Statistically significant?
Of course not. But Chuck, perhaps you too might spend some anecdotal time finding out how easy it is to steer Gnubg 2-ply into positions like this:
Move number 107: Human on roll, cube decision?
| Bot | 29 |
|---|---|
| |
| Human | 160 |
Position ID: /wMAEABs2x0AAQ Match ID: UgkAAAAAAAAA | |
• Human doubles
How did the human get a 4-cube? By beavering an impetuous double. Not a "backgame double" -- I think Gnubg plays normal backgames pretty well. But this game never reached a "normal" backgame position. Blue had 13 checkers back (with some help from the bot which repeatedly hit checkers on Blue's own ace and deuce points), and Gnubg does not know how to prevent Blue from building a counterprime.
Above, 2-ply will happily take your "no double/take" 8-cube. Roll it out if you want. A 2-ply rollout will either have this as a close take or a close pass. I've got it as an 0.014 pass after 144 trials. But I know you can do better than that, because you won't be making the bot's rollout misplays that caused it to lose 2.3% gammons and (it's therefore likely) some additional single games too.
But positions like that one aren't your bread and butter in a milk-the-bot session. In the position below, Gnubg 2- and 4-ply will beaver your double.:
| Bot | 50 |
|---|---|
| |
| Human | 247 |
Position ID: 3+cBAAEAsG1bCQ Match ID: UgkAAAAAAAAA | |
• Human doubles
Alert: wrong double ( -0.9270)!
| Cube decision | |||
|---|---|---|---|
| 4-ply cubeless equity | -0.3684 | ||
| 0.4700 0.0059 0.0000 - 0.5300 0.3080 0.0064 | |||
| Cubeful equities: | |||
| 1. | No double | -0.2083 | |
| 2. | Double, pass | +1.0000 | +1.2083 |
| 3. | Double, take | -1.1352 | -0.9270 |
| Proper cube action: | No redouble, beaver (43.4%) | ||
The exercise does not produce a normal gammon loss vs. single win trade-off. It's not lose 8 vs. win 4. More like lose 8 vs. win 16.