{"id":"W7130688282","doi":"10.1109/swc65939.2025.00060","title":"Automated Grading: Methods, Implementations, and Opportunities in Higher Education","year":2025,"lang":"","type":"article","venue":"","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Athabasca University; University of Alberta","funders":"Innovation Fund; Government of Alberta","keywords":"Grading (engineering); Workflow; Automation; Higher education","routes":{"ca_aff":true,"ca_fund":true,"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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001244105,0.000169637,0.0002355729,0.0004199532,0.0004490176,0.000353163,0.0001955196,0.0001117165,0.004958665],"category_scores_gemma":[0.00002889114,0.0001765381,0.00004322787,0.0007710555,0.000219549,0.0004479818,0.0001161153,0.0001085988,0.000008200418],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001943298,"about_ca_system_score_gemma":0.0008784817,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002142525,"about_ca_topic_score_gemma":0.0006262799,"domain_scores_codex":[0.9980768,0.0005956417,0.0004709235,0.0003074959,0.0002167568,0.0003323917],"domain_scores_gemma":[0.9992519,0.0003303485,0.0001137092,0.0001226142,0.00009045124,0.00009096483],"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":[0.00000699074,0.0002069922,0.06073177,0.0000409774,0.00006392636,6.392243e-7,0.008135544,2.014429e-7,0.0001062979,0.8232691,0.04385399,0.06358358],"study_design_scores_gemma":[0.0008586159,0.00004282149,0.4169882,0.0001403867,0.0001484272,2.126212e-7,0.2193445,0.0006289387,0.00005983645,0.02404656,0.337351,0.000390468],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.07199609,0.004006597,0.0001992251,0.05881608,0.004604949,0.001028785,0.00000577685,0.0002968792,0.8590456],"genre_scores_gemma":[0.7875844,0.004004033,0.007172965,0.002046136,0.0001226423,0.00008843729,0.00002318228,0.000008851966,0.1989493],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.7992225,"threshold_uncertainty_score":0.9959509,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1780403127498506,"score_gpt":0.5101519021792482,"score_spread":0.3321115894293977,"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."}}