I had actually checked the live cube equities in my Snowie rollouts. What I glossed over in my rush however was the negative sign in front of the -0.004 so I figured the difference as .008 instead of .016.
It is interesting to assume Snowie is somehow donking up the two down play. (I don't believe it as bad as SW reports) Perhaps someone has the time now as I do not but have you checked how it (and the other bots) play the immediate replies?
David has checked a lot of evals, so his opinion would be useful here, but from what I've seen, GnuBG misplays significantly more replies than Snowie. I don't know about XG. OTOH, it's natural to suspect Snowie's [S D18] 75k result when Gnu and XG agree on [S D7].
I have checked Snowie's replies to opening 32S, 32D, and 32Z with all reply rolls (21*3 = 63 evals). There were two misreplies to 32D:
32D-33E (-.028)
32D-51S (-.004)
Snowie made zero misreplies to 32S or 32Z. Therefore, 32D gains (.028/36 + .004/18) = .001, adjusting Snowie value only slightly, from [S Z17 D18] to [S Z17 D17].
I'm clumsy with Gnu and haven't yet used XG, so I leave it to someone else to check evals (for all three plays, for all twenty-one rolls) for those bots. To save you/whomever time, the correct replies (by agg) are listed below:
32D: 11N 21$ 22N 31P 32ZD 33C 41$ 42P 43D 44B 51$ 52S 53P 54H 55P 61P 62S 63H 64H 65R 66B
32S: 11P 21X 22N 31P 32H 33B 41X 42P 43H 44C 51S 52H 53P 54H 55A 61P 62X 63H 64H 65R 66B
32Z: 11e 21S 22N 31P 32K 33A 41U 42P 43X 44B 51R 52S 53P 54H 55A 61P 62S 63H 64H 65R 66B
When you/whoever report back Gnu's and/or XG's errors (if any), I'll use agg to peg error sizes and do the math.
Nack
| White played opening 32D | |
|---|---|
| |
| Blue on roll | |
| White played opening 32S | |
|---|---|
| |
| Blue on roll | |
| White played opening 32Z | |
|---|---|
| |
| Blue on roll | |