{"id":"W4413180511","doi":"10.23919/cycon65856.2025.11103589","title":"Vulnerability Patch Verification for Military Software Systems Through AI-Driven Code-Level Rule Generation","year":2025,"lang":"en","type":"article","venue":"","topic":"Software Reliability and Analysis Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Defence Research and Development Canada; McGill University; Queen's University","funders":"","keywords":"Computer science; Code (set theory); Code review; Vulnerability (computing); Static program analysis; Software; Software engineering; Programming language; Software development; Computer security","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":[0.002419523,0.001312032,0.0008858522,0.001957131,0.0006189354,0.001900099,0.00278147,0.001271496,0.003835069],"category_scores_gemma":[0.01721523,0.0008985632,0.002317935,0.0006799657,0.001735515,0.00209153,0.00238221,0.002385768,0.001838573],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001223502,"about_ca_system_score_gemma":0.003565857,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007718921,"about_ca_topic_score_gemma":0.01334684,"domain_scores_codex":[0.9960381,0.0007235954,0.0002798401,0.0007537618,0.001913436,0.0002912508],"domain_scores_gemma":[0.988806,0.006404099,0.0008280666,0.001964648,0.001791929,0.0002052272],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0004168763,0.0007722915,0.01508263,0.001066192,0.0003118343,0.001526778,0.0008562815,0.4358758,0.06596983,0.03289909,0.0173235,0.4278988],"study_design_scores_gemma":[0.00003785338,0.00006174868,0.0003956808,0.00004545326,0.00004062198,0.0001198005,0.00003680394,0.9717926,0.01470007,0.009129336,0.003615176,0.00002489332],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03593659,0.0001965375,0.9125161,0.0003281702,0.0001050076,0.0007191772,0.0009541204,0.04561821,0.003626127],"genre_scores_gemma":[0.2228862,0.0002346235,0.7665761,0.0003788294,0.00004511051,0.0006001132,0.003859747,0.00285952,0.002559809],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007718921,"threshold_uncertainty_score":0.01534796,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07424323189957673,"score_gpt":0.3458651863693802,"score_spread":0.2716219544698035,"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."}}