diff --git a/docs/guides/matrix-intake.md b/docs/guides/matrix-intake.md
index 895f273..d5f5c7a 100644
--- a/docs/guides/matrix-intake.md
+++ b/docs/guides/matrix-intake.md
@@ -87,8 +87,14 @@ Spark). See *Fuzzy matching* below. Tests green (27/27 backend + the offline bot
    clears the room's pre-existing backlog. The intake room keeps **no durable pending state** (the
    proposal store is in-memory, lost on restart), so every message in it is stale after a restart and
    safe to redact — it clears **every** `m.room.message` (text + card images), bot's and humans'
-   alike. Run on the Spark: `docker compose run --rm matrix-intake python -u
-   backend/matrix_intake/redact_intake.py [--apply]`.
+   alike. Run on the Spark (the compose **service** is `intake`, not the container name
+   `matrix-intake`): `docker compose run --rm intake python -u
+   backend/matrix_intake/redact_intake.py [--apply]`. **Redaction is rate-limited** by the
+   homeserver (~4s `429` backoff each — matrix-nio sleeps through it), so a large room takes minutes
+   and a single SSH run can time out partway; the script is **idempotent** (already-redacted events
+   have empty content → skipped), so just re-run `--apply` until a dry-run reports `0`. First live
+   purge (2026-06-20) cleared a 104-message room over two passes; the bot already held the room's
+   redact/mod power (users' notes + a card image cleared with **0 failures**).
 
 ## Business-card capture (M3 — image intake)