[Bug-gnubg] Bad 0-ply evaluation (no bug)

[Christopher D. Yep]

No bug to report, but here's a very bad 0-ply evaluation.  No backgammon 
program is perfect, but the size of this 0-ply mis-evaluation motivated me 
to post it to the group.  Any idea why GNU's 0-ply evaluation is so far off 
the mark in this position?

    GNU Backgammon  Position ID: cMMHBDzemwAAAA
                    Match ID   : QQkPAAAAAAAA
    +24-23-22-21-20-19------18-17-16-15-14-13-+  O: gnubg (Cube: 2)
    |             O  O |   |          O       |  0 points
    |             O  O |   |          O       |
    |             O    |   |          O       |
    |                  |   |          O       |
    |                  |   |          O       |
    |                  |BAR|                  |v
    |                  |   |                  |
  X | O  X  X          |   |                  |
  X | O  X  X          |   |                  |
  X | O  X  X  X       |   |                  |  Rolled 63
  X | O  X  X  X     X |   |    O             |  0 points
    +-1--2--3--4--5--6-------7--8--9-10-11-12-+  X: Chris

    1. Cubeful 0-ply    6/3 4/off                    Eq.:  +2.099
        92.2%  87.0%  41.1% -   7.8%   0.0%   0.0%
        0-ply cubeful [expert]
    2. Cubeful 0-ply    6/off 3/off                  Eq.:  +1.544 ( -0.556)
        87.4%  65.5%  18.3% -  12.6%   0.0%   0.0%
        0-ply cubeful [expert]

GNU 0-ply strongly likes 6/3 4/off!!

The above position is perhaps slightly unusual (though not "complex" in my 
opinion), but a close variant (moving O's stack from O's 10 point to O's 9 
point) occurred in an actual game that I played vs. GNU.  Even in this 
variation, GNU still prefers 6/3 4/off at 0-ply (though only by .070).  If 
the stack is moved to the 7, 8, 11, 12, or 13 pts., the evaluation is then 
in the expected order (6/off 3/off the top move).  Note that this 
evaluation oddity disappears when evaluating at higher plies (lookahead 
catches this obvious mis-evaluation).

Chris