I saw that a recent Sage update included a new backgame engine: “New backgame engine! Sage's new engine excels at backgames and containment games- even deep backgames.” It’s exciting to see some engine developments after XG has stalled over the last 15 years, and even more so for an update targeting a notorious weak spot for backgammon engines.
I tested a snake position I found on Reddit a couple of months back, and the rollouts still show substantial differences between XG2 and Sage in estimated winning chances—and especially gammon rates. Their preferred moves differ too.
                      |
| ||||
| XGID=----a-ABCBBCB----------eh-:1:-1:1:63:0:0:3:0:10 | |||||
| Blue to play 63 | |||||
| 1. | Rollout1 | 12/9 12/6 | eq: +0.661 | |||
| ||||||
| 2. | Rollout1 | 12/6 8/5 | eq: +0.629 (-0.032) | |||
| ||||||
| 1 12960 Games rolled with Variance Reduction. Moves: 3-ply, cube decisions: XG Roller | ||||||
eXtreme Gammon Version: 2.19.211.pre-release
XG estimates White’s winning chances at 13.61% after 12/9 12/6 and 14.63% after 12/6 8/5, compared with Sage’s 9.5% and 8.8%, respectively. The gammon estimates were particularly striking: XG gives White 3.94% and 5.12%, versus Sage’s 0.9% and 0.6%.
We tend to treat rollouts as the “truth,” but differences like these raise interesting questions. I’d welcome others testing relevant positions and sharing their results.