navin-engine

Your agent writes the patch. This proves it.

An AI coding agent can produce a plausible change in seconds. Nothing in that loop tells you whether the service still survives a restart under load, or whether the "optimization" actually made anything faster. `navin-engine` is the missing half: it boots your app in a throwaway git worktree, attacks it, measures it, applies a candidate patch, measures again, and reports what changed with numbers instead of adjectives.

AI proposes. Navin proves.

Isolated	Every candidate runs in a disposable git worktree. Your working tree is never written to.
Measured	Baseline and candidate under identical load, repeated to separate a real gain from noise.
Regression-gated	Break one passing test, answer one byte differently on replayed traffic, and the candidate is rejecte