{"id":"W7118245466","doi":"10.1080/0969594x.2025.2607856","title":"Assessment quality","year":2025,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Quality (philosophy); Quality assessment; Educational assessment; Standards-based assessment; Risk assessment","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3105044,0.000925481,0.003634484,0.01309789,0.003612871,0.01325931,0.00449139,0.001939799,0.03887416],"category_scores_gemma":[0.7017449,0.001051137,0.004400853,0.01509002,0.004153308,0.007958222,0.009859078,0.004695031,0.006956594],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01667343,"about_ca_system_score_gemma":0.0517221,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01794302,"about_ca_topic_score_gemma":0.01264636,"domain_scores_codex":[0.4925662,0.202292,0.1040022,0.02841787,0.1661517,0.00656992],"domain_scores_gemma":[0.2101776,0.2514169,0.05892863,0.1050829,0.3636531,0.01074091],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0009483342,0.0001918094,0.04966017,0.01215331,0.001188339,0.00008937832,0.00557376,0.001442349,0.0004933787,0.04421252,0.1195124,0.7645343],"study_design_scores_gemma":[0.0005024382,0.0005320287,0.07816793,0.02354003,0.001182503,0.0006052424,0.002414229,0.003333947,0.002630342,0.04081749,0.8459522,0.0003216537],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.04233078,0.05418697,0.3900444,0.08648247,0.01255701,0.03622507,0.06034259,0.004896308,0.3129345],"genre_scores_gemma":[0.5798776,0.02202249,0.2474038,0.02427992,0.004864691,0.03580578,0.03879227,0.00305727,0.04389617],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3105044,"threshold_uncertainty_score":0.8502708,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09899369132097026,"score_gpt":0.5616235891664735,"score_spread":0.4626298978455032,"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."}}