One possibility is to require that for any particular play or cube decision, a player may challenge xg's analysis and play a prop. I tried the snake scenario (below) and ended up with an ER of 28.3, including a quintella on the first roll. Part of that may be deserved, as I managed to allow xg some potential 17-1 jokers. But the odd part (at least to me) is the xg rated itself at 31.4.

              
                   
 | is eXtremeGammon
score: 0 pip: 38 | Money session Jacoby Beaver | pip: 311 score: 0
is blotto | |
| XGID=-a----------------BBCBCCn-:0:0:1:63:0:0:3:0:10 |
to play 63 |
| 1. | XG Roller | 22/16 20/17 | eq: -0.430 |
| Player: Opponent: | 50.09% (G:1.28% B:0.00%) 49.91% (G:40.61% B:1.45%) | |
|
| 2. | XG Roller | 23/17 20/17 | eq: -0.899 (-0.469) |
| Player: Opponent: | 40.48% (G:1.36% B:0.00%) 59.52% (G:53.27% B:2.03%) | |
|
| 3. | 3 ply | 21/15 20/17 | eq: -0.877 (-0.447) |
| Player: Opponent: | 44.60% (G:1.13% B:0.00%) 55.40% (G:47.30% B:2.06%) | |
|
| 4. | 3 ply | 20/11 | eq: -1.000 (-0.570) |
| Player: Opponent: | 40.98% (G:1.18% B:0.00%) 59.02% (G:50.21% B:2.71%) | |
|
| 5. | 3 ply | 20/17 19/13 | eq: -1.000 (-0.570) |
| Player: Opponent: | 40.84% (G:1.19% B:0.00%) 59.16% (G:50.15% B:4.84%) | |
|
eXtreme Gammon Version: 1.20