When using error handling in combination with asynch work items as we do for a failed publication the process did not get persisted correctly. So the persisted process was still pointing to the publication work item, whilst we were already dealing with the error and updating the task in Pulse.

As a result shutting down the system with a failed publication in Pulse would result in completed processes never being correctly removed from JCR. And probably some other weirdness as the workItem was not completed correctly either.