🚀 go-pugleaf

In article <7V2qXB1w165w@tornado.welly.gen.nz>, sai@tornado.welly.gen.nz (Simon McAuliffe) writes:

....

> This is kinda frightening...  Every single test failed miserably,
> supporting the alternative hypothesis.  Even if I tried my best I
> couldn't fiddle these tests.  These tests simply cannot be passed at any
> significance level.
> 
> It looks like _THIS_IS_ what Dr. Sidelnikov also did, and it means one
> of three things:  (ranked in order of likelihood as I see it)
> 
>    1)  Both Dr. Sidelnikov and I stuffed up.
> 
>    2)  This correlation test isn't appropriate for this situation.
> 
>    3)  Something funny is going on and either the RNG is badly
>        implemented or IDEA has a problem.
> 
> Number 1 sounds like the most likely and number 3 is just plain scary.
> 
> Well, after all this, perhaps Dr. Sidelnikov isn't just a crock. If I
> didn't make any mistakes, there may be something to what he says after
> all.  Can somebody who really knows some real statistics shed a brighter
> light on the subject?

An interesting test.  A useful control would be to plug in DES instead
of IDEA, and see what happens.  It's pretty well accepted that DES is
a very good random number generator; if it fails your test, I'd look
hard at cases 1 and 2.  If DES passes, on the other hand, case 3 looks
plausible.