A curated collection of solid doubles, clear takes, about in the middle of the doubling window. From
a) move 1-5,
b) move 6-10,
c) move 11-15 (perhaps)
Maybe include variations to show minimum modifications needed to turn this into a marginal double or a marginal take.
Perhaps use to automatically generate position types. Perhaps try to obtain a really large set of positions from matches. Maybe do the clustering also on game level: What type of game was this?
Distribution of cube mistakes: Are my doubles closer to no double than to pass? (I.e. on the aggressive side?) How many takes are passes and by how much? Are those distributions different between me and opponents?
Distribution of checker play mistakes: Do we see a fat tail of a few really big mistakes?
Distribution of mistakes by lasse and opponents -- who play better
Size and frequency of opening mistakes?
Count "actual cube action" by "proper cube action" -- how frequent and how big are the possible mistake types? (Sum to compare total contributions).
Is it profitable to double a little too early (do we get passes enough)?
Is it profitable to double slightly too good (do we get takes enough)?
Rare birds: Examples of "too good, take" situations. Examples of really big mistakes.
For a number of position categories (opening, blitzes, holding games, primes, close-outs, ...) display a set of positions each in three variations: Borderline double, solid D/T, borderline take
Take, for example, LH Kval 2024, and compute confidence interval for all the observed PRs. Perhaps use my own matches to estimate standard errors on checker play and cube errors.