I see, I was misreading the spec and was thinking it was more complex than it is. Even so, both for complexity and performance reasons it's best to keep relays very simple. A DVM would also allow you to introduce the new format with universal support, rather than only getting support from relay operators that opt-in.

And yeah, I see how other algorithms could actually compose with this. So other than the REQ update, I think this is quite good.