Also: if we find that neglecting the shell model makes no big difference, then we can hope to be able to simulate a bigger system and so we can hope to solve the reproducibility problem we had in 19990831/19990902/19990909 (assuming that the problem is due to an insufficient statistics...)