[lstephen666]

Thanks, I hadn't seen the paradise one before. I had looked through the party and stars pages.

[ QUOTE ]
At Paradise Poker, we chose to use a 31 bit random function (coming from a 2016 bit seed that has entropy constantly added to it - see the random number generation information below for more details), thereby decreasing the bias even further (by a factor of 65536), as well as shuffling the deck more than once. We found that shuffling twice removes 99% of the bias, and shuffling 3 times makes it pretty much immeasurable, then we decided to triple it for good measure and bump it up to 10 times to keep the marketing people happy. These two changes result in a shuffled deck with no detectable bias.

[/ QUOTE ]

This paragraph worries me a bit. I really don't like how they just decided to "triple it and bump it up 10 times". Having done a bit of computer security I know that implementing the wrong type of security 30 times, is no better than doing it once (i.e., fixing one security hole a bad way 30 times, is no better than just fixing it the right way once (esp., if there are still other holes)). It would be good if they provided a link to some mathematical proof that this is actually the case.

Also, it still seems like there would be some bias here (although insignificant and undetectable, I don't think it's zero), when it would be easy enough to get zero bias (using their 0-63 number generator example, how about just ignoring numbers 52 or over?).

It's good to see some inputs like network delay in their RNG.

So, I like their detail, but I think a lot of it is unnecessary 'show' for the user.