I would say that the main exceptions to "Stick's rule" have the following properties:
1. The 3a player loses 5% pseudocubeless gammons or less.
2. The take decision is close to the take/drop borderline.
If these two conditions are satisfied then it can be a big double even if the 3a player wins a ton of gammons.
I haven't asked the bot about this position at 3a8a, 3a9a, etc., but I'm sure that at those scores it will no longer be close to the take/drop borderline (because the takepoint changes), and so it will be a big mistake to double. But I suspect it will be possible to craft a similar position where the above two conditions will be satisfied, and then I predict the bot will indicate that the 3a player should double.
                       |
| ||||
| XGID=bB-BbBBBB----A---abd-b-bB-:0:0:1:00:4:0:0:7:10 | |||||
| Blue on roll, cube action? | |||||
| Analyzed in Rollout | No double | Double/Take |
| Player Winning Chances: | 75.67% (G:47.47% B:2.06%) | 75.71% (G:50.36% B:7.96%) |
| Opponent Winning Chances: | 24.33% (G:5.04% B:0.17%) | 24.29% (G:5.29% B:0.63%) |
| Cubeless Equities | +0.963 | +1.600 |
| Cubeful Equities | ||
| No double: | +0.915 (-0.085) | ±0.010 (+0.905..+0.925) |
| Double/Take: | +1.005 (+0.005) | ±0.021 (+0.985..+1.026) |
| Double/Pass: | +1.000 | |
| Best Cube action: Double / Pass | ||
| Rollout details | ||
| 1296 Games rolled with Variance Reduction. Dice Seed: 271828 Moves: 3-ply, cube decisions: XG Roller | ||
| Double Decision confidence: | 100.0% | |
| Take Decision confidence: | 69.4% | |
| Duration: 31 minutes 39 seconds | ||
eXtreme Gammon Version: 2.19.208.pre-release, MET: Kazaross XG2