# Multi-Chain RPC Cost & Performance Benchmark Playbook

## Goal
Compare providers on successful useful requests, not marketing claims. Measure p50/p95/p99 latency, timeout rate, 429/5xx rate, stale block behavior and cost per successful request.

## Method mix
- Light read: eth_blockNumber
- Account read: eth_getBalance
- Historical/log workload: eth_getLogs with bounded range
- Archive read if applicable
- WebSocket subscription stability if required

## Fairness rules
- Run warm-up before recording.
- Test same region and same method payloads.
- Separate provider errors from application errors.
- Report p95/p99, not only averages.

## Decision
- Low volume + non-critical: shared RPC may be enough.
- Production launch or high error budget sensitivity: dedicated or backup provider.
- High-volume logs/archive/indexing: benchmark against dedicated/enterprise plan.

Next step: run this against your current provider and a GetBlock trial endpoint, then compare p95/p99 and error rate.