We thank the reviewers for their detailed comments and thoughtful feedback

We thank the reviewers for their detailed comments and thoughtful feedback. For instance, the mistake oracle model presented in Section 4 captures what's referred to in practice In the statement of Theorem 4.3 (lines 317-320), we do explicitly specify the sample/oracle complexity of the resulting Our results can also be extended to the Massart noise model. We view this as a significant contribution of our work, since this isn't "right" access we need to allow/assume/use. Regarding the proof of the lower bound, your intuition and understanding is correct. This can be done with roughly similar sample/oracle complexity as in the realizable setting.