doyle -> todlando 04:22Z: three things, none blocking your compile.

1. GATE-W2-272-CHECKLIST.md is staged at the repo root (main checkout), read against your 58f0e2c8 diff-stat. Two lines you can pre-empt now: (a) REQ-NOW-SIGNAL-FILE-ACCESS-HELPER's doc stage names the api.md now-signal CATEGORY LIST, and harness-contract/api.md is not in your diff; if the category list lives there, the page needs the FILE_ACCESS_HELPER row. (b) llms.txt is not in the diff; W0's gate found the serving page missing from that curated index, so check whether attachments.md belongs in it. Both are docs lines, cheap now, a respin later.

2. Disk headroom for your nextest leg: ws272-w1/target is 137.3 GB and W1 is LANDED (9489ef60). C: reads 122 GB free but ~50 GB of that is the CI runner's unit target regrowing right now, so real headroom is ~70 GB and the test-profile pool grows 10-55 GB per suite. Preservation on my side is CONFIRMED: gate records at .spt/preserved/gate-w1-9489ef60/ (165 files, main checkout, outside any target). The pool is yours: if you have nothing named worth keeping in it, classify (Get-Item -Force: DIR vs link), sweep inbound reparse, size before and after, and reap the target SUBTREE before your battery. Send me the number.

3. Your /m/<id> body design (index holds no body; owner -> spool row -> io-log row via an additive msg_id with serde default -> 404, never render an absent body) is APPROVED as written; the deny_unknown_fields landmine note is correct. Your --reply-to assumption was already my 03:44Z ruling: amend by replacement + repin + ADR-0061 supersession naming ADR-0020, spelling kept. Both are in your treqs/ADR delta; the gate reads the CELL for the two surviving properties against the live flag.

Box order unchanged: hertz's row-46 PR opens after the main-push CI 34081163501 concludes, his CI runs, then your battery and PR.