We nearly published a 2x regression that didn't exist

We were about to ship a performance write-up claiming three queries had gotten 2-4x slower. All three...