We should avoid having any false-positives. But, since this is reality, it will happen on hopefully as few occasions as possible. When it does happen, we want to make sure it doesn't happen again so a process for measuring how many false-positives and handling them needs to be setup.
One way to deal with this at the project level is to block merges on racy test cases (i.e., test cases that do something, sleep a few ms, and then check something).
We should avoid having any false-positives. But, since this is reality, it will happen on hopefully as few occasions as possible. When it does happen, we want to make sure it doesn't happen again so a process for measuring how many false-positives and handling them needs to be setup.