{"id":"W7125641124","doi":"10.1109/cascon66301.2025.00040","title":"Evaluating Generative AI for CS1 Code Grading: Direct Vs Reverse Methods","year":2025,"lang":"","type":"article","venue":"","topic":"Teaching and Learning Programming","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia, Okanagan Campus; University of British Columbia; Okanagan University College; Okanagan Science Centre","funders":"","keywords":"Code (set theory); Source code; Generative grammar; Feature (linguistics)","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009682138,0.0009585255,0.0006560714,0.002311779,0.0006235808,0.002758547,0.001959857,0.001583359,0.004830955],"category_scores_gemma":[0.08993855,0.0003889074,0.0009361923,0.001034789,0.001239386,0.002431227,0.001824163,0.00207747,0.001750116],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001680606,"about_ca_system_score_gemma":0.002429975,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00314727,"about_ca_topic_score_gemma":0.005508754,"domain_scores_codex":[0.990846,0.004193089,0.0005860825,0.00115625,0.002942174,0.0002763665],"domain_scores_gemma":[0.8356512,0.1319706,0.005908984,0.01037832,0.01295806,0.003132861],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.005832918,0.005037329,0.05156249,0.001100868,0.000465751,0.00006682522,0.002014732,0.02438317,0.01130139,0.008867479,0.005562149,0.8838049],"study_design_scores_gemma":[0.004286277,0.02152842,0.0962504,0.0009272464,0.001430828,0.0007130861,0.002311646,0.7047668,0.09412262,0.04777228,0.02545339,0.0004370055],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8622429,0.001229894,0.1055781,0.0005020378,0.0003558076,0.000882439,0.0003983798,0.003493426,0.02531694],"genre_scores_gemma":[0.9171343,0.0002781302,0.07570168,0.0001782122,0.00005812533,0.000344999,0.0005600611,0.0003341803,0.00541031],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009682138,"threshold_uncertainty_score":0.05120462,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1356687935760371,"score_gpt":0.4992251913451128,"score_spread":0.3635563977690757,"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."}}