My suggestion was to up the play strength, not the number of games rolled out. Taken as a pair your rollout and mine suggest that the base evaluations XG has are just confused. For the 6/1 5/3 play the loss of equity in races and the loss when a hit do come are probably not being accounted for enough. It shouldn't matter too much though since it does have a positive of saving a 6 and those downsides often don't come into play.
Compared to your "standard settings" rollout showing 6/1 5/3 beating 7/2 3/1 by .035 my rollout shows it losing by .017, so more than a .05 swing when stronger settings were used. I suspect that even stronger settings would move the difference even higher.
          
 
          
 | | White is Player 2
score: 0
pip: 105 | Unlimited Game
Jacoby Beaver | pip: 90
score: 0
Blue is Player 1 |
|
| XGID=--ACBBBABb---Ba--abcbb-b--:2:-1:1:52:0:0:3:0:10 |
| Blue to play 52 |
| 1. | Rollout1 | 7/2 3/1 | eq: +0.3665 |
| Player:
Opponent: | 72.35% (G:0.25% B:0.01%)
27.65% (G:0.49% B:0.01%) | Conf.: ±0.0044 (+0.3621...+0.3709) - [100.0%]
Duration: 2 hours 12 minutes |
|
| 2. | Rollout1 | 6/1 5/3 | eq: +0.3496 (-0.0169) |
| Player:
Opponent: | 71.58% (G:0.30% B:0.00%)
28.42% (G:0.59% B:0.02%) | Conf.: ±0.0049 (+0.3447...+0.3545) - [0.0%]
Duration: 2 hours 32 minutes |
|
| 3. | XG Roller++ | 8/3 6/4 | eq: +0.3597 (-0.0068) |
| Player:
Opponent: | 72.39% (G:0.41% B:0.00%)
27.61% (G:0.69% B:0.02%) | |
|
| 4. | XG Roller++ | 7/5 6/1 | eq: +0.3594 (-0.0071) |
| Player:
Opponent: | 72.34% (G:0.37% B:0.00%)
27.66% (G:0.64% B:0.01%) | |
|
| 5. | XG Roller++ | 6/4 6/1 | eq: +0.3549 (-0.0116) |
| Player:
Opponent: | 72.16% (G:0.24% B:0.00%)
27.84% (G:0.51% B:0.01%) | |
|
| |
1 1296 Games rolled with Variance Reduction.
First 6 moves and cube decisions: XG Roller+
Remaining moves and cube decisions: XG Roller
Search interval: Huge
|
eXtreme Gammon Version: 2.10