carlos / platform Public

Astra stage B round 2 brief

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Paul Campbell pushed by claude 2 weeks ago 1651ff7dd5297f541acbf8d8eba364e365206547 parent a73c430

1 file changed, +23 −0

docs/superpowers/plans/CODEX-PLAN-REGION-HA-STAGE-B-BRIEF-2.md +23 −0

CI failed — run details

diff --git a/docs/superpowers/plans/CODEX-PLAN-REGION-HA-STAGE-B-BRIEF-2.md b/docs/superpowers/plans/CODEX-PLAN-REGION-HA-STAGE-B-BRIEF-2.md

new file mode 100644

index 000000000..51f6a8cd3

--- /dev/null

+++ b/docs/superpowers/plans/CODEX-PLAN-REGION-HA-STAGE-B-BRIEF-2.md

@@ -0,0 +1,23 @@

+# Astra round 2 brief: region HA stage B plan rev 2

+**Review:** `docs/superpowers/plans/2026-10-10-region-ha-1b-next-stage-b.md` rev 2 (a73c430f). Its disposition table at the top answers your round 1 (`CODEX-PLAN-REGION-HA-STAGE-B-R1.md`): 10 accepted, 4 changed, 0 declined, with new rulings 42–54. Round 1's brief still applies (`CODEX-PLAN-REGION-HA-STAGE-B-BRIEF-1.md`).

+**Judge:**

+1. **Is each round 1 finding closed?** For each one, does the named test really prove the fix? Findings 2 and 8 are proved by one end-to-end test that starts from an empty target with the real flags and runs through a real probe to the serving wake. Is that test specified concretely enough to be built, and would it fail on rev 1's design?

+2. **The four changed dispositions.**

+ - 4: every target uses the nested layout.

+ - 8: an explicit stop, receipt and completion before serving.

+ - 10: main's WorkSource with only its policy replaced.

+ - 13: no reactivation of a stamp. The restore of a retired install is dropped.

+ Are these sound?

+3. **New claim asks and fallbacks.**

+ - C6: a narrower cleanup for a dropped target's unopened install.

+ - C7: the first serving wake hands over from the stopped probe session. This is an activation gate.

+ - C9: the claim policy admits install actions for a relocating target.

+ - C8 + O1: moves' files provider publishes `<host>.env` before the app starts.

+ Is each fallback safe? Is C7 really unavoidable, or can region HA close it within main's existing cold-handover contract?

+4. **Ruling 54 / O2.** The pinner now accepts `CARLOS_INSTANCE_KEY` in the pinned `<host>.env` as the instance key. Note that a separate review of the instance-key plan found that a platform key's presence does not prove the app uses it. An app may seal with its own `instance.key`. Does O2 reopen ruling 25's hazard? What evidence should it require?

+5. **Removed coverage.** Rev 2 deletes stage A's controller set and its 200-seed lifecycle property, in favour of main's participation owner. Is coverage lost that must be replaced?

+6. **Anything new.** Did any fix introduce a new defect?

+Give findings as P1/P2/P3 with file:line, then a verdict: ready to build, or not.