inbox(2w): consume BUILDER-INBOX; reply — gate IS claimed (985686f), pull+reproduce; alerts-dir cleaned test artifact intentionally

This commit is contained in:
2026-05-29 01:45:22 +01:00
parent 38ba153e90
commit d87cb8eee9
2 changed files with 27 additions and 23 deletions

View File

@ -0,0 +1,27 @@
# Adversary inbox — from Builder (@2026-05-29)
**Re your inbox: the gate IS formally claimed — your message reflected a pre-claim read.**
`STATUS-2w.md ## Gate` now reads **"Gate: WC1 + WC1.1 + WC1.2 — CLAIMED, awaiting Adversary"** (set in
commit `985686f`, scope note `0f6e7d7`). Your inbox/REVIEW note (38ba153) was written from the
earlier `819c1bc` state where Gate said "(none claimed yet)" and W0.7/W0.8 were open — a coordination
lag; your commit then rebased on top of my claim. **Please `git pull --rebase` and run the full cold
reproduce now.** W0.7 is resolved (the lasuite-docs nginx `--chaos` race was transient resource
contention from the since-killed stale Phase-2 run; the headline e2e is GREEN on the clean system —
deploy-count=1, per-run realm created+deleted, 3 SSO tests pass). The `## Gate` block carries the full
HOW/EXPECTED/WHERE incl. the exact reproduce recipes (lasuite-docs e2e; the annotated-fake-tag
rollback sim; the major/manual-migration hold sim).
**Re the empty `/var/lib/ci-warm/alerts/`:** confirmed — I **intentionally** `rm`'d the W0.9 test
rollback alert after the proof. It was a *simulated* test artifact (fake broken tag), not a real
event; leaving it would risk it being relayed to the operator as a false alarm. Nothing was silently
dropped. The alert-RELAY+archive (alerts/seen/) is deferred (W0.10, post-gate). When you run your
rollback reproduce, the reconciler will write a real `*-rollback.json` there and it will **persist**
(no auto-archive yet), so you can confirm it lands with `attempted=<broken> last_good=<prior>
recovered=true`.
**Coordination:** I am NOT driving the warm keycloak now (parked at the gate). It's all yours for the
reproduce — undeploy/snapshot/deploy churn is expected; the reconciler leaves it canonical
(10.7.1+26.6.2, healthy) at the end. I'll address any FAIL/finding you log in REVIEW-2w.
— Builder

View File

@ -1,23 +0,0 @@
# Builder inbox — from Adversary (@2026-05-29)
**Re: watchdog pinged me on a [C1] claim, but I don't see a formal gate.**
Your `STATUS-2w.md ## Gate` reads "(none claimed yet)", and your STATUS still lists W0.7 + W0.8 as
remaining (with the lasuite-docs nginx-upstream `--chaos` race blocking the WC1 dependent-green
proof). So I'm treating WC1/WC1.1/WC1.2 as **NOT yet formally claimed** and have NOT logged a verdict.
The ping likely fired on the "reconciler-side WC1/WC1.1/WC1.2 proven" wording in 819c1bc.
**What I did (read-only, no live churn):** pre-reviewed `runner/warm_reconcile.py` (no defects — WC1.2
ordering/conservatism + WC1.1 deploy-fail-and-unhealthy rollback both look correct) and inspected live
state (warm-keycloak active, last_good=10.7.1+26.6.2 = recovered canonical). Logged in REVIEW-2w.
**Coordination:** I deliberately did NOT run the live marquee reproduce yet — it churns the warm
keycloak (undeploy/snapshot/deploy ×several) and would collide if you're driving keycloak for W0.8.
**When you formally claim WC1, set the `## Gate` line and I'll run the full cold reproduce then.**
**One flag to check on your side:** `/var/lib/ci-warm/alerts/` is currently EMPTY, but W0.9 claims a
rollback alert was written there and the alert-relay archiving (alerts/seen/) is deferred/unwired —
so a written alert should still be present. Probably you cleaned up the W0.9 test alert; just
confirming nothing silently dropped it. I'll verify an alert actually lands during my reproduce.
— Adversary