{"id":"W1979809802","doi":"10.1007/s10459-008-9129-z","title":"The effect of reused questions on repeat examinees","year":2008,"lang":"en","type":"article","venue":"Advances in Health Sciences Education","topic":"Radiology practices and education","field":"Medicine","cited_by":36,"is_retracted":false,"has_abstract":false,"ca_institutions":"Medical Council of Canada","funders":"","keywords":"Test (biology); Psychology; Multiple choice; Block (permutation group theory); Social psychology; Medicine; Medical education; Significant difference; Mathematics","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":[0.001442491,0.0000651841,0.0001441023,0.0001278981,0.0004264156,0.000005332687,0.0001109708,0.00002793378,0.000006265362],"category_scores_gemma":[0.001506722,0.00003965809,0.00002083453,0.0004914394,0.000402276,0.0002737549,0.00000599444,0.0001216322,0.000005989604],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001149129,"about_ca_system_score_gemma":0.0007554717,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002548824,"about_ca_topic_score_gemma":0.00006157844,"domain_scores_codex":[0.9989341,0.0002116915,0.0003035662,0.0001919585,0.000179572,0.0001790353],"domain_scores_gemma":[0.9982063,0.001210189,0.000244821,0.0002315065,0.00004542352,0.00006174808],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002075881,0.0002436448,0.8194948,0.0002363285,0.000004012496,7.509261e-7,0.001842139,0.0009629598,0.00009011855,0.003147449,0.001903652,0.1718666],"study_design_scores_gemma":[0.0005667035,0.00669553,0.9048448,0.0006172574,0.00001197555,0.0002173087,0.001968123,0.0005278037,0.0009395809,0.0008247686,0.08266183,0.0001243019],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9714665,0.01218291,0.00001650996,0.01059287,0.001450657,0.0004300544,3.175248e-7,0.00001397676,0.003846237],"genre_scores_gemma":[0.9826258,0.01572059,0.0006802051,0.0003811086,0.0001431785,0.00007491836,0.000004288516,0.000003241813,0.0003666401],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1717422,"threshold_uncertainty_score":0.3279687,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0248753314000006,"score_gpt":0.4252534904192208,"score_spread":0.4003781590192202,"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."}}