RISE release acceptance protocol

This protocol closes the judgments that automation cannot make. A build may be technically green while the release remains blocked. Never replace a missing human verdict with a generated value.

1. Exact-edition certification

1. Run npm run release:prepare-certification.
2. Review out/release/archive-certification-candidates.json, prioritizing the three Keystone editions, then completing 10–15 shelf candidates.
3. For each edition, compare the named reference and pinned revision against RISE: opening, closing, division structure, representative internal spans, and every detector/signature disposition.
4. Copy only completed rise.archive-certification.v1 records into src/content/archive/certifications.json, keyed by workId.
5. Run npm run release:check. A changed edition identity must withdraw stale certification automatically.

The reviewer—not the preparation script—sets edition choice, rationale, comparison verdicts, disposition count/reviewer, detector verdict, and dates.

2. Acoustic acceptance

1. Run npm run release:prepare-acoustic-review.
2. Open out/release/acoustic-review/index.html and listen to every phrase.
3. Mark a phrase Defect for clipping, boundary errors, mispronunciation, corrupt audio, or unacceptable level. Correct and regenerate the affected source before continuing; regenerated bytes create a new review identity.
4. When all phrases pass, enter the reviewer name and export acoustic-review-evidence.json.
5. Run npm run release:record-acoustic-review -- --input <downloaded-json>.

The recorder accepts only an exact voice/review hash, exactly one passing decision for every expected phrase, zero defects, a reviewer, and a timestamp. It writes the complete decision ledger into release-evidence.json.

3. Real-device matrix

Witness the release build on all five required families:

• iPhone Safari;
• Android Chrome;
• desktop Chrome;
• desktop Safari;
• desktop Firefox.

For each, test landing-to-Keystone entry, cold reload and browser history, recitation and reveal independence, pause/resume, division continuation, reduced motion where available, and a complete short reading. Record passed: true, reviewer, and ISO completion time in release-evidence.json.

4. Stranger corridor

Observe at least three unprompted participants. All must understand entry, complete a reading, and understand the next action without coaching. Record the aggregate attestation, reviewer, and completion time in release-evidence.json.

5. Final admission

Using the pinned Node version and FFmpeg binary, run:

npm run test:run
npm run build
npm run test:e2e
npm run release:check

Release only when release:check reports zero blocked and zero human gates, the worktree is intentionally reviewed, and the deployed candidate is the same commit and asset set that produced the evidence.

---

Published from docs/RELEASE-ACCEPTANCE-PROTOCOL.md. Edits made here are overwritten on the next push to main.