Benchmark outcomes log
Keep these in your own repo (e.g. docs/perf-outcomes/). One file per locked-in perf change. Numbers are machine-specific; patterns and rules are what we keep.
When to append
After Phase D (lock-in), not after investigation-only runs.
Filename
YYYY-MM-DD-<project-or-domain>-<short-slug>.md
Example: 2026-01-15-checkout-form-validation.md
Template (copy per entry)
# <title>
- **Date:** YYYY-MM-DD
- **Project:** repo or product name
- **User-visible action:** e.g. Enter / Copy in picker
- **Suites:** `bun tools/benchmark/<suite>.mjs` (+ live trace if used)
- **Runtime:** Bun x.x · happy-dom / Chrome trace
## Locked in
| Change | Module | Ship gate |
|---|---|---|
| … | `src/…` | pass / fail + why |
## Round 0 → shipped (µs/op or ms)
| Label | Before | After | × vs ref |
|---|---|---|---|
## Learnings → perf rules
- Rule N: … (portable; no golden µs)
## Live trace (optional)
Link or summary from [perf-live-traces.md](perf-live-traces.md).
## Next ideas (not shipped)
- …Perf rules index
Derived rules live in perf-rules.md. After each lock-in, add or tighten a rule if the learning generalizes — treat it like lint policy for hot paths.