Young's model illustrates one of the supposed virtues of production systems. They are modular. That is, since all rules are syntactically independent, you can easily add new ones without having to re-write old ones. Because of how the recognise-resolve-act cycle works, it will call these new rules automatically if their conditions match STM.
Contrast to (e.g.) improving a Pascal program. If you want to add a new procedure (or predicate definition), you can do so easily. But the program will not call it just because it's there. You must also insert into the existing old code, statements which call the new procedure (or predicate). In production systems, that isn't necessary.
Note that this is a somewhat different use of the word from e.g. cognitive modularity.
In passing, I should say that Matt Elton, in his 3rd year project, appears to have found that Young's seriation system is not as modular as he claims. He claims that if you analyse the stage 1 rules, you find they have more than the minimum number of variables etc needed to achieve stage 1 behaviour. The extra parts are ``hooks'' which are necessary so that the additional rules will interface smoothly with stage 1 to give stage 2 and stage 3 behaviours.