Re: [Bug-gnubg] Odd rollout results of classic reference position

[Joern Thyssen]

On Mon, May 19, 2003 at 03:06:56PM +0100, Ian Shaw wrote
> > 
> > I agree that it seems like X should be able to cash more 
> > games later by retaining cube access.  I've seen this show up with GNUBg in 
> > other positions
> > as well.
> 
> I'm concerned that there's a real problem here. Has anyone got a
> convincing explanation as to how one can win more games by
> relinquishing the cube?

Hmm, I think I wrote something about this a long time ago...

Anyway, the primary output of a cubeful rollout is the cubeful equity --
no more, no less.

The cubeless percentages and cubeless equity are calculated anyway,
because most users like to see them, but note that the rollout is
truncated at "double, drop"-decisions in the live cube rollout, thus the
percentages are not really "true" cubeless ones, as the rollouts is
truncated. gnubg uses the cubeless percentages calculated from the
double, drop decision -- in your case a 2-ply decision.

However, note that the "no double" branch is often truncated after very
few rolls if you're well into the market window, and as you indicate
yourself, gnubg doesn't understand the position that well, so the
cubeless percentages may be garbage when the rollout is truncated
after,say, 3 rolls. The "take" branch will often produce better
percentages as the rollout will proceed for more plies (and often even
played to the bitter end).

This is the main reason for the difference in this case. For positions
with gammons (and non-zero gammon price) there may also be differences due
to chequerplay according to score. The leader and trailer may play very
different at, say, -2,-4, in the "no double" and in the "take" branch
leading to very different win and gammon percentages.  Of course, this
is not the case in your position as it's a gammonless race.

Jørn