For example, for an integer, they may reason about the range of an integer but be unable to reason about whether an integer is odd/even or a multiple of 10 or a prime number.

That’s wrong. You can implement a parity domain and perform computations on that domain along your interval domain. Great article otherwise :).

Edit: I think I misunderstood the author’s point. He’s not arguing that a parity domain can’t exist but that it has to be implemented - the abstract interpreter won’t magically make the parity domain emerge by itself. That is very true.