[Bug-gnubg] Confusing GNU rollout results !!

[Ian Shaw]

    GNU Backgammon  Position ID: zA7yBAPIM+YwQA
                    Match ID   : cAlyAaAASAAA
    +24-23-22-21-20-19------18-17-16-15-14-13-+  O: Woolsey
    |       O     O  O |   |    X        O  X |  10 points
    |       O     O  O |   |    X           X |  
    |                O |   |                X |  
    |                  |   |                  |  
    |                  |   |                  |  
    |                  |BAR|                  |v 11 point match (Cube: 1)
    |                  |   |                  |  
    |                X |   |                O |  
    |                X |   |                O |  
    |             O  X |   |    X        X  O |  Rolled 44
    |          X  O  X | X |    X     O  X  O |  9 points
    +-1--2--3--4--5--6-------7--8--9-10-11-12-+  X: Readers

The following was posted by on GammonLine:
Neil Kazaross -- Thursday, 12 December 2002, at 3:12 p.m.

1. Rollout bar/17 8/4(2) MWC: 28.54% 

0.502 0.055 0.002 - 0.498 0.112 0.008 CL 28.54% 

[0.001 0.001 0.000 - 0.001 0.002 0.001 CL 0.28%] 

Full cubeless rollout (trunc. at one-sided bearoff) with var.redn., with rotate 
rolls 

5184 games, seed 1154634112 using Mersenne Twister 

Play and cube: 0-ply cubeless 

2. Rollout bar/21 13/9(3) MWC: 28.24% ( -0.30%) 

0.496 0.064 0.004 - 0.504 0.143 0.012 CL 28.24% 

[0.001 0.001 0.001 - 0.001 0.002 0.001 CL 0.30%] 

Full cubeless rollout (trunc. at one-sided bearoff) with var.redn., with rotate 
rolls 

5184 games, seed 1154634112 using Mersenne Twister 

Play and cube: 0-ply cubeless 

3. Rollout bar/17 11/7(2) MWC: 28.12% ( -0.41%) 

0.513 0.047 0.003 - 0.487 0.082 0.005 CL 28.12% 

[0.001 0.001 0.001 - 0.001 0.002 0.001 CL 0.24%] 

Full cubeless rollout (trunc. at one-sided bearoff) with var.redn., with rotate 
rolls 

5184 games, seed 1154634112 using Mersenne Twister 

Play and cube: 0-ply cubeless 

I meant to launch this 1 ply..but anyhow..these results are confusing. All that 
matters at this score are our side's wins and gammons and our gammon value is 
1. Therefore we simply can add up our wins and gammons and the play with that 
highest resulting total will be the best play. However, the MWC reported by GNU 
doesn't reflect that. 

Let's look at play 2 where it is simple to do the math. We win the match 6.4% 
on our G's plus half of the remaining 43.2% wins or 21.6 + 6.4 =28 % and GNU is 
reporting this as 28.24% !! Is GNU doing two different rollouts or is something 
screwing up with truncating to the bearoff database ? 

What's wrong ? ....neilkaz.... 
----------------------------------------
I've checked the evaluations, and they seem to be correct (within the limits of 
rounding error).

    1. Cubeful 2-ply    bar/17 11/7(2)               MWC:  28.85%  [28.855}
       0.521 0.056 0.001 - 0.479 0.079 0.001
        2-ply cubeful 100% speed [world class]
    2. Cubeful 2-ply    bar/21 13/9(3)               MWC:  28.72% ( -0.13%)   
[28.70%]
       0.494 0.080 0.002 - 0.506 0.136 0.004
        2-ply cubeful 100% speed [world class]
    3. Cubeful 2-ply    bar/17 8/4(2)                MWC:  28.24% ( -0.61%)   
[28.20%]
       0.499 0.065 0.001 - 0.501 0.100 0.002
        2-ply cubeful 100% speed [world class]

--Ian