|
| ||||||||||||||||
- [H X10 P38] ~10*5
[H X7 P62] /10*1
I believe the difference in the D/P margins is more attributable to the variance in Gnu's 1k trials than to Snowie's rollout being with no live cube.
Key: H = Hit and down (13/10 8/7), X = hit and split (24/21 8/7*), P = Point (8/5 6/5). Error sizes are in thousandths. After the brackets, the bot symbols are ~ is Snowie, and / is GnuBG. The number of trials is in thousands (rounded down), with an asterisk being a repeater. For example, "[H X10 P38] ~10*5" means that H is best and X is -.010 according to a Snowie rollout of 10000+ trials, and P is -.038 after 5000+ trials.
For reference, the Snowie result (again, no live cube) for 53P-63S-31 (right-hand position) is
- [X H1 P20] ~10
Blue has greater incentive to split against the 10pt builder because of the direct threat to hit it. As you can see, the effect on the X/H margin is only .011, though, with 6 duplication being double-edged.
In Backgammon Openings, pp. 40-41, 64P-32Z-31 and 64P-63S-31 are analyzed and compared. The latter position is similar to the right-hand position diagrammed above, but X beats H by more than .030. The deeper Blue's inside point, the more willing is to split rather than try to make a prime.
Nack