# #11 — Own Robinhood Chain node: study (one page, Ben decides)
_ppfeeds, 8 Oct 2026. Constraint: paid dRPC works but by Ben's decision only the AI picker uses paid; all other feeds run on free sources. This asks what an own node buys the FREE-source feeds._

## Free stack today
free dRPC (keyless) -> public RH node -> picker-box relay (public node via 2nd IP, VPC-only). AIMD pacing per public route; 429 widens gap, billing/403 parks source 60s. Free dRPC caps batch at 3, getLogs at 10k.

## Measured now (live, not estimated)
- v3live is AT the chain tip on the current mix: lag_blocks 0, pools 8235.
- BUT v3live's RPC calls are mostly PAID: paid 25943 vs public 975 + public_pb/relay 1810 + free-dRPC 452. This contradicts "only the picker uses paid." Could not confirm RPCPOOL_PAID env on v3live (research-box went unreachable mid-check). THIS IS THE KEY THING TO RESOLVE — it changes the whole study.
- Free-source friction real but survivable: ~130 logged 429s on v3live, clustered, most recent 13:04 UTC. Pacing absorbs them at cost of latency, not data loss.

## What an own node removes
- 10k-getLogs cap and missing debug_traceTransaction on public node (limits backfills/gap-fills and trace-based features; #24 kit-launch analysis is trace-adjacent).
- 429s / batch-of-3 ceiling -> own node sets own limits.
- Dependence on dRPC billing state + shared public node capacity.

## Cost / latency sketch
- Self-hosted RH-chain node: 1 DO droplet, likely 8-16 vCPU + NVMe, est $100-250/mo (archive = disk-heavy, expensive part). Hardware specifics NOT verified — follow-up before any number is trusted.
- Latency: LAN-local node removes RTT + rate-limit backoff; reliable sub-block-time — the main win for #24's <2s-after-block target.

## Recommendation (Ben decides)
1. FIRST resolve the v3live-paid discrepancy. If v3live is quietly on paid, the "free feeds struggle" premise is partly false and a node is less urgent. If it's meant to be free and isn't, that config fix is worth more than a node.
2. A node is most justified by #24's trace/launch-block needs + the 10k cap, NOT routine v3live/pretag keeping (free sources hold the tip today).
3. Before spend: confirm RH-chain real archive node specs + 1-week disk-growth estimate.

## Open (couldn't close — box unreachable)
RPCPOOL_PAID state on v3live; whether feeds hit the 10k getLogs wall in logs; verified node hardware reqs.