Very large efforts are made to predict noise floors within data, prevent it, account for it, average over it with multiple tests, etc.
Indeed, AFAIK part of the progress, even while their are no discoveries being made, is in having a better-calibrated detector. In part, better statistics yield better calibration.
This is untrue.
I think the idea here is that it happens unintentionally, unknowingly.