The Audit You Can Audit: In Depth

The full problem analysis behind the brief. Four failure modes of codebase audits — evidence decay, analytical silos, the findings cliff, and, new with agentic AI, the unaccountable auditor — grounded in four decades of software-decay research and the multi-agent-failure literature. It traces the "who audits the auditor" problem from Juvenal through agency theory to AI scalable-oversight research, and derives five principles for a trustworthy audit.