Reverse Entropy v1.0
A comment on my blog suggested, once the model has been running a bit, reverse the velocity vectors and see if the particles return to their initial conditions. Assuming the model is deterministic, it should.
This is a first attempt at trying that, but I didn't initially realize that the starting state involves considerable overlap of collision boundaries, which the "physics" of the model don't normally allow, so it's not possible under any condition to re-create that starting position.
Even so, it's clear the particles don't exactly backtrack, although they do return to close to the starting position. That almost surprised me.