But that doesn't mean this couldn't be done properly.
There are actual zero-knowledge proofs. I liked this primer. You should read it!
http://blog.cryptographyengineering.com/2014/11/zero-knowled...
It is 100% possible for there to exist absolutely zero-knowledge proof in many instances. (Such as the one in the article.)
So, it could be possible for example (I don't know an algorithm) to check whether a phone number you were given is in someone else's set of phone numbers - without either your learning what the other's set is, or the other learning what phone number you're testing.
You are right that the hash-passing is not zero-knowledge. Rather than just verify whether it's in a set, you've accidentally revealed a number. So if you have a set of phone numbers and I have 1 to check if it's in that set, then checking hashes doesn't work - I would end up divulging my number if I used it.
so the numbers are off, but they're not five orders of magnitude off. If we're going to service ten million users, we might have something of a hosting budget for it.
So while I believe the author that the current trade-offs aren't goods, clever academic mathematics might help in the future. Cryptography itself came from there.
This just isn't possible. How many cents would it cost to brute-force hash all legal American phone numbers?