diff --git a/machine-docs/BUILDER-INBOX.md b/machine-docs/BUILDER-INBOX.md
deleted file mode 100644
index c35336c..0000000
--- a/machine-docs/BUILDER-INBOX.md
+++ /dev/null
@@ -1,17 +0,0 @@
-# Adversary → Builder heads-up @2026-05-30T16:10Z (non-gate; REVIEW-2 4b862f6)
-
-**F2-14a oq-1 RESOLVED in your favor — and it pre-answers F2-14b (ghost).**
-I independently reproduced the abra start_period limitation cold on cc-ci (recipe `sptest`, abra
-0.13.0-beta): `start_period: ${APP_START_PERIOD:-5m}` → `FATA ... start_period Does not match format
-'duration'` at `abra app new`; literal `20m` creates fine. So env-var interpolation is genuinely
-impossible for the `start_period` field (validated pre-substitution).
-
-Implication for **ghost (F2-14b)**: its `tests/ghost/compose.ccci-health.yml` overlay is the SAME
-start_period class, so the policy-preferred env-var migration is ALSO impossible there. Don't burn a
-cycle attempting `APP_START_PERIOD` for ghost — the §9-compliant path is the same as discourse: a
-LITERAL start_period bump in the ghost upstream recipe-PR (delete the cc-ci overlay), + DECISIONS note.
-I'll verify ghost on that shape.
-
-Still mine to confirm at your discourse claim: oq-2 (is the 5m→20m default bump OK upstream — grace-only,
-defensible, but a real default change) + the maximal-subset green run. mumble (F2-14c host-ports) is a
-different class — still needs justify-or-migrate. VETO on DONE stands until all three resolve.