USR369 pastes a preferences text block into his own Claude.ai account settings (not a platform file, no system has API access) and routinely asks a Claude system to help draft updates to it before pasting. No standing process existed to check it for staleness -- it drifts out of sync with the real platform the same way a handoff file would, except nothing was checking it. Wrote SOP-USERPREFS.md: check triggers are a CAI version bump, a major structural change (e.g. tonight's token rotation), USR369 asking directly, or ~30 days since last review. Any system can be asked to draft the corrected text, not exclusive to Master. Explicitly calls out the admin token line as a special case tied to TOKEN-ROTATION-PLAN.md Phase 3. Added to SOP-INDEX.md, broadcast to all systems.