{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.004424936,0.00015398,0.0003421264,0.0002687155,0.0001252656,0.0001187739,0.00009035831,0.0001260699,0.00008103303],"category_scores_gemma":[0.01137784,0.0001452935,0.00005440595,0.000624461,0.0000696046,0.0006222947,0.00008937658,0.0006079813,0.00000856379],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004046254,"about_ca_system_score_gemma":0.006729583,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00147244,"about_ca_topic_score_gemma":0.00011755,"domain_scores_codex":[0.9972739,0.001076244,0.0007672189,0.0003773188,0.0002763967,0.0002289566],"domain_scores_gemma":[0.9915717,0.00704345,0.0004022627,0.0004968881,0.0003228628,0.0001628472],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.0002148063,0.002043153,0.08875023,0.0001796482,0.00008801065,0.000002710492,0.0001837042,0.00000476088,0.0002924133,0.8654642,0.0003656472,0.04241073],"study_design_scores_gemma":[0.0008734473,0.00007078849,0.5200362,0.00007683155,0.00009641235,0.00001087166,0.001944038,0.0001169969,0.000008087694,0.002349391,0.4743272,0.00008971213],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.1923656,0.0005430828,0.0008745148,0.309757,0.0005623841,0.0008463145,0.00001092087,0.00005609962,0.4949841],"genre_scores_gemma":[0.9325048,0.001131605,0.0204473,0.0404893,0.0002842409,0.0001326576,0.00003477881,0.00001024576,0.004965098],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8631148,"threshold_uncertainty_score":0.9989014,"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."}}