Currently, if a prioritized PR fails CI, and is retries because the CI failure is (possibly) due to a flake, Hoff forgets the merge priority.
It is possible to pass the priority as an argument to the retry command (@mergebot retry with priority), but that is easy to forget, especially in the high-stress situations where we need priority merges.
I think it would make sense to remember the priority of the merge when retrying a PR.