{"id":"W2054837185","doi":"10.1111/medu.12302","title":"Test‐enhanced learning and its effect on comprehension and diagnostic accuracy","year":2014,"lang":"en","type":"article","venue":"Medical Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":79,"is_retracted":false,"has_abstract":true,"ca_institutions":"Memorial University of Newfoundland; The Wilson Centre; University of Toronto","funders":"","keywords":"Test (biology); Comprehension; Psychology; Educational measurement; Computer science; Medicine; Curriculum; Pedagogy; Biology","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0005054049,0.0001111231,0.0002401309,0.00004855934,0.00007887316,0.00001669182,0.00003067621,0.000128594,0.0001655977],"category_scores_gemma":[0.6486655,0.00008239048,0.00002529163,0.00008157187,0.00006832831,0.00002679842,0.00002986945,0.0003370021,0.00007811368],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001822282,"about_ca_system_score_gemma":0.0001527194,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001763237,"about_ca_topic_score_gemma":0.000001408482,"domain_scores_codex":[0.9989514,0.000121733,0.0001908825,0.0002622215,0.0003244959,0.0001492313],"domain_scores_gemma":[0.9062041,0.093037,0.00007955039,0.0001322311,0.00008357173,0.0004635403],"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":[0.00007101214,0.0004983216,0.06489488,0.0001136044,0.000008946924,0.000003935281,0.0001433976,9.695962e-7,0.0006571527,0.0002308286,0.002860802,0.9305161],"study_design_scores_gemma":[0.004834308,0.006162371,0.9424399,0.0119519,0.0002222706,0.0001253425,0.00006249571,0.00671288,0.003539034,0.0003936136,0.02323541,0.0003204697],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9910201,0.0006396182,0.00008436534,0.005772832,0.0003276016,0.0002435568,2.580552e-7,0.00005643431,0.001855209],"genre_scores_gemma":[0.9950945,0.0005423735,0.00006440981,0.003595456,0.0003618505,0.00003417271,0.00002975547,0.00001238129,0.0002651133],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9301957,"threshold_uncertainty_score":0.3542939,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.009704307280209934,"score_gpt":0.3583462339113984,"score_spread":0.3486419266311885,"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."}}