Dan Taflin
The "run bits" test as presented here is far too simplistic to be of any use. All of the "skews" between left and right halves are well within a single standard deviation for a binomial distribution. That, and the fact that the skews were toward different halves in the different tests seems to more strongly support the null hypothesis of no skew than anything else.
On the other hand, the cascade rate test result does concern me. I have routinely assumed that even a small change in input text should result in a radical change in hash output. We'll have to be much more careful to salt every hash input to avoid difficulties.
