Stale bare-name GUC reference in db/specs/appevents-pruner-design.md, deliberately excluded from a621cb6 to avoid sweeping another lane's uncommitted work

db/specs/appevents-pruner-design.md:90 describes the actor uuid as coming from the app.user_id GUC. terra_23 (applied 2026-08-16, a621cb6) renamed terra's archive-context GUCs to app.terraUserId / app.terraArchiveReason with a HARD cutover and no dual-accept window, so that sentence is false as of the apply. WHY IT IS NOT ALREADY FIXED, and this is a ratified deviation, not an oversight: coder-terra-cc was instructed to fold all three stale bare-name surfaces into a621cb6 and folded two (db/backup-db.sh:164, db/schema.md:49). They deliberately excluded this one and flagged it rather than burying it. The file holds ~87 uncommitted insertions from another lane, and gitpush --only is per-FILE, never per-hunk - including it would have swept another lane's in-progress work into a commit carrying db-terra-cc's Authored-By: and coder-terra-cc's message. That is the commons rule against committing another lane's stray files, applied correctly. pm-terra-cc RATIFIED the exclusion. SEVERITY, stated so nobody prices it above what it is: mildest of the three. It is descriptive prose about attribution provenance, not an operator instruction anyone executes. The dangerous one was db/backup-db.sh - restore-path guidance that would have made an operator conclude the tooling was broken - and that one shipped. WHAT TO DO: whoever owns the 87 uncommitted insertions lands them first, then corrects line ~90 to name app.terraUserId. Do not correct it before that edit lands - the same per-file collision returns. If the uncommitted work is abandoned rather than landed, say so and the fix becomes a trivial standalone push. WHY FILED AT ALL: the correction otherwise exists only as a closing paragraph in a status message, which is the form of deliverable that never happens. It is a documentation defect with a known cause, a known fix and a blocked-until condition, so it gets a row.