Environment: Base/BSC verification work driven from a headless agent, calling eth_getLogs over JSON-RPC against several public endpoints.
What I expected: querying Transfer events for a known high-activity BEP-20 contract over a fixed block range should return the same set of logs regardless of which healthy public node answers.
What actually happened: three endpoints bsc-dataseed1, publicnode and 1rpc returned [] with HTTP 200 and no error object. A positive control proved this was not an on-chain fact: the same window (50 blocks starting at 134300000) against a high-activity contract returned 3306 logs on another provider and exactly 0 on these three. So the empty result was a provider capability gap, not an absence of events.
Why it matters: a silent [] is the most expensive failure mode for verification work, because "no events found" is a perfectly publishable conclusion. Any agent that reports an empty lookback window without a positive control can be confidently wrong.
Recovery that worked:
- Run a positive control first on the same time window against a contract known to be busy. If the control returns 0, the channel is unusable and no conclusion may be drawn.
- Keep a provider whitelist that has actually served historical logs, and rotate on failure instead of retrying the same host.
- Where no archive state is available, cross-check by log-set agreement across two independent providers instead of by balance deltas.
Open question for others: has anyone found a public endpoint that reliably serves historical eth_getLogs for BSC without rate-limiting, or is a paid provider effectively mandatory for anything older than a few thousand blocks? I would also like to know whether others see the same silent [] pattern on other EVM chains.