What makes an IRP specific from other components, is the existence of a compatibility matrix between its users. This implies some form of deferred binding: at compile-time (i.e. when the config records used for labelling are produced), the various peers accessing it are not fully bound. This allows later to change the exact version of some peers without invalidating the others.
Ensuring the compatibility relies upon the expressiveness of the syntax used in the IRPs. SCM only allows to track the combinations actually used (e.g. during validation testing).
In other words, IRPs are contracts, not dependencies. They are based on semantics, they are anticipations of the behaviour.