Let’s say the probability of a recount in state 1 changing the result of the initially tabulated vote in favor of losing candidate B is 0.2% (I made the 0.2% figure up just for illustration). But B needs to have recounts in four states do the same thing in order to affect the outcome of the election, and to keep it simple, let’s assume the margins that he’s behind are the same in all four states. So in each state, the probability of that happening is 0.2%.
But the probability of that happening for candidate B in four states is NOT 0.2% — indeed, far from it. It is 0.2%x0.2%x0.2%x0.2%. That means it is a very small number, 0.0016%.
This is known as the principle of conjunctive probabilities, first discovered by Fermat.
This assumes the four state counting processes and any errors in them revealed in a recount are independent of each other.