quantify-impact

Read Poteto. For performance optimization, read the experiment loop before edits. For repeated candidates, use hill-climb.

Assess every PR creation/update and session PR check, published or not. Reuse valid evidence; rerun invalid comparisons. Assessment grants no publication authority.

Flow

1. Evidence opportunity scan: benchmark only when a cheap metric can clear a predeclared worthwhile delta and variance. Tiny copy/style/test work gets a value sentence.
2. Lock claim before coding: thesis, primary metric, guardrail, scenario, minimum worthwhile delta. Never pick the winner afterward.
3. Product lane + Codebase lane: one must improve; the other must not materially regress.
   • Product: capability, task success, repro, errors, steps, latency, resources.
   • Codebase: verifiable contracts, change coordination, dependencies, warnings, leaks, bundle/build/test cost; follow verifiability, not size/complexity scores.
4. Proportional rigor: sentence for obvious value; deterministic repro/count for correctness; paired REFERENCE.md benchmarks for runtime. Measure explicit performance claims.
5. Base: measure before coding or reconstruct merge-base. Use the same scenario, fixture, config, and machine for base/candidate.
6. Compare: claim only threshold-clearing deltas with method, environment and noise. Diagnostic attempts are not proven impact; tests/proxies alone do not prove speed.
7. Decide:
   • Worthwhile gain: Value proven.
   • Ambiguous/negligible performance claim: Value not proven; no metric-shopping. Revise a hypothesis once with evidence, then scrap it; distinct supported leads remain open.
   • No useful metric and no performance claim: normal value summary, no impact artifact.
   • Regression: fix, narrow, or stop.

Filter guardrails; omit negligible movement.

For runtime/compiler-cost measurements, load performance analysis.

PR output

State user/maintainer value and evidence; tiny copy/style edits use visual proof, not invented metrics. Only threshold-clearing evidence uses:

## Proven impact
| Metric | Before | After | Delta |
|---|---:|---:|---:|
| <direct metric> | <base> | <candidate> | <absolute and %> |
**Value proven:** <product or codebase benefit>
Method: `<command, fixture, runs, environment>`.

Publish via /commit-push-pr or /stacked-prs; /make-pr-easy-to-review reuses on request. Keep raw data locally. No empty tables or negligible claimed gains; unproven claims say Value not proven.