{"id":"W4389099788","doi":"10.3138/jvme-2023-0061","title":"Measures for Quality Assurance of Electronic Examinations in a Veterinary Medical Curriculum","year":2023,"lang":"en","type":"article","venue":"Journal of Veterinary Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Reliability (semiconductor); Quality assurance; Quality (philosophy); Medicine; Status quo; Medical education; Medical physics; Curriculum; Psychology; External quality assessment; Pathology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0660622,0.0005430425,0.0007774582,0.007760778,0.0008015124,0.001698675,0.00104139,0.0005483402,0.002106656],"category_scores_gemma":[0.1817282,0.0003220651,0.0009650297,0.006970638,0.0008504243,0.001961714,0.001710596,0.000775455,0.0003959051],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001965722,"about_ca_system_score_gemma":0.002657823,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001078739,"about_ca_topic_score_gemma":0.001305766,"domain_scores_codex":[0.9346269,0.03228902,0.0113327,0.002295529,0.01845589,0.0009999936],"domain_scores_gemma":[0.7251344,0.1512993,0.05959805,0.01624654,0.04530853,0.002413264],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006443747,0.001088348,0.7102377,0.0013342,0.0003307768,0.00005746257,0.003352157,0.001817637,0.002343165,0.001620258,0.001908122,0.2752657],"study_design_scores_gemma":[0.0000371788,0.001848292,0.9872023,0.000324778,0.00009281701,0.0001066511,0.001515455,0.002856284,0.002466052,0.0005655782,0.002947352,0.00003714474],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9061909,0.002514753,0.06812798,0.0007060096,0.0001861764,0.004279342,0.00186769,0.0004162211,0.01571107],"genre_scores_gemma":[0.9350265,0.0004138644,0.06005727,0.00009309565,0.00007839724,0.002370444,0.001096343,0.00002987709,0.0008341694],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0660622,"threshold_uncertainty_score":0.3493745,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08635386521717388,"score_gpt":0.4648867203011789,"score_spread":0.3785328550840051,"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."}}