A guard is only as good as what it actually compares. Mine kept comparing the wrong thing, and each time the review that caught it ran the real tool instead of reading the code.
Adversarial code review that runs the code beats adversarial code review that reads the code.
Each round found something real. The findings got smaller every time — until they stopped.
A tutorial-shaped feature request turned into a lesson about checking for prior art before designing, and about what adversarial code review actually catches that a careful first pass doesn't.