{"id":"W4416613148","doi":"10.1080/02602938.2025.2587246","title":"What should we be assessing exactly? Higher education staff narratives on gen AI integration of assessment in a postplagiarism era","year":2025,"lang":"en","type":"article","venue":"Assessment & Evaluation in Higher Education","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Calgary Laboratory Services; Brock University; University of Calgary","funders":"","keywords":"Higher education; Narrative; Qualitative research; Professional development; Semi-structured interview; Focus group; Educational 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":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.006313928,0.0003390222,0.0004279833,0.001175366,0.0003614619,0.0005433304,0.0004172318,0.0009324633,0.002461789],"category_scores_gemma":[0.0002046023,0.0003505508,0.000099832,0.001657201,0.0001776194,0.003297493,0.00004791109,0.002105267,0.000007966151],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003698055,"about_ca_system_score_gemma":0.01197936,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002945645,"about_ca_topic_score_gemma":0.001234901,"domain_scores_codex":[0.9930885,0.002947254,0.001120415,0.000686025,0.001734975,0.0004228311],"domain_scores_gemma":[0.9970575,0.0006961958,0.000623391,0.0004119959,0.001111712,0.0000991878],"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":[0.00006242713,0.00358112,0.02445444,0.000115606,0.00006542753,4.58381e-7,0.03261785,0.0005876833,0.003047376,0.7837499,0.02250677,0.1292109],"study_design_scores_gemma":[0.001032031,0.0001592426,0.6647667,0.00261934,0.0001212115,3.421576e-7,0.1155022,0.001277647,0.0004483693,0.1477941,0.06572633,0.0005525155],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5385379,0.001419474,0.0008999713,0.2981523,0.01665317,0.003179238,0.00001104099,0.00008901145,0.141058],"genre_scores_gemma":[0.9815541,0.0005632706,0.00313843,0.003691182,0.0004630474,0.0008635154,0.0003892441,0.00002335266,0.009313853],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6403123,"threshold_uncertainty_score":0.9998946,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1128112722653669,"score_gpt":0.4954028071868408,"score_spread":0.3825915349214739,"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."}}