Thanks, Tim, for revealing this bust; better to find out now than even later. :)
In 2005/2006, I had a very slow computer; moreover, Snowie-4 is a snail compared to XG. In fiddling around with positions and variants, I mostly relied on truncated rollouts as a guide. Full rollouts I farmed out to Paul Weaver and our merry band of rollateers.
Having constructed a position that, by truncated rollouts, seemed to produce a different best play at each of the four key scores, I managed to convince Paul that it was worth doing a full rollout for it, even though we had many other positions (for our book) competing for limited CPU time. What I eventually got back (carefully documented though later lost in a crash) was a 10k rollout of two plays and a 2k rollout of the other three plays, all cubefully adjusted (no live cube). That may sound shorthanded, but the result seemed clear enough for the money score:
- [V B6 @44 E46 K126] ~10*2
...where V = 24/21(3) 21/18, B = 24/21(2) 13/10(2), @ = 24/18(2), E = 24/21(3) 6/3, and K = 13/4*/1*. (Refer to diagram below.)
Notice that the (DMP-only-supported) anchoring play is –.044: this is the play that emerges as best/tied in an XG2 money rollout!
| |
|---|---|
| Published R4 position (busted) | |
- ..[V B6 @44 E46 K126]
d[@ B8 V12 K38 E66]
s[E V16 B27 @89 K212]
g[K B53 @62 E151]
Paul found GnuBG too difficult to use, and I couldn't even download it onto my gimpy computer. Fortunately, a guy named Stick (whom I didn't know yet) offered to roll out the position for Paul at all four scores on GnuBG. Not long after, Paul reported to me that Stick had Gnu-verified that this was a definite four-score position. That information, combined with the data we already had from Snowie, seemed sufficient, and a year later we published the position in Backgammon Openings (Match Play section, p. 98).
Assuming that XG2 3-ply's money rollout of this position is trustworthy (and I have no independent reason to believe otherwise), it is mystifying that the money rollouts of both Snowie and GnuBG had @ (anchor, 24/18(2)) out of the running. At some point, out of curiosity, I will reroll it on both of the older bots.
Stick has not been able to find (at least not yet) those GnuBG rollouts from 2006, but he was able to determine (from a no-longer-functioning link where he had posted the rollouts for Paul) that the version number was 14.3 devel, one that often produced corrupted rollouts (as we discovered only years later -- indeed, some are posted on the navigable early game rollout compendium of this site).
With XG2 on tap now, it is worlds faster and easier to hunt for and verify (or bust) T4's. Preliminary indications suggest that sliding White's 9pt spare to her 10pt as shown in the diagram below (thereby enabling Blue to duplicate 3s when he plays V), yields a T4 and maybe even an R4 (real four-score position). This is one of the many variants I tried before that didn't seem to work when my tools were stone knives and bearskins. I'll report again on this new T4 candidate when I get further along.
| |
|---|---|
| New T4 candidate | |
Nack