Stable findings make regression policy possible.

Why raw counts fail

Line movement, formatting changes, and detector revisions should not make every existing observation appear new. Vyper Guard derives a stable finding identity from the rule, source context, and semantic location, then compares the current scan with a reviewed baseline.

Policy follows review

A baseline is not dismissal. Existing findings remain visible while the release gate focuses on qualifying regressions. Accepted findings carry reasons and expiry in managed workflow.