>> Sorry, but it's horrible. Let me write the formula you use here:
How so? As in "horribly simple"? My main goal was to make an algorithm that had predictable, generic properties that worked on any number of bits. I believe I've achieved that.
>> Given a number, the next number can be calculated by simply filling in the formula if you know how many numbers have been generated already. Not only that, but it shouldn't be hard to find the seed either if you get the first number.
I think that falls under the second category of "disadvantages" that I listed. Security was not my focus here at all. The types of flaws I am concerned with are lapses in periodicity, bad distribution, and evidence that it is not in fact "bit width neutral".
>> If not, you need two to predict the rest.
Are you sure? Can you guess what follows these two numbers?
4290416596
806034681
>> But, I don't think a potential flaw found be found in the screenies (unless it's really bad).
Well, yes and no. Plotting the values is on the one hand a suprisingly effective tool for evaluating the basic properties of a generator, since our eyes are remarkably good at recognizing patterns (or the lack thereof). On the other hand, there are certainly subtle mathematical relationships that would be missed using this "method" and a more rigorous analysis would be necessary to find them.
>> Let's see what NIST thinks about it
Good idea. I had gotten a hold of one of their programs recently only to find that it had a very primitive interface and was difficult to work with, but the package you linked to looks like it might be a newer one, so I think I'll give it a shot.
