{"id":"W2111648217","doi":"10.1111/j.1745-3992.2010.00181.x","title":"Developing Score Reports for Cognitive Diagnostic Assessments","year":2010,"lang":"en","type":"article","venue":"Educational Measurement Issues and Practice","topic":"Data Visualization and Analytics","field":"Computer Science","cited_by":68,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Computer science; Context (archaeology); Test (biology); Cognition; Diagnostic test; Hierarchy; Sample (material); Knowledge management; Data science; Psychology; 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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.001502616,0.00009186042,0.00008233803,0.000053845,0.0002089524,0.0003417542,0.0001405197,0.00003120907,0.00003432103],"category_scores_gemma":[0.01656031,0.00008870458,0.00001818056,0.0001446674,0.00002473697,0.001010508,0.00005850535,0.00008721004,0.0000113225],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003265068,"about_ca_system_score_gemma":0.0005229164,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000304628,"about_ca_topic_score_gemma":0.00002100082,"domain_scores_codex":[0.9988559,0.00006240166,0.0002096835,0.0002877398,0.0004547391,0.0001294728],"domain_scores_gemma":[0.9966125,0.001709468,0.0001961055,0.0001772344,0.00122394,0.00008072382],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002093245,0.0005529723,0.006985798,0.00009661648,0.0001417021,0.000008806072,0.0008880873,0.000002938094,0.0006654864,0.93872,0.02511074,0.02680593],"study_design_scores_gemma":[0.000409533,0.0001098099,0.02471811,0.0001847337,0.0001083729,0.0002051385,0.0004468932,0.001091455,0.001070919,0.03520958,0.9360324,0.0004130511],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.002726362,0.0007574659,0.9029152,0.08679445,0.00275249,0.0006183059,0.00001181337,0.00005981429,0.00336412],"genre_scores_gemma":[0.6360547,0.0002561946,0.3560744,0.005791929,0.0007179109,0.0001686242,0.0001381618,0.00001663455,0.0007814888],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9109216,"threshold_uncertainty_score":0.9917236,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.165780093144807,"score_gpt":0.4552123800807414,"score_spread":0.2894322869359344,"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."}}