.0003870145,0.001583282,0.6692547,0.0007399789,0.0003342342,0.0003828974,0.001002513,0.294882,0.01001821,0.002821715,0.01838444,0.0002090574],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5873595,0.001092328,0.3492661,0.001999439,0.000308813,0.00976218,0.02378569,0.01469789,0.0117283],"genre_scores_gemma":[0.6363358,0.0003937367,0.338857,0.0003095404,0.0001331948,0.00697579,0.01406763,0.0003449854,0.002582359],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.964502,"threshold_uncertainty_score":0.1877335,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1173736023270483,"score_gpt":0.4999297432818502,"score_spread":0.382556140954802,"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."}}