Phase-1b: require full cold-start re-verification of all Phase-1 D1–D10 as the final gate
RL3 strengthened: after lint/review findings are responded to and fixed, the Adversary independently re-verifies EVERY Phase-1 Definition-of-Done item (D1–D10) from a cold start to the same bar as Phase 1's own DONE (fresh PASS + evidence in REVIEW.md), proving the cleanup regressed nothing. 1b cannot be DONE until all D1–D10 are re-confirmed green post-cleanup. Method/W2 updated to make the ordering explicit (tooling -> fixes -> re-verify). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
@ -45,9 +45,13 @@ in `REVIEW.md`):
|
||||
over the codebase; every **blocking** finding is fixed; every **advisory** finding is either
|
||||
fixed or moved to `BACKLOG.md`/`IDEAS.md` as `[idea]` with a one-line rationale. Findings +
|
||||
resolutions recorded in `REVIEW.md`.
|
||||
- [ ] **RL3 — Still green, nothing weakened.** After the cleanup, the full Phase-1 Definition of Done
|
||||
(`plan.md` §2 D1–D10) **still holds** — the Adversary re-confirms a representative `!testme` run
|
||||
is still green and no test/assertion was softened or skipped to satisfy a lint/review nit.
|
||||
- [ ] **RL3 — Full Phase-1 re-verification (the final gate).** This is the **last step before 1b is
|
||||
marked DONE**, performed *after* all lint/review findings (RL1+RL2) have been responded to and
|
||||
resolved. The Adversary **independently re-verifies every Phase-1 Definition-of-Done item
|
||||
D1–D10 from a cold start** — re-running each acceptance check itself, to the **same bar as
|
||||
Phase 1's own DONE** (fresh PASS + evidence + timestamps logged in `REVIEW.md` within 24h). 1b
|
||||
cannot be DONE until **all of D1–D10 are re-confirmed green post-cleanup** and the Adversary
|
||||
confirms no test/assertion was softened, skipped, or otherwise regressed by the cleanup.
|
||||
- [ ] **RL4 — Documented.** `docs/` notes how to run lint/format locally and that CI enforces it; any
|
||||
accepted deviations from the review checklist are in `DECISIONS.md`.
|
||||
|
||||
@ -62,7 +66,9 @@ When RL1–RL4 hold and are confirmed, write `## DONE` to Phase-1b `STATUS.md`.
|
||||
style should just be auto-fixed — don't deliberate over it.)
|
||||
2. **Then the review checklist (§3).** Read the code against the checklist; classify each finding
|
||||
**blocking** vs **advisory**; fix blocking, triage advisory.
|
||||
3. **Re-verify.** Confirm Phase-1 §2 still passes (RL3) — cleanup must not regress behavior or tests.
|
||||
3. **Re-verify everything last.** Once all fixes have landed, the Adversary re-runs the **entire**
|
||||
Phase-1 D1–D10 acceptance suite from a cold start (RL3) — this is the final gate; cleanup must
|
||||
regress nothing. Order matters: tooling → review responses/fixes → *then* full re-verification.
|
||||
4. **Bound it.** Cap iterations; this is a pass, not a rewrite. Record dead-ends/deviations and stop.
|
||||
|
||||
---
|
||||
@ -103,8 +109,10 @@ Blocking unless noted; these are *plan-relevant invariants visible only by readi
|
||||
checkout.
|
||||
- **W1 — Review + fixes.** Checklist (§3) run; blocking findings fixed; advisory triaged. *Accept:*
|
||||
`REVIEW.md` lists findings + resolutions; no blocking item open.
|
||||
- **W2 — Re-verify + document.** Phase-1 §2 still green; docs updated; deviations in `DECISIONS.md`.
|
||||
*Accept:* a `!testme` run is still green post-cleanup; flip Phase-1b `STATUS.md` to `## DONE`.
|
||||
- **W2 — Full re-verification + document.** *After* W0+W1 fixes land, the Adversary re-verifies **all
|
||||
of D1–D10** cold (RL3) and docs/deviations are recorded. *Accept:* every Phase-1 D1–D10 item has a
|
||||
fresh post-cleanup PASS + evidence in `REVIEW.md` (nothing weakened); then flip Phase-1b
|
||||
`STATUS.md` to `## DONE`.
|
||||
|
||||
---
|
||||
|
||||
|
||||
Reference in New Issue
Block a user