Statistics you can defend
Holm-Bonferroni, Benjamini-Hochberg and Šidák
Three corrections controlling two different things. Which you want depends on how many variants you are running and what a false winner costs.
Holm-Bonferroni is the strictest of the three. It sorts the comparisons and works down from the tightest threshold, stopping at the first one that fails, and controls the chance of making any false claim at all. It is the right default when a false winner would be shipped and lived with.
Benjamini-Hochberg works from the other end, and controls the share of false positives among the results you call winners rather than the chance of any. That is a weaker guarantee and a more useful one when several variants are live: it is meaningfully less likely to throw away a real effect, at the cost of admitting the occasional bad one.
Šidák applies one threshold to every comparison instead of a sequence of them. It is slightly less strict than a plain Bonferroni and assumes the comparisons are independent, which is a reasonable assumption when your variants are genuinely different ideas and a poor one when they are variations on a theme.
None is also an option, and it is honest about what it costs: with several comparisons at an uncorrected threshold, the console states outright how likely at least one false winner becomes. That is a defensible choice on a single-variant test and rarely a defensible one otherwise.