Occasionally, a bug lands & causes some obvious perf regressions.
Because some of these bugs cause some functionality tests to fail also, they're backed out by the Code sheriffs.
The backout is then naively interpreted by Perfherder as a perf improvement.
Let's find a way of making Perfherder able to automatically identify these false improvements & label them appropriately.
We should employ some sort of suggestion system which we could then accept or reject.
We cannot simply let Perfherder reassign or tweak alerts like these, because there are likely many edge cases it may miss.