[Bug-gnubg] Re: with a Little Help From My Friends

[Philippe Michel]

On Mon, 27 Dec 2010, Wilson Amaral Jorge wrote:

> I need understand why there is a divergence so great between the numbers 
> showed in the (Analyse)Market Window and Evaluation/Hint. Better 
> explaining: in a particular game -2/-4, cube 2 with the trailer, the 
> Market Window showed Cash Point 50% and Too Good 82.495%.
>
> However, when the trailer was only with 61.4% of GWC chances, he asked a 
> Hint and it was Too Good. From 61.4% to 82.495% is a huge difference, 
> maybe I'm not understanding something.

What may be misleading here is the 61.4% number : it is a cubeless number, 
the percentage of games you would win *if you never redouble*.

In reality, you will redouble quite often, when the oppenent rolls 
something that decrease your gammons chances but you're still above the 
50% cash point.

For instance, suppose you roll 6-2 and he rolls 3-3. He will anchor up on 
the 18 or 21 point ; your gammon chances become much smaller but you can 
still cash and that's what you should do.

In the cubeless calculation that gave the 61.4%, this continuation is 
counted for something like 55% (your cubeless GWC) but in reality it is 
worth 100% (your GWC with cube access).

When every continuation is accounted for, it looks like you're indeed too 
good (according 3 or 4 ply evaluation, 0, 1 and 2 ply say double/pass). 
But you won't win 61% of the games with 17% gammons like the cubeless 
numbers suggest, it will be more like 90% games (almost all cashes) and 
relatively few gammons.

How gnubg estimates cubeful equity/GWC is described here, for instance :
http://www.gnubg.org/win32/gnubg/gnubg.html#Cubeful%20equities