When ON, every risk tier shifts ONE level more autonomous. That means:
| Action type | Auto OFF (now) | Auto FULL |
|---|---|---|
| FB ad budget +20% on winner | Telegram approval needed | Fires immediately, FYI only |
| FB ad pause when CPA > $25 | Telegram approval needed | Fires immediately, FYI only |
| V6 single-listing price push | Auto + FYI | Silent auto |
| Silence nudge DM to student | Telegram approval needed | Auto-sent via ManyChat if linked |
| Cart-close email drafts | Telegram approval needed | Drafted + FYI, still paste into Systeme |
| Cart-close price flip at T-0 | CRITICAL — double confirm | Still CRITICAL (unchanged) |
| New ad campaign creation | CRITICAL | Still CRITICAL (unchanged) |
| Any single action > $500 impact | CRITICAL | Still CRITICAL (unchanged) |
Edit the droplet .env and add:
Then: systemctl restart business-brain and Brain starts operating with one-tier-lower approval requirements across everything.
Remove the line OR set BRAIN_AUTO_MODE=off and restart. All approval gates return to default.
audit_log so nothing happens invisibly