Why is Iterative-deepening A* optimal, even without monotonicity?

Question

Why is it that Iterative-deepening A* is optimal, even without monotonicity? How can I be sure that the first goal reached is the optimal one?