I recall that what XG means by 2-ply is what GNU means by 1-ply, so maybe that is part of the confusion.
What a human beginner can see is not necessarily a good guide to what a neural net can see. Beginners can be taught to roll primes, but look what XG does below.
          
 
    
 | | White is Player 2
score: 0
pip: 11 | | Unlimited Game | pip: 300
score: 0
Blue is Player 1 |
|
| XGID=--------------aA-ABCBABBA-:1:1:1:23:0:0:0:0:10 |
| Blue to play 23 |
| 1. | 3-ply | 24/21 15/13 | eq: -2.079 |
| Player:
Opponent: | 4.79% (G:0.00% B:0.00%)
95.21% (G:69.37% B:49.82%) | |
|
| 2. | 3-ply | 24/21 18/16 | eq: -2.394 (-0.314) |
| Player:
Opponent: | 2.51% (G:0.00% B:0.00%)
97.49% (G:81.93% B:63.33%) | |
|
| 3. | 3-ply | 21/18 15/13 | eq: -2.446 (-0.367) |
| Player:
Opponent: | 0.96% (G:0.00% B:0.00%)
99.04% (G:80.33% B:66.51%) | |
|
| 4. | 3-ply | 24/21 19/17 | eq: -2.501 (-0.422) |
| Player:
Opponent: | 1.43% (G:0.00% B:0.00%)
98.57% (G:89.97% B:63.48%) | |
|
| 5. | 3-ply | 19/16 15/13 | eq: -2.531 (-0.452) |
| Player:
Opponent: | 1.40% (G:0.00% B:0.00%)
98.60% (G:81.46% B:74.95%) | |
|
| 6. | 3-ply | 18/16 15/12 | eq: -2.639 (-0.560) |
| Player:
Opponent: | 1.31% (G:0.00% B:0.00%)
98.69% (G:87.32% B:79.65%) | |
|
| 7. | 3-ply | 19/16 18/16 | eq: -2.669 (-0.590) |
| Player:
Opponent: | 0.82% (G:0.00% B:0.00%)
99.18% (G:88.40% B:80.43%) | |
|
| 8. | 3-ply | 18/16 17/14* | eq: -2.675 (-0.596) |
| Player:
Opponent: | 0.86% (G:0.00% B:0.00%)
99.14% (G:89.93% B:79.60%) | |
|
eXtreme Gammon Version: 2.19.211.pre-release