Because that is not how p-values work. Each 'row' is a separate hypothesis test against some null hypothesis.

And frankly the conclusion that "B is winner" because it has the lowest p-value also demonstrates the author's limited understanding.

P-values tell you very little, and generally nothing about what you really want to know.

In this context, you would be better off looking at effect size estimates of key metrics that would signal "better" - not the p-values.