Cranked up the error rate and gave in a couple of more runs, no issues. It was happening for around one in three recoveries before, so I'd say it's solid now.

Great. Took a while but we worked our way through that 🙂. Properly re-initializing on recovery is often a gotcha. Luckily the design of the task state machine means that there aren’t really any concurrency/race conditions to worry about so there weren’t all that many potential causes.

Nice. Thanks for your help on that.

Pushed a commit with a fix. I’ll release 0.13.1 in the next day or two, but you should be fine to use that snapshot for now.

👍