The topic of adjusting for one's opponent comes up fairly often on this forum, but AFAIK there has been very little concrete analysis of opponent-dependent checker-play decisions. For cube decisions, there has of course been much discussion of fish METs, chouette strategy, reading the opponent, etc. But for checker plays, the closest I've seen is some discussion by opening-play junkies, and even then the discussion is largely focused around adjusting for general differences in skill, rather than adjusting for specific weaknesses that an opponent is known for.
To illustrate that there is a large amount of untapped potential in this subject, I'm going to give a fairly detailed analysis of the position below that arose in a casual game that I played recently (on a real board against a human opponent!).
| White | 88 |
|---|---|
| |
| Blue | 99 |
| GNUBg Id: bWuGAwC27SLAAA:cIk6AAAAAAAA XGID=--BBBBCA--A--c---bBbabbba-:0:0:1:65:0:0:0:1:1 | |
There's a clear conceptual decision here: Blue can either break anchor with something like 18/7, or hold the anchor and play something like 10/4 6/1.
The way most players would "analyze" this position after the game would be to make a pilgrimage to an electronic Delphi and ask the oracle. I won't give the full prophecy here but will just summarize it: 18/7 is about 0.05 back, a blunder. Case closed.
What the oracle didn't know, though, was that I knew that my opponent was very shy about leaving shots in this kind of position. I counted pips, so both of us were aware that the pip count would be dead even after I played my roll. Suppose I were to play 18/7. Then I knew that my opponent would play 8/7* 5/1 with a roll of 41 and 12/7* 5/1 with a roll of 54—he wasn't that paranoid about getting hit back. But with 61 51 55 65, I was sure that he would not hit. (For 65 and 61, I was much more confident of what he would do than of what the bot would do.) The crucial question was, what would he do with rolls that would allow him to hit and cover his 5pt, but leave me with a direct shot from the bar at his blot on the 1pt? 11 21 31 52 53. I felt fairly confident—though not certain—that he would be spooked by the prospect of leaving a direct shot in light of my strong forward position, and would play safe.
If we assume that my opponent won't hit with 11 21 31 52 53, what does that imply about my decision? Clearly, since the race is close, Blue gains by playing 18/7 whenever White fails to hit. Nine fewer shots is a lot; is that enough to swing the decision in favor of 18/7?
There's no way to answer this question definitively, but as an approximation, I first obtained an XGR+ evaluation for each of the twenty-one rolls after each of 10/4 6/1 and 18/7. Then for each roll for which hitting was legal (including, for completeness, 51 55 65, when the bot doesn't hit), I recorded what XGR+ thought were the best hitting and the best non-hitting plays. Then I manually averaged the equities after 18/7 assuming that my opponent would play like the bot did except for 11 21 31 52 53, where I instead inserted XGR+'s favorite non-hitting play.
There are a couple of reasons that this calculation is bogus. First of all, there are some other rolls which I'm not sure my opponent would play the same way that XGR+ does. For example, after 10/4 6/1, I could imagine my opponent using an ace to play 5/4. Secondly, and more importantly, the XGR+ evaluations don't give an accurate estimate of the equity of me versus my opponent. Nevertheless, I think this exercise does give us some interesting information. I found that 18/7 switched from being about 0.05 back to being about 0.03 ahead. So 18/7 could be the right play against this particular opponent.
Unfortunately for the cause of science, my opponent rolled something that didn't allow him to hit, so I never got to find out what he would have done with 11 21 31 52 53.
By the way, against a super-aggressive opponent who would hit with anything, even 51 or 55, the same calculation indicates that 18/7 is still between 0.03 and 0.04 back.