I'm wondering if you've noticed an uptick in issues with API nodes over the last few weeks.

Yes, I've also noticed that. I need to improve the logic for handling node errors in the coming days, so that automatically switching to other nodes only happens after a few request retries.

The cause for the duplicate account operations was the different field orders on different nodes (I'm building a hash sum from the operation's data to have a unique id for each operation. When receiving an operation from an other node with different order, the hash changed and the already known operation was seen as a new one).

I've changed it today, so that the fields are being sorted before building the hash. Hopefully we will no longer see any duplicates from now on.