[Bug-gnubg] Bearoff anomaly(bug?)

[Michael Petch]

Before beginning.
1) Go to Settings menu/Players. Set Player 0 as GNU Backgammon at and
Expert Level. Player 1 should be human
2) Go to Settings/Analysis. Uncheck "Same as Analysis" and select Expert
3) Go to Settings/Options/Dice and Use Manual Dice

4) Start with this pasting this position:

 GNU Backgammon  Position ID: 34sPAAB/AAAAAA
                 Match ID   : UQlgAAAAAAAE
 +13-14-15-16-17-18------19-20-21-22-23-24-+     O: gnubg
 |                  |   | O        O  O  O |     0 points
 |                  |   | O           O  O |
 |                  |   | O           O  O |
 |                  |   | O           O  O |
 |                  |   | O              O |
v|                  |BAR|                  |     3 point match
 |                  |   |                7 | X
 |                  |   |                X | X
 |                  |   |                X | XX
 |                  |   |                X | XX  On roll
 |                  |   |                X | XX  0 points
 +12-11-10--9--8--7-------6--5--4--3--2--1-+     X: moi (Cube: 2)

Evaluation before roll says:

        Win     W(g)    W(bg)   L(g)    L(bg)   MWC       Cubeful
static: 1.000000 0.004300 0.000000 0.000000 0.000000  75.22489%  75.22489%

**********************************
Please Note: At this point it shows "moi" has chances to gammon which
is correct.
**********************************

5) Roll a 21 (or any non double)

**********************************
First issue. The bot will attempt to resign a regular win, although it
can still be gammoned.
**********************************

6) Reject the resign
7) Roll 54

 GNU Backgammon  Position ID: HwAAwPfiAwAAAA
                 Match ID   : EQFxAAAAAAAE
 +12-11-10--9--8--7-------6--5--4--3--2--1-+     O: gnubg
 |                  |   | O        O  O  O |     0 points
 |                  |   | O           O  O |     Rolled 24
 |                  |   | O           O  O |
 |                  |   | O           O  O |
 |                  |   | O              O |
^|                  |BAR|                  |     3 point match
 |                  |   |                X | XX
 |                  |   |                X | XX
 |                  |   |                X | XX
 |                  |   |                X | XX
 |                  |   |                X | XX  0 points
 +13-14-15-16-17-18------19-20-21-22-23-24-+     X: moi (Cube:2)

Evaluation shows:

       Win     W(g)    W(bg)   L(g)    L(bg)   MWC       Cubeful
static: 0.000000 0.000000 0.000000 0.000000 0.000000  24.88210%  24.88210%

**********************************
Second Issue (in the data above). At this point it is obvious why it
tried to resign one point. It seems the bearoff code has determined this
position can't be gammoned (no gammon losses possible according to the
evaluation. Clearly resigning without being assured the save is bad.
**********************************

So the question becomes, why is the gammon value 0 for gnubg's position
when the roll before (moi) actually knew it was still gammonish. It
seems like the bearoff database may have the right gammon related data,
but the calculations on that data yield a wrong result? It doesn't seem
normal to me.

I happened to identify this anomaly when I analysed the results of about
10 million positions from 21000+ 11pt matches played bot vs bot at 0ply.
These positions also exhibit the same behavior:

HwAAwPfiAwAAAA:MAFpARAAOAAE
LwAAQHfeAwAAAA:sAFxAaAAAAAE
vwEAAL5vNAAAAA:MAF5AUAAIAAE
LwAAgOX7AwAAAA:8AlzAVAAUAAE
HwAAwFt+AwAAAA:AQFuAaAAGAAE
dwAAgGv2BwAAAA:AYFmAQAAIAAE
dwAAgDf5BwAAAA:UQltASAAAAAE
XwAAgP88AwAAAA:UYlmATAAMAAE
LwAAwOfHAwAAAA:AQF5AWAAIAAE
LwAAgDv+AwAAAA:AQFyAVAAGAAE
XwAAgL2PBwAAAA:UQltAVAAUAAE

I do know this anomaly isn't related to Philippe Michel's small change
to the bearoff code to fix an undefined left shift with a negative
value. I tried a version of GNUBG that predates that change. As well
generating a one sided bear off database with no gammon values has the
same result.

If you use higher plies it will begin to notice the position is gammonable.

Is this normal behavior or am I missing something?

-- 
Michael Petch
CApp::Sysware Consulting Ltd.
OpenPGP FingerPrint=D81C 6A0D 987E 7DA5 3219 6715 466A 2ACE 5CAE 3304