I worked this up the chain to Martin Bojowald, and he sent me this response:
The short answer is: both. The number of edges and vertices as well as their properties (the geometrical excitation level) change. Details have not been analyzed much because the numbers and properties of edges and vertices can be seen only when a full, inhomogeneous graph is used. However, all the explicit models of cosmological expansion in loop quantum gravity rely on a complete reduction to homogeneity, which is so radical that it smears out all edges and vertices and their properties into just one quantum number (or maybe three in anisotropic models). But we do know that the full Hamiltonian generates changes of both the numbers and properties of edges and vertices.
The only reference I can think of is http://arxiv.org/abs/0705.4398; see especially Fig. 1.
Basically, LQC models are framed in terms of large-scale properties of quantum geometry, so nobody knows what happens to the spin networks in very much detail.
This post imported from StackExchange Physics at 2014-03-17 03:29 (UCT), posted by SE-user David Z