Re: [Bug-gnubg] Why does it move first about 80% of the time UPDATE

[Russ Allbery]

"Wayne" <address@hidden> writes:

> I am still finding the distribution is not 50:50 on the initial roll,
> but have observed a significant change when I set the dice option to
> www.random.org.

You didn't answer Ralph's question: over how many trials?

This observation is meaningless without knowing how many samples you've
taken.  Pure random numbers generate long streaks far more frequently than
human intuition thinks they will.  (This is an old trick for
distinguishing human-generated random coin tosses from truly random coin
tosses.  Human-generated strings of coin tosses will have fewer and
shorter streaks than they "should" if they were random.)

> I note something from my systems engineering: Nancy Leveson at MIT
> states it is impossible to verify that software provides a safe outcome.

This is certainly true in general, but random number generators are
mathematical formulas and are, next to cryptography, one of the most
intensively studied and verified computer algorithms in existence.

GNU Backgammon uses very standard and well-studied techniques for random
number generation.  The chances that you've found some bias is small; the
likelihood is much higher that you've not taken an adequate number of
samples and the observation will disappear if you take enough samples.

-- 
Russ Allbery (address@hidden)              <http://www.eyrie.org/~eagle/>