TL;DR: They're assuming that spacetime is not quantized, but is as GR describes it. This produces a consistent theory that predicts that measured mass will fluctuate very slightly.
A wave in the ocean doesnt exist any more than any other part of the water that isnt in the wave. We just focus on the wave because our brains are good at seeing moving shapes for hunting.
The substrate graph is updated by travelling rule applicators which dont require energy. Theyre like the updating edge of stacking faults between planes in crystalography.
Gravity slowing down light is just due to the travel speed of rule applicators being fixed. Higher graph density but fixed node updates per time means it takes longer to update a given volume.
Until then we are stuck at a funky high level abstraction. The table of quarks and particles we have is like what would happen if we were trying to figure out what our computers were made of by interacting with a url bar, and coming to the conclusion that the fundamental components of a computer are like... a table of HTTP response status codes.
Not to say that its useless, just that its not zoomed in enough.
The rule applicator just traverses the graph and updates a node based on adjacent nodes.
Time for us is just the difference in relative sequence lengths between updates applied to volumes at our scale.
I've always thought entanglement makes more intuitive sense on a graph substrate of some kind, where the "spooky action at a distance" is actually just an anomalous edge in the graph connecting two vertices which would otherwise be more distantly separated. This begs the question of how the edges in the graph came to be or whether they're modifiable at all though.
As for weird behaviour like far reaching spookiness. Could be multiple overlapping substrates at different scales.
Or, the rule directly includes a mechanism for remote information transmission.