I always use the exact binomial tests because calculating them is cheap. It is well-known that the normal approximation fails when the number of samples (n) or the probability of success (p) is low (my rule of thumb is n × p ≥ 7)—with qualifications. I document the practical implication in...