After further experimentations with the MultiwaySystem function, I have a hypothesis that might explain the observations I made: the function may not have a bug after all, but the book would be lacking critically important information about when the individual causal graphs should be isomorphic to each other.

Here are the facts:

1) Considering rule BA => AB and initial condition BBBAA:

-- For t = 4: Not all graph instances are isomorphic to each other, since there are 2 types of graphs. -- For t = 5: All 5 graph instances are isomorphic to each other and to those of Section 5.9. -- From t = 6: All graph instances are isomorphic to each other and to those for t = 5. -- Convergence of the evolution and causal graphs (i.e., their reaching a fixed point) only happens at t = 6.

3) Considering the same rule BA => AB, but now initial condition BBBAA:

-- Convergence of the evolution and causal graphs (i.e., their reaching a fixed point) only happens at t = 16. -- For t = 13: Not all graph instances are isomorphic to each other. -- For t = 14: All 24024 graph instances are isomorphic to each other. -- For t = 15: All 24024 graph instances are isomorphic to each other. -- For t = 16: All individual 24024 causal graph instances are isomorphic to the three ones shown in Section 5.9.

Conclusion:

It seems that isomorphism among the causal graph instances can only be guaranteed from the time step (and onwards) the evolution (and, therefore, its associated causal graph) has reached a fixed point. However, isomorphism may happen before the convergence.

So, the conditions by which isomorphism happens are not discussed in the sections related to causal graphs (especially Section 5.9).