Asymptotically Optimal Adversarial Strategies for the Probability Estimation Framework