Post by Paul Epstein
Tim has a website on errors in Robertie's book.
It would be fun to look at the most dramatic examples.
1. According to the bots, what is the greatest checkerplay error among
2. What is the greatest cube-action error?
3. Is a pass ever mistaken for a ND/T (or vice versa)?
4. Is a TG ever mistaken for a D/T (or vice versa)?
It's been too long ago for me to remember the answers to these questions,
and you should also bear in mind that GNU rollouts (which is what I relied
on, in almost all cases) might disagree with XG rollouts. Plus there are
situations where the bot says the Robertie missed a double before the roll;
how "dramatic" is that? Is it more "dramatic" if Robertie doesn't even
mention the bot's top play, or if Robertie ranks the top three plays and
gets them exactly reversed? There is one position where Robertie writes,
"The right play is hitting two." In fact, in that position, there are two
plays that hit two, and the bot's preferred play is the double hit that
Robertie doesn't even consider.
That said, I think there were very few checker plays where GNU thought that
Robertie's play was in the -0.1 range (and again, an XG rollout might yield
a different verdict). There were quite a few in the -0.05 to -0.08 range as
I think the answers to #3 and #4 are no, although I'm not certain. I'm pretty
sure that there were some cube-action errors in whopper-with-cheese range,
but few if any in the double-whopper range.