Optimization proof · observed 2026-08-14

Every project gets its own evidence boundary.

These are separate local trials, not one blended headline. Each page names the exact flow, revision, source candidate, patch cost, correctness scope, repeated measurement, decision, availability, and limitation. A rejected experiment remains visible.

How Optimize works

Anime List retained 35.8% fewer A two-file import-boundary change materially reduced one local Vite navigation, while a separate production build showed almost no shipped-bundle movement. 2 files · 62 gross lines · net 58 · 0 dependencies 2026-08-14 Free AI confirmed 29.8% faster A one-pass model-selection loop removed an intermediate array and improved every tested registry size without growing the patch. 1 file · 45 gross lines · net -7 · 0 dependencies 2026-08-14 Starboard rejected 5.3% slower A plausible incremental token scanner passed focused correctness but ran slower, so CodeVetter rejected it and restored the original code. 1 file · 19 gross lines · net 5 · 0 dependencies 2026-08-14

These three receipts are the intentionally small public set. Historical research capability is labeled separately from the compact runtime path currently shipped in CodeVetter.