I was going to argue and say that you shouldn't be overliteral. Obviously you can translate what he said to "insert some identifiable token". It doesn't matter what the sentinel token is, but selecting one does require a bit of knowledge of the input domain or the input set.

In general, though, I'll assume that you meant that in the spirit of saying that there's no a priori way to verify a sorted set that's faster than O(n), and that there's no sentinel token value that exists which would be useful for in all possible input domains using a posteriori validation.