{"id":"W4404293607","doi":"10.36834/cmej.72320","title":"User experience of the Written Exam Question Quality tool to inform the writing of new written-exam questions","year":2024,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Sherbrooke","funders":"","keywords":"Task (project management); Perception; Quality (philosophy); Computer science; Test (biology); Psychology; Medical education; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"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":[0.01724738,0.0005762068,0.0003671203,0.001209421,0.001188029,0.001896883,0.0007400457,0.0008429214,0.003097872],"category_scores_gemma":[0.07230172,0.0002948194,0.0004635259,0.0007638625,0.0009662091,0.001324143,0.001661381,0.0006276614,0.0005052807],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001605073,"about_ca_system_score_gemma":0.002263772,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01062214,"about_ca_topic_score_gemma":0.01647446,"domain_scores_codex":[0.9893174,0.007509888,0.0006089594,0.0003837474,0.00153778,0.0006423612],"domain_scores_gemma":[0.9116214,0.06709557,0.004116105,0.00295495,0.01057298,0.00363899],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.001393985,0.001390454,0.1897189,0.001273581,0.00009051539,0.002245578,0.4405807,0.001090398,0.04158248,0.000713594,0.00682658,0.3130932],"study_design_scores_gemma":[0.0004350403,0.009949629,0.4883216,0.001720809,0.0003595323,0.005320792,0.3348487,0.01510653,0.03595036,0.001282629,0.1058783,0.0008261569],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9868204,0.0001686172,0.008657854,0.0004608817,0.00002564468,0.000264702,0.0001307263,0.0004188679,0.003052268],"genre_scores_gemma":[0.9797928,0.0001274303,0.01793782,0.0002586868,0.00001818092,0.0001826882,0.000132068,0.0000696219,0.001480645],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01724738,"threshold_uncertainty_score":0.09121394,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03389104683518359,"score_gpt":0.4116215754441866,"score_spread":0.3777305286090031,"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."}}