Moving sink arming inside the cron ingest race invalidated three comments that now asserted the opposite of the code (this ships to a public upstream): - cron/+server.ts: dropped the claim that racing arming+ingest keeps "the whole tick" under the interval. The race bounds only the raced section (arming + ingest); the tick also spans the un-raced bot/notify/drip stages. State that the bound guards the search-endpoint hazard, not the tick's total duration. - meili-sink.ts (SETTINGS_TIMEOUT_MS doc): the arming fetch no longer runs "before the ingest race" — it runs inside it. Describe the current topology: the bound makes arming self-terminate within the shared budget, and also backs the xrpc arming path (outside any race). Also correct the false "upsert/remove only run inside contrail.ingest under the cron race": upsert is also reached from the geocode drip (after the race) and xrpc notify re-ingests (no race), so those write paths stay unbounded — a pre-existing exposure. - meili-sink.ts (applyMeiliSettings doc): "the arming path runs outside the cron ingest race" is now false for cron; distinguish cron (inside the race) from xrpc (outside any race). Comments only; no logic change. check: 0 errors / 7 pre-existing warnings; test: 283 passed / 4 skipped (unchanged).