advert: inferring Piece Value in Chess and Chess Variants (incl. Antichess) • page 1/1 • www.antichess.org • lichess.org

Hello fellow antichess nerds;
My preprint on inferring piece value via logistic regression is now on arxiv: https://arxiv.org/abs/2509.04691, with sections on Antichess, Atomic, Horde, 960 and Standard chess. Happy to field any questions,

--Steven

Hello fellow antichess nerds; My preprint on inferring piece value via logistic regression is now on arxiv: https://arxiv.org/abs/2509.04691, with sections on Antichess, Atomic, Horde, 960 and Standard chess. Happy to field any questions, --Steven

shabbychess

Some takeaways:

I had to jump through some technical hoops to use this observational data and come to a causal conclusion.
Inferring quiescent positions at scale is difficult and throws some shade on this endeavor.
Nonetheless, it seems that bishops are worth more than knights in Standard.
Antichess and Atomic are much more positional than Standard and 960 in the sense that piece values are smaller. (This should be obvious to a casual observer.)
I find positive value for kings in the middlegame in Antichess; piece values are generally negative, except rooks in the endgame.
There are still a lot of unanswered questions here, but the methodology has some promise.

Some takeaways: 1. I had to jump through some technical hoops to use this observational data and come to a causal conclusion. 2. Inferring quiescent positions at scale is difficult and throws some shade on this endeavor. 3. Nonetheless, it seems that bishops are worth more than knights in Standard. 4. Antichess and Atomic are much more positional than Standard and 960 in the sense that piece values are smaller. (This should be obvious to a casual observer.) 5. I find _positive_ value for kings in the middlegame in Antichess; piece values are generally negative, except rooks in the endgame. 6. There are still a lot of unanswered questions here, but the methodology has some promise.