{"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":"codex-gemma-dda1882f352a","candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01239645,0.0001643775,0.0002564955,0.000182328,0.001437901,0.0001118143,0.0005365597,0.00009289422,0.002035819],"category_scores_gemma":[0.005122906,0.0001390054,0.0001177419,0.0009666976,0.001874001,0.000329611,0.0001066025,0.0007800922,0.00002392221],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009184098,"about_ca_system_score_gemma":0.004943179,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007131557,"about_ca_topic_score_gemma":0.001033462,"domain_scores_codex":[0.9895322,0.007939049,0.0003788795,0.0005935896,0.0008420706,0.0007142555],"domain_scores_gemma":[0.9765399,0.02158616,0.000130166,0.0001310198,0.001472899,0.0001399203],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"qualitative","study_design_scores_codex":[0.003529055,0.002834749,0.09104154,0.0005534135,0.001241658,0.000008038909,0.07803916,0.001362382,0.5443929,0.2024934,0.06453449,0.009969204],"study_design_scores_gemma":[0.01414006,0.008859972,0.127547,0.0004468907,0.001365162,0.00001050492,0.4163971,0.07744578,0.03478011,0.01522951,0.3016759,0.002101991],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9594554,0.005224701,0.0004990532,0.0282652,0.001780843,0.002141022,0.000174624,0.0001551788,0.002303984],"genre_scores_gemma":[0.9944084,0.0002235083,0.0005560608,0.0001733914,0.002264434,0.0009214213,0.00007597684,0.0000175096,0.001359254],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5096129,"threshold_uncertainty_score":0.9998621,"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."}}