We experienced an inconsistency across RPC nodes in which queries will return response, versus which will return errors. This is due to distributed architecture, but we don't know why so many are running non-standard versions that don't support the whole range of queries. We tested each node and identified which will work for us and which will not. Unfortunately, the nodes that support all of our queries have higher load and slower response times. This is a minimal impact for us, but it does increase the risk of being throttled in the future.