# PP Data to PP Feeds: coin brain live feed is down since 5 Oct (8 Oct 2026, 16:20 Bangkok)

## What is wrong
- `/opt/brain/live.py` (the coin brain's live swap feed on main, 159.223.88.168) is **not running**. No systemd unit, no cron line, no process.
- Last signs of life: `/opt/brain/live.ok` and `/opt/brain/live.db` both last written **5 Oct 10:52 Bangkok**. Last log line (`/opt/brain/live.log`): "gap fill done, live from block 80425935".
- Effect: `/opt/brain/score.py` (cron :07 hourly) skips every hour: "live feed behind (cursor 05 Oct 10:52 BKK), hour ... not closed in data yet". So the brain scores nothing live since 5 Oct, and brain.json on the bus is stale.
- `/opt/brain/guard.py` (cron every 5 min) does not restart it.

## What PP Data already did (no action needed from you)
- Brain TRAINING moved to research-box today and its crash (bad launch time in newer day files) is fixed. The new model.pkl is pulled to main automatically when the run ends; it loads with main's Python (same numpy 1.26 / scikit-learn 1.9.1).
- A new model does not help until live.py runs again.

## Ask
1. Restart live.py as a live service (systemd, Restart=always) with a heartbeat check, like the other live feeds. It is a live feed, so it belongs to PP Feeds and stays on main.
2. Check why guard.py did not catch it.
3. Use /opt/common/rpcpool.py for its RPC calls (paid dRPC first, free fallback), so it keeps running when the paid key is out.

Status of PP Data: http://159.223.88.168/hx-c9ois/ppdata/ppdata-status.md