Do you think that the technique of using the last 4 bits to choose a further pseudorandom 31 bits from the rest of the hash MIGHT mitigate SOME future weakness as a PRNG? Or do you have confidence it is completely useless? Or neither, of course.