If political polling data gives you information about the outcome of an election -- if it can make your probability estimate more accurate than that of someone

If political polling data gives you information about the outcome of an election -- if it can make your probability estimate more accurate than that of someone who hasn't seen any such data -- then I don't see what's so all-important about independent testability. Evidence is evidence; the possibility that a poll is skewed merely weakens its evidentiary value, by increasing the probability that the poll would have those results if they're not true.