[Bug-gnubg] quasi-random dice for the Initial Position

[Robert-Jan Veldhuizen]

Hi there, I also posted this on GOL.

I've understood from Jim Segrave that quasi-random dice just discards 
doubles for the opening rolls, effectively rotating 30 rolls.

However, I still have a question about this. Normally when I do a rollout, 
I would get one complete rotation for the first roll when I set 36 trials.

Suppose I use 36 trials for the opening roll. Will it now try to rotate the 
opening roll, find out after 30 trials that one rotation is done and start 
another rotation for the remaining 6 trials that it can't complete? Or will 
it use random dice for the remaining 6 trials?

Now, the same question for using 30 trials for the opening roll. Will this 
in practice produce exactly one rotation of non-double rolls? Or will gnubg 
think 30 trials is not enough for a complete rotation and just use random 
dice then?

More realistically, what exactly happens when I use 1080 trials from the 
initial position? Will this rollout recognize it can do a complete rotation 
of the first two rolls (30*36)? Or will it think less than 1296 trials is 
not enough for rotation of both rolls and just rotate the first roll?

What happens when one use 1296 trials? Will this give potentially skewed 
results perhaps?

Any comments welcome, especially from the gnubg team. It's hard to find out 
what gnubg does just by running tests here. Joern Thyssen's earlier 
explanation doesn't mention rotation with 30 rolls instead of 36, so I have 
some worries about what GNUBG exactly does here, and whether it's consistent.

TIA,

--
Robert-Jan Veldhuizen