An amplitude-amplification procedure starts with a state whose marked subspace has success probability p. Ignoring implementation details and rounding, how does the number of iterations scale when p is small?