When we split failed jobs, the job with the highest ID will no longer be
the one with the highest max_value