Seems reasonable to do the sums in moderate modulo sizes, then for those that pass, give them to a larger modulo system, and only then check them "properly." A

Seems reasonable to do the sums in moderate modulo sizes, then for those that pass, give them to a larger modulo system, and only then check them "properly." Also seems resonable to reduce the work by precomputing many, many results for small modulo values. Hmm. Can easily speed up by a factor of 10 with some work, but that's not really interesting. Besides, I actually believe the conjecture.