In this game: en.lichess.org/XW7YxjNw/black#119, Stockfish is wrongly saying that black's move 60... Kxf6 was a blunder and that Ke6 would have avoided checkmate with an evaluation of +0.1. In fact, it's a simple K+P endgame that white can win easily with 61. f7.
Not a bug
You can read the last few threads for more detail, but basically it intentionally evaluates the first repetition as a draw instead of the 3rd repetition. Or in this case, the best move would have led to the repetition
You can read the last few threads for more detail, but basically it intentionally evaluates the first repetition as a draw instead of the 3rd repetition. Or in this case, the best move would have led to the repetition
I don't see a repetition in this game. But I do think this is sort of out of lichess' hands. They don't develop stockfish. Stockfish is an open source chess engine (one of the strongest available) developed by someone else.
It may be out of their hands, but definitely of interest to lichess to pursue a fix (with stockfish developers) or revert to a previous version of stockfish if that's the problem. I've been seeing this issue quite often recently - the evaluation given after pressing the computer analysis button is completely off.
This issue has been repeatedly raised in the official-stockfish issue tracker and repeatedly dismissed because "fixing" it weakens Stockfish for competitive play.
I have repeatedly implemented threefold repetition in the version of Stockfish uses because for analysis it's useful to know the difference between twofold and threefold repetition; however, upstream changes from official-stockfish repeatedly break threefold repetition. I have once again implemented threefold repetition detection.
It's expensive to manually test for this every time a change is made, and I don't know how to automate a test for it; and official-stockfish don't consider this a bug and they keep developing changes which break this. If someone knows something I don't, please advise!
I have repeatedly implemented threefold repetition in the version of Stockfish uses because for analysis it's useful to know the difference between twofold and threefold repetition; however, upstream changes from official-stockfish repeatedly break threefold repetition. I have once again implemented threefold repetition detection.
It's expensive to manually test for this every time a change is made, and I don't know how to automate a test for it; and official-stockfish don't consider this a bug and they keep developing changes which break this. If someone knows something I don't, please advise!
Yeah I've run into many of those when reading an endgame manual...
One minor note:
"Stockfish is wrongly saying that black's move 60... Kxf6 was a blunder..."
Stockfish doesn't annotate games, but the game annotator heavily relies upon Stockfish evaluations. So while an evaluation of "cp 0" may be counterintuitive, like every evaluation (other than a "mate(d) in N") it's allowed to be subjective.
"Stockfish is wrongly saying that black's move 60... Kxf6 was a blunder..."
Stockfish doesn't annotate games, but the game annotator heavily relies upon Stockfish evaluations. So while an evaluation of "cp 0" may be counterintuitive, like every evaluation (other than a "mate(d) in N") it's allowed to be subjective.
Toadofsky, it's worth noting that the examples I've seen (and the one I posted above) don't have to do with three-fold repetition.
@klooth
But the best move according to Stockfish leads to a repetition, so it counts it as one anyway
But the best move according to Stockfish leads to a repetition, so it counts it as one anyway
This topic has been archived and can no longer be replied to.