Consider the following code:

std::vector<int> v1 = {1, 2, 3, 4};
std::vector<int> v2 = {1, 2, 3, 5};
auto result = std::mismatch(v1.begin(), v1.begin() + 2, v2.begin(), v2.begin() + 2);

The current wording of [mismatch] seems to require result to be {v1.begin() + 3, v2.begin() + 3}, because 3 is the smallest integer n such that *(v1.begin() + n) != *(v2.begin + n). In other words, if there's a mismatch that's reachable from first1 and first2, then std::mismatch must find and return it, even if it's beyond the end iterators passed by the user.

This is doubly unimplementable: the library has no way of knowing that it's safe to keep going past the end of the user-supplied range, and even if it could, doing so would violate the complexity requirements. More importantly, it would violate the fundamental convention that STL algorithms operate on user-supplied ranges, not on the underlying containers.