I'm no expert, but one area where you can never trust XGR++ evaluations is backgame cubes. Just one example to illustrate the idea:
                      |
| ||||
| XGID=-bbEDF--------a---bbbbb---:0:0:1:00:0:0:3:0:10 | |||||
| Blue on roll, cube action? | |||||
| Analyzed in Rollout | No double | Double/Take |
| Player Winning Chances: | 60.02% (G:36.05% B:9.17%) | 58.82% (G:35.38% B:9.20%) |
| Opponent Winning Chances: | 39.98% (G:0.00% B:0.00%) | 41.18% (G:0.00% B:0.00%) |
| Cubeless Equities | +0.653 | +1.244 |
| Cubeful Equities | ||
| No double: | +0.879 | ±0.006 (+0.873..+0.886) |
| Double/Take: | +0.858 (-0.021) | ±0.011 (+0.848..+0.869) |
| Double/Pass: | +1.000 (+0.121) | |
| Best Cube action: No double / Take | ||
| Percentage of wrong pass needed to make the double decision right: 13.0% | ||
| Rollout details | ||
| 5184 Games rolled with Variance Reduction. Dice Seed: 81232284 Moves and cube decisions: XG Roller+ Search interval: Huge | ||
| Double Decision confidence: | 100.0% | |
| Take Decision confidence: | 100.0% | |
| Duration: 10 hours 43 minutes | ||
eXtreme Gammon Version: 2.10
|
| ||||
| XGID=-bbEDF--------a---bbbbb---:0:0:1:00:0:0:3:0:10 | |||||
| Blue on roll, cube action? | |||||
| Analyzed in XG Roller++ | No double | Double/Take |
| Player Winning Chances: | 60.88% (G:36.75% B:9.43%) | 61.53% (G:37.44% B:9.33%) |
| Opponent Winning Chances: | 39.12% (G:0.00% B:0.00%) | 38.47% (G:0.00% B:0.00%) |
| Cubeless Equities | +0.679 | +1.397 |
| Cubeful Equities | ||
| No double: | +0.956 (-0.044) | |
| Double/Take: | +1.106 (+0.106) | |
| Double/Pass: | +1.000 | |
| Best Cube action: Double / Pass | ||
eXtreme Gammon Version: 2.10
That, in particular, brings a question of how you estimate players' PR in a match that involved backgames? It would be way off unless you apply strong settings rollouts.