The following position is from Magriel, Chapter 17 (Slotting), Position 9.


(Side note: In Magriel, the original problem was simply "X [Black] to play 1", and the Black checker on the bar in the diagram above, was on the 20 point. Since there's no way to get XG to play a single number, I set the roll to 5-1 and put a Black checker on the bar to force an entry on the 20-point with the 5.)

The oddity is this:

If I set the conditions to Unlimited Games, with Jacoby and Beavers turned off, then XG at 3-ply thinks that the slotting play (Bar/20 7/6) is virtually equal to the non-slotting play Bar/19. Only at 4-ply does XG think that the slotting play is superior.

Now if I change the conditions to match play to 11 points with a score of 0-0, a score that should be virtually identical to money play with no Jacoby or Beavers, then XG can immediately see at 3-ply that the slotting play is better than the non-slotting play.

What's going on here? The benefits of slotting should be visible at 3-ply: Black slots, White dances, Black covers (3 plies).

I have no idea about the backgammon side, but mathematically there could be a large cube in the calculation somewhere that's messing with things. Try cranking the match to 15 points or higher and see how far you have to go to eliminate that artifact.

I cranked it up to 99 and the result is the same as for 11. The same discrepancy is there. There must be something else going on.

Well... there goes my one idea. Maybe someone else will figure it out.

The cubeless equities are pretty close comparing 3-ply money and 3-ply match, so it looks like for matches, it's using a different cube model and that's affecting things. Slotting is much better almost entirely because you get an efficient cube after the dances, so the cubeful equities are very sensitive to whether the bot understands that or not at a given ply.

(The reason why the cubeless equities don't match exactly between money and match is because the cubeless equities are adjusted for score in a match, I believe.)

Thank you, that explanation looks correct. If I move the cube to White's side, then the discrepancy vanishes.