Statistics you can defend
Multiple-comparison correction
Test four variants at 0.05 each and you have roughly a one-in-five chance of a false winner. A correction is what stops that.
Every additional variant is another chance for noise to look like a winner. With four comparisons judged at 0.05 apiece, the chance that at least one crosses by luck is close to twenty per cent, and the winner you ship is as likely to be an artefact as an improvement.
A correction tightens the bar to compensate. Pertento offers three, plus none, and applying one changes only the significant verdict. The p-value, the z-score and the intervals are untouched, so switching methods never changes the evidence, only the standard you are holding it to.
With a single variant all four reduce to the same threshold. Rather than leaving you to discover that by testing each option, the console says so directly, and states the actual threshold the chosen method is applying across your number of comparisons.
One correction is pinned rather than chosen wherever the platform acts on its own behalf. The attention queue, the implementation queue and the automated stopping rule all read the same method, because a queue cannot offer a dropdown and two screens that disagree about who won is the bug this avoids.