The docstring for the FisherExactTest is misleading as "success probabilities" usually refers to a/(a+c) (i.e. successes/total attempts).