So I have not worked yet trough all the details of the theory and I hope this is not something resolved through the representation by multiway causal graph.
In the representations of states and events there is first a lot of redundancy, because the whole unaltered part of the graph ( or string) was copied after every alteration. Only after causal invariance is declared update events are described as something elementary which can be ordered differently, however if updates do not overlap in the nodes they interact with of course those updates order is independent.
So instead of looking at the problem like the constant generation of a new graph trough iterations, why not look at it as a layering of fitting pieces atop an existing structure? You could see it as a reinvention of spacetime, like when you represent a 1D cellulae automata your usually show its timeline and not a single state.
This representation might save on a lot of redundant, identical parts of a universe because the localization of events might express itself better.
The idea of avoiding reduce was explored by Wolfram here
He only considers global states of the systems, I also remember similar state graphs concerning cellular automata. They are useful to show rules of transition, but not for representing concrete examples.
I am talking more about local update events , for example the rule AB->B
if we start with ABAB then we can go BAB or ABB and finally arrive at BB
hovwever what we actually have there is AB->B|AB -> B. Only because we use different Foliations in the Causal Graph we gain a bunch of different global states. So all I am saying is, simple write those updates into the space graph instead and afterwards chose your slice to gain the foliation/timeslice of your choice.
It is also possible to see the multiway graph as fixing an already existing structure. For example, consider the graph of all positive integers and assume that there is an arrow from u to v if and only if v is a multiple of u and v is not equal to u. This is equivalent to begin with 1 and to generate all the multiples of each vertex each time.