This won't please the fans of W, but I think is quite clear.
Thanks, Neil, for the XG 4-ply rollout. It agrees with Snowie, and somewhat disagrees with GnuBG and XG 3-ply.
(It's funny how often XG's 3-ply and 4-ply seem to produce different results, though in this case your 4-ply was only 15k so the jury may still be out on this one).
Here is data for 62S-32 (see left-hand diagram below):
Money
Snowie .. [U S11 W14 Z31 $33 T39 %39 H40] 36kUSW 20k
GnuBG .. [W S1 U1 H22 Z25 $28 T29 %40] 46kUSW 5k
XG 3-ply [U S3 W4 T26 Z27 H30 $34] 46kUSW 5k
XG 4-ply [U S14 W15 H38 T39 Z42] 15kUSW 5k
Other scores
d[W S5 U7] 10k
s[U S3] 15k
g[$ %7 S9 H11] 10k
Legend: U = Up (24/22 24/21), S = Split (24/21 13/11), W = Wild (24/21 6/4), Z = reverse split (24/22 13/10), $ = Slot (13/10 6/4), T = sTack/Tower (13/8), % = alt slot (13/11 8/5), H = Hit (6/1*).
Error sizes are in thousandths. If the score is non-money, a lower-case letter appears before the brackets (d = DMP, s = gammon save, and g = gammon go). After the brackets is the number of trials (in thousands, rounded down). For example, "[U S11 W14...] 36kUSW 20k" means that U is best, S is -.011 and W is -.014 after 36000+ trials (and the other plays were rolled out 20000+).
"Fans of W" can still be content that W is best at DMP (to the degree that GnuBG's rollout can be trusted), and that anyway W is close enough for money to be quite playable -- indeed I make it only -.007 or so ("tied" according to the B/W error scale) after amalgamating all current bot data.
Finally, for 63S-32 (White has a 10pt builder instead of an 11pt builder, shown in the right-hand diagram below), Blue's W is in a dead heat with S and Z (and maybe U). My estimated bot-average margins for plays within .01 are listed in the captions.
|
| ||||||||||||||||
