[Cryptography] ideas for (long) Nothing up my sleeve numbers

[tpb-crypto at laposte.net]

> Message du 02/04/14 22:54
> De : "Natanael" 
> 
> In case the numbers in Pi would be shown to have an exploitable structure
> for your use case, you can switch constant. In case it would still be
> unique enough and you have a trusted hash function, you can simply just use
> that hash on the output.

How about this. You open ten threads to compute ten constants (square root of 2, pi, e, ...), these threads are opened in random order (ask the system a random number and use the last digit to choose which thread to fire up each iteration).

Let them compute for say 5 seconds, then ask the system random number for one sample, take the last digit of that sample and stop the thread that corresponds to that digit.

Now you use the last digit of this first calculated constant to stop the next thread, and on, and on, until you have finished stopping all threads.

Then you take digits of the calculated constants of your choice near the end of each constant (size and position could be randomly selected too) and mix them together to make a seed for a good pseudo-random routine.

Once it is fed into the pseudo-random routine it is thrown away so it can't be found in the system there after.

You can store the computations done by the threads, so the next time you call the constant generation routines they will continue from where they have stopped.

Does it sound like some good routine for random number generation, at least not depending so much in the outside world, except the order the threads are created and the stopping of the first thread?

Suggestions about which pseudo-random algorithms could be used with it? Couldn't we start one pseudo-random routine and then feed it into another?