This is proving to be an interesting read, but I have much different design goals in mind than the author. I'm interested in simulation, and this sort of RNG is

This is proving to be an interesting read, but I have much different design goals in mind than the author. I'm interested in simulation, and this sort of RNG is useless for that purpose. I would go as far to say this approach is useless for any scientific purpose. With great state comes great reproducibility. (I couldn't resist!) In the abstract the author suggests that PRNGs rely on "on unprovable assumptions about “one-way functions”" Any PRNG worth using should have solid number theoretic footing. For example, Multiply with Carry generators are based on the decimal sequence produced by dividing integer by a very large prime.