Reproducible builds make diagnosing performance regressions less horrible, because they are less likely to have been caused by random shuffling of code, and any

Reproducible builds make diagnosing performance regressions less horrible, because they are less likely to have been caused by random shuffling of code, and any random shuffling is reproducible when trying to bisect.