{"id":"W4412537309","doi":"10.1109/ast66626.2025.00009","title":"Simulink Mutation Testing using CodeBERT","year":2025,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Computer science; Mutation; Mutation testing; Software testing; Genetics; Programming language; Biology; Software","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":[],"consensus_categories":[],"category_scores_codex":[0.0002039093,0.00007056567,0.00007387061,0.0001144635,0.0001249776,0.0001201501,0.0003302413,0.00003602977,0.000001849357],"category_scores_gemma":[0.00097669,0.00006498171,0.00001960581,0.0006518514,0.00001606215,0.0001823123,0.0001474645,0.00006452179,0.000006729596],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003485706,"about_ca_system_score_gemma":0.00008131497,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002163784,"about_ca_topic_score_gemma":0.000001943549,"domain_scores_codex":[0.9994019,0.00002410231,0.0001344048,0.0002098281,0.00009159669,0.0001381344],"domain_scores_gemma":[0.9988787,0.0006526228,0.00003611354,0.0002929928,0.0001161118,0.00002348241],"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":[0.000003210099,0.00009600011,0.03544721,0.00006341295,0.00002496008,0.00005922086,0.0003627513,0.006756281,0.002222113,0.07052505,0.01938564,0.8650541],"study_design_scores_gemma":[0.00005910492,0.00001339039,0.0007354192,0.0000594396,0.000002772131,0.00001190746,0.000001239639,0.9071652,0.001709044,0.08999945,0.0001659048,0.00007712879],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.005552154,0.00002860457,0.9714063,0.0002660608,0.000137467,0.0000603983,9.289548e-8,0.01189379,0.01065508],"genre_scores_gemma":[0.4120785,1.261906e-7,0.5871512,0.0005639076,0.00001035376,0.000002068681,2.331384e-7,0.000002271766,0.0001912856],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9004089,"threshold_uncertainty_score":0.2649877,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04942265958739796,"score_gpt":0.3214422032418111,"score_spread":0.2720195436544131,"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."}}