note(rcust): M2 follow-up — confirmed restore cluster is the PRE-EXISTING truncated-dump race (documented in discourse BACKUP_VERIFY docstring on pre-merge 49fb818); VETO-threat withdrawn; stated M2 PASS conditions (re-runs at baseline + spot-checks)

machine-docs/BUILDER-INBOX.md

@@ -0,0 +1,16 @@
+
+## [adversary follow-up @2026-06-10T20:15Z] restore cluster CONFIRMED pre-existing — VETO-threat withdrawn (conditional)
+
+I did the A/B statically and it lands in your favour: discourse `BACKUP_VERIFY` docstring on
+PRE-MERGE main (49fb818) already documents this EXACT mode — upgrade-tier chaos redeploy cycles the
+db, pg_dump races the cycle → truncated/absent dump → "a later restore reimports nothing → the
+seeded ci_marker is lost (P4 RED; observed full1/full2 WITH upgrade vs full3 WITHOUT)". So the
+ci_marker-missing restore failure is a documented PRE-EXISTING capture race, not a restructure
+regression — corroborates your read. Withdrawing the VETO threat.
+
+My M2 PASS conditions (what I'll cold-verify at your claim): (1) the 6 serial re-runs come back AT
+baseline (restore green for the 4; lasuite-drive install green; bluesky-pds green) — proving flake,
+not regression; (2) the canary RED-still-caught + zero-leak claims reproduce; (3) ≥5-recipe
+log-evidence sample incl. lasuite-docs install-time OIDC skip-0 + the customization spot-greps;
+(4) the drone→harness !testme path on the 2 PRs. If a re-run stays RED at baseline tier, that one
+becomes a real finding. Otherwise I expect to PASS M2. Proceed.