Two Types of Passwords

I am tempted to say I "like" this, but danger lurks in amateurs discussing aspects of Cryptography. I would want to see a discussion led by people in the field, about the merits of this approach.

It looks like the author avoids sharing even super-encrypted state outside of the locally strong-store box, and has made an architectural decision to segment passwords/phrases into two: the ones which get onto the strong box you hold, and the ones you send over the wire to login to systems.

But, what worries me is the implied uniqueness of the derived strings: if a family of things share xxxx-yyy-zzz-<unique> then the functionally unique input to the PRNG or whatever mix he used to generate the string, is the length of <unique> and there is a large component of input which is shared with other things: In my weak understanding of crypto, this is traditionally held to be a weakness: the apparent length of uniqueness is not as long as it looks, and loss of even one of these tokens may reveal facts about the seed data.

TL;DR want to see competent cryptography on the products of this scheme and the impact of the <common-prefix> <unique> inputs to the produced passwords.