Neil seems to be correct that bots underestimate backgames. At least this is a good example of it.
Four-ply says this is a thin (.009) cube. Full rollout (below) tells a very different story: cubing is not exactly a blunder, but it's a .06 error.
                        |
| ||||
| XGID=-cdBCDC-B-A-a-----bbba----:0:0:1:00:0:0:3:0:10 | |||||
| Blue on roll, cube action? | |||||
| Analyzed in 4-ply | No double | Double/Take |
| Player Winning Chances: | 54.37% (G:32.99% B:14.28%) | 54.49% (G:32.94% B:14.44%) |
| Opponent Winning Chances: | 45.63% (G:0.94% B:0.05%) | 45.51% (G:0.91% B:0.05%) |
| Cubeless Equities | +0.550 | +1.108 |
| Cubeful Equities | ||
| No double: | +0.712 (-0.009) | |
 Double/Take: | +0.721 | |
| Double/Pass: | +1.000 (+0.279) | |
| Best Cube action: Double / Take | ||
eXtreme Gammon Version: 2.10
|
| ||||
| XGID=-cdBCDC-B-A-a-----bbba----:0:0:1:00:0:0:3:0:10 | |||||
| Blue on roll, cube action? | |||||
| Analyzed in Rollout | No double | Double/Take |
| Player Winning Chances: | 52.93% (G:31.55% B:14.14%) | 52.51% (G:31.03% B:13.84%) |
| Opponent Winning Chances: | 47.07% (G:0.61% B:0.02%) | 47.49% (G:0.62% B:0.01%) |
| Cubeless Equities | +0.509 | +0.985 |
| Cubeful Equities | ||
| No double: | +0.605 | ±0.037 (+0.568..+0.642) |
| Double/Take: | +0.541 (-0.064) | ±0.039 (+0.502..+0.581) |
| Double/Pass: | +1.000 (+0.395) | |
| Best Cube action: No double / Take | ||
| Percentage of wrong pass needed to make the double decision right: 12.2% | ||
| Rollout details | ||
| 1296 Games rolled with Variance Reduction. Moves: 3-ply, cube decisions: XG Roller | ||
| Double Decision confidence: | 99.0% | |
| Take Decision confidence: | 100.0% | |
| Duration: 14 minutes 47 seconds | ||
eXtreme Gammon Version: 2.10