{"id":"W4402519017","doi":"10.1002/prp2.70004","title":"Evaluating the quality of multiple‐choice question pilot database: A global educator‐created tool for concept‐based pharmacology learning","year":2024,"lang":"en","type":"article","venue":"Pharmacology Research & Perspectives","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta; Western University","funders":"Monash University","keywords":"Context (archaeology); Quality (philosophy); Computer science; Process (computing); Cognition; Medical education; Medicine","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":[],"consensus_categories":[],"category_scores_codex":[0.1054074,0.0007274072,0.001080982,0.003592609,0.0006599054,0.001552956,0.001413098,0.0008778421,0.003620601],"category_scores_gemma":[0.1896388,0.0005352316,0.00096937,0.002135347,0.0007880165,0.00210517,0.002713092,0.0008086949,0.0007282798],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001660058,"about_ca_system_score_gemma":0.002603122,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008750628,"about_ca_topic_score_gemma":0.001438253,"domain_scores_codex":[0.941727,0.04217236,0.005849931,0.002267753,0.007233344,0.0007497672],"domain_scores_gemma":[0.6801427,0.2443776,0.01545065,0.01469149,0.04265309,0.002684584],"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":[0.004186061,0.006187249,0.2334571,0.002161182,0.0003421229,0.0004566873,0.02286455,0.005359887,0.01297497,0.001740621,0.008891524,0.701378],"study_design_scores_gemma":[0.004465909,0.03256967,0.7778056,0.001801699,0.0008552299,0.0008704518,0.01830169,0.07205619,0.04417588,0.003132352,0.0434491,0.0005163282],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9112073,0.0002373484,0.06326631,0.0005787601,0.0001220291,0.01743964,0.001948991,0.001285179,0.003914389],"genre_scores_gemma":[0.8360671,0.0001415875,0.1428085,0.0002821013,0.00004818109,0.0181177,0.001470802,0.0001447586,0.0009192217],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1054074,"threshold_uncertainty_score":0.5574541,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4582604395447449,"score_gpt":0.6897479715448398,"score_spread":0.2314875320000949,"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."}}