Kees recently published the results of his study regarding the FIBS
rating calculator, and one of the things that came up was the disparity
in the weight of cube errors on the MWC as opposed to checker errors.
Many already knew of this, unquestionably, but here was an actual ration
on the importance. I understand some may have reservations on the
absolute preciseness of this, but let's put them aside for argument's
sake.
The idea of doubling according to one's opponent is very old news, and
one outstanding example are the Jacobs/Trice METs for rating differences
of 100 or 50 points. Still, wouldn't it be possible, for the program to
use the FIBS calculator, or even pre-entered ratings values, and
literally adapt the MET double/take points accordingly? Meaning it would
attempt, however fallaciously, to do what humans do, and adapt its
doubling according to the opponent's play. Obviously, if it is trying to
judge the player's play as opposed to working with a pre-known value, it
would be forced to be evaluating the player's ability *while playing*,
and keep a running score. Still, this might be interesting no? If the
bot judged that so far I was playing 200 points below it, it would adapt
it's doubling accordingly, in order to maximize my chances of making
checker play errors.