{"id":"W4413304054","doi":"10.1080/0969594x.2025.2549158","title":"Diversity: a necessary imperative for assessment research","year":2025,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Diversity (politics); Data science; Engineering ethics; Management science; Computer science; Sociology; Engineering; Anthropology","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","sts"],"consensus_categories":[],"category_scores_codex":[0.02525773,0.0001768253,0.0002804151,0.001556178,0.001825767,0.0006116584,0.000613855,0.0001173123,0.0001691306],"category_scores_gemma":[0.01567403,0.0001535441,0.00006103333,0.002260363,0.0001579875,0.002082949,0.00112477,0.0005452217,0.00001669689],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008331924,"about_ca_system_score_gemma":0.01020887,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009658303,"about_ca_topic_score_gemma":0.0003781455,"domain_scores_codex":[0.9943515,0.002190822,0.0008040874,0.0006702925,0.001559229,0.0004240656],"domain_scores_gemma":[0.9763798,0.02039728,0.0003561782,0.0006250004,0.00210383,0.0001378779],"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":[0.0001079399,0.0007498134,0.05762774,0.00003452412,0.00004095611,6.210196e-7,0.004673721,0.0001712208,0.00007685571,0.8599273,0.006510029,0.0700793],"study_design_scores_gemma":[0.0008322081,0.0001620205,0.2861777,0.00005923019,0.0000280088,0.000005791066,0.04507035,0.005298485,0.00004607498,0.1125585,0.5495876,0.0001740563],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2229315,0.0003580497,0.01053343,0.2522392,0.001347215,0.002424146,0.00002960952,0.00004255144,0.5100943],"genre_scores_gemma":[0.9600778,0.0006724112,0.01874498,0.006185903,0.0002196507,0.0005732451,0.00002753024,0.000008870305,0.01348966],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7473688,"threshold_uncertainty_score":0.9994737,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4676913733950703,"score_gpt":0.6664621529759441,"score_spread":0.1987707795808739,"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."}}