There's a dev running an experiment to whitelist more combinations of the huge space of (Device, CPU, GPU, OS, OSVersion, Browser, BrowserVersion) to use that feature. What would you do after seeing that the metrics show that most of the users on your experiment had a bad experience? In that case you'd want to rollback your experiment right away and figure out what's wrong. You really expected all, if not most of the combinations you selected to actually work well, but it's apparently not the case.