[Bug-gnubg] Evaluation inconsistency

[Albert Silver]

I was going over a match and asked GNU to analyze a cube decision at
3-ply just out of curiosity. Here is what it said:


Cube analysis
3-ply cubeless equity  +0.345 (Money:  +0.343)
   72.5%   0.0%   0.0% -  27.5%   9.7%   0.9%
Cubeful equities:
1. No double            +0.679
2. Double, pass         +1.000  ( +0.321)
3. Double, take         +0.671  ( -0.007)
Proper cube action: No redouble, take (2.2%)

So far no problem. Doubling is borderline and I have 72.5% winning
chances. Still, I decided to run a rollout:

Cube analysis
Rollout cubeless equity  +0.369 (Money:  +0.362)

Cubeful equities:
1. No double            +0.862
2. Double, pass         +1.000  ( +0.138)
3. Double, take         +0.747  ( -0.115)
Proper cube action: No redouble, take (45.4%)
Rollout details:
Player SubtleOne owns 2-cube:
   73.0%   0.0%   0.0% -  27.0%   8.9%   0.9% CL  +0.369 CF  +0.862
 [  0.3%   0.0%   0.0% -   0.3%   0.2%   0.1% CL   0.008 CF   0.016]
Player MartinKrainer owns 4-cube:
   74.0%   0.0%   0.0% -  26.0%   8.8%   0.9% CL  +0.847 CF  +0.747
 [  0.3%   0.0%   0.0% -   0.3%   0.2%   0.1% CL   0.016 CF   0.019]
Full cubeful rollout with var.redn.
648 games, Mersenne Twister dice gen. with seed 1179198464 and
quasi-random dice
Play: 0-ply cubeful [expert]
Cube: 0-ply cubeful [expert]

Now I'm really confused. According to this everything is improved. I
have better winning chances, and I get gammoned less, yet now doubling
is a big mistake. Why?

Mind you, it's perfectly ok that I doubled wrong, I just don't
understand the numbers.

                                                Albert