I know it's tempting to think that what you describe is what bots are doing. But according to all my information and evidence it isn't. Granted, I'm speaking mostly about gnubg. Here is a piece of evidence:
The score (after 0 games) is: gnubg 0, stiebler 0 (match to 1 point, post-Crawford play)
Move number 2: stiebler to play 13
| gnubg | 162 |
|---|---|
| |
| stiebler | 167 |
Position ID: 4HPkASLgc/ABMA Match ID: cIksAAAAAAAA | |
• stiebler moves 8/5 6/5
3-ply gnubg-style:
| # | Ply | Move | Equity | |
|---|---|---|---|---|
| • | 1 | 3 | 8/5 6/5 | -0,00262 |
| 0,49869 0,13996 0,00497 - 0,50131 0,13220 0,00571 | ||||
3-ply Rich-style:
| # | Ply | Move | Equity | |||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| • | 1 | R | 8/5 6/5 | -0,00279 | ||||||||||||||||||
| ||||||||||||||||||||||
| Truncated cubeful rollout (depth 1) without var.redn. | ||||||||||||||||||||||
| 36 games, Mersenne Twister dice gen. with seed 903957018 and quasi-random dice | ||||||||||||||||||||||
| Play: world class 2-ply cubeful prune [world class] | ||||||||||||||||||||||
| keep the first 0 0-ply moves and up to 8 more moves within equity 0,16 | ||||||||||||||||||||||
| Skip pruning for 1-ply moves. | ||||||||||||||||||||||
| Cube: 2-ply cubeful prune [world class] | ||||||||||||||||||||||
3-ply Maik-style:
| # | Ply | Move | Equity | |||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| • | 1 | R | 8/5 6/5 | -0,00262 | ||||||||||||||||||
| ||||||||||||||||||||||
| Truncated cubeful rollout (depth 3) without var.redn. | ||||||||||||||||||||||
| 46656 games, Mersenne Twister dice gen. with seed 903957018 and quasi-random dice | ||||||||||||||||||||||
| Play: 0-ply cubeful [expert] | ||||||||||||||||||||||
| Cube: 0-ply cubeful [expert] | ||||||||||||||||||||||
I'm much less sure what XG does. I have seen evidence that a least the cube decisions are more sophisticated than only 1-ply in the lookahead. Don't remember exactly anymore.
Of course, Rich-style evals would be much more computationally expensive than Maik-style evals. I cannot really imagine that speed monster XG does Rich-style if gnubg does not.