# To: Price Predict Feeds. From: Manager (Claude), for Ben, 8 Oct 2026

What happened today and what changed in your area (also in http://159.223.88.168/hx-c9ois/ops-2026-10-08.md).

## Outage and fixes
- 06:35 the paid dRPC balance ran out; crewlive, ponstape and signer used only paid dRPC and stopped for about 6.5 h. ~12:10 Alchemy hit its monthly cap and rhfast fell 17 min behind. Ben's rule: paid sources are a bonus; collection on new coins must never stop.
- `/opt/common/rpcpool.py`: one sender with fallback (paid if allowed, free dRPC, public node; 401/402/403/429 = skip that source 60 s). crewmap.py, ponstape.py, signer.py use it. rhfast has its own fallback list and backoff. crewmap live catches up in 5000-block steps. All gaps were refilled.
- 17:56 to 18:21 livewatch crash-looped because `alchemy` was removed from rpc_keys.json. Fixed: providers are optional, BILLING checks are info only (no Telegram), collector checks read the -wal file (WAL databases).

## Guards (Ben's system rule: redundant monitoring + redundant fixing by importance, Telegram only when every layer failed)
Added to the guard mesh on main (`/opt/guard/checks.json`): S rhfast (heartbeat fast.ok 90 s), S livewatch (30 s), M crewlive and ponstape (-wal 10 min), M price-predict-rh (status.json 5 min), M runtrack, rhledger, klines. Layers per service: systemd restart, livewatch (5 s), livewatch2 (20 s timer), guard (30 s), remote mesh from 3 other boxes (1 min), Mac backup check.

## Asks
1. Any collector you add: use rpcpool (never a single provider), write a heartbeat file, and tell me its name so it gets a guard check.
2. Keep every new collector off the live box (firstmin) and keep research off main.