When the situation described in SKARA-1616 occurs, we enter an infinite series of retries for a PR, because once a user is inactive, there is no reason to believe it will ever stop. OTOH, we can't be sure that a single failure actually means that the user is inactive and the PR essentially dead. I think it would be worth having a mechanism that could keep count on how many times a PR has failed to be evaluated by a WorkItem and start to throttle the retries. This could simply be handled with a specialized exception type (subclass of RuntimeException) which we can check for in the error handling method of WorkItem and call the quarantine method of the poller. It would be nice if we could design a more generalized handling of this though, maybe as part of the poller itself, so we don't need to repeat a lot of the implementation in different bots.