here is my study: lichess.org/study/PVwBFUn5 depth = 88 cloud mode. And the recommendation is wrong. Is your Sf 11 doing the same?
So, among atomic players, these bugs have been known for a while. There have been several regressions in atomic sf over the past year. So far the solution has just been to run an older version of stockfish locally that doesn't have these bugs. When someone measured modern sf for atomic against the strongest sf for atomic from a year ago using an elo meter, the difference was a regression of about 450 rating points. younameit has given an example of one bug that seems to be related to move order that results in check getting priority over blowing up the king. However, there is also a promotion bug that may or may not be related to this bug. A month ago @Gannet posted a position on discord that showcased the promotion bug. Here is Gannet's bug report:
"I think I may have found the bug in modern SF that causes at least some of the super wacky evals
there have been issues with promotion in atomic versions of SF as well as e.p. king explosions in previous versions of SF so I wouldn't be surprised if this was related
regardless, it falsely giving mate in 1 here is probably the most blatant thing I've seen yet
If someone wants to report this on github I suspect this kind of position might be prove very key in finally sorting out some of its eval issues.
also I managed to mess up the move order bar by using "make mainline" in one of these positions and didn't know if anyone had encountered this issue before.
pretty sure 26 < 32 last I checked
for the 1 person wondering, this is the line involved: 1. Nf3 f6 2. Nc3 Nh6 3. g4 c6 4. d4 d6 5. Ne5 dxe5 6. d5 e5 7. d6 Qb6 8. e3 Bf5 9. Ne4 Qd4 10. Bd3 Qa4 11. b3 Qa5+ 12. b4 Qa4 13. d7+ Kd8 14. Bc4 Qxc2 15. Bg8 Bxe4 16. Bxh7 a5 17. h4 axb4 18. Rb1 b5 19. h5 b4 20. h6 gxh6 21. Rh7 Bg7 22. Rh8+ Bf8 23. g5 b3 24. a4 Ra5 25. f4 Rc5 26. Kf2 Rc1 27. Rh7 Bc5 28. Kg2 Rg1+ 29. Kh2 Rg2+ 30. Kh1 exf4 31. Re7 Kc7
I don't remember lichess having many bugs crop up basically at all in the old days. Ever since lichess switched to their v2 codebase it seems like these have just begun appearing. I do have to wonder if part of that is the massive increase in traffic lichess has been getting over the years though as well, and them trying to save on processing power at the expense of making things much more shaky in on their backend."
For convenience, I have imported that pgn to a study. lichess.org/study/ieF8GmdY/bzAhc7KW
In this position sf thinks 32.d8=Q is checkmate, but it doesn't see that the bishop can take the rook and blow up both the rook and the queen checking the king, and then the position is completely winning for black.
P.S.
Something of note is that while skoudone and I were playing around with multi-variant stockfish earlier this year, we discovered that after reverting this patch github.com/ddugovic/Stockfish/commit/a9344db3a9722a6ac900a2f12987b6d8b3428d42 modern atomic sf improved quite a bit, but it was not completely fixed. So while this patch did bug atomic sf, it isn't solely responsible for all of the bugs.
"I think I may have found the bug in modern SF that causes at least some of the super wacky evals
there have been issues with promotion in atomic versions of SF as well as e.p. king explosions in previous versions of SF so I wouldn't be surprised if this was related
regardless, it falsely giving mate in 1 here is probably the most blatant thing I've seen yet
If someone wants to report this on github I suspect this kind of position might be prove very key in finally sorting out some of its eval issues.
also I managed to mess up the move order bar by using "make mainline" in one of these positions and didn't know if anyone had encountered this issue before.
pretty sure 26 < 32 last I checked
for the 1 person wondering, this is the line involved: 1. Nf3 f6 2. Nc3 Nh6 3. g4 c6 4. d4 d6 5. Ne5 dxe5 6. d5 e5 7. d6 Qb6 8. e3 Bf5 9. Ne4 Qd4 10. Bd3 Qa4 11. b3 Qa5+ 12. b4 Qa4 13. d7+ Kd8 14. Bc4 Qxc2 15. Bg8 Bxe4 16. Bxh7 a5 17. h4 axb4 18. Rb1 b5 19. h5 b4 20. h6 gxh6 21. Rh7 Bg7 22. Rh8+ Bf8 23. g5 b3 24. a4 Ra5 25. f4 Rc5 26. Kf2 Rc1 27. Rh7 Bc5 28. Kg2 Rg1+ 29. Kh2 Rg2+ 30. Kh1 exf4 31. Re7 Kc7
I don't remember lichess having many bugs crop up basically at all in the old days. Ever since lichess switched to their v2 codebase it seems like these have just begun appearing. I do have to wonder if part of that is the massive increase in traffic lichess has been getting over the years though as well, and them trying to save on processing power at the expense of making things much more shaky in on their backend."
For convenience, I have imported that pgn to a study. lichess.org/study/ieF8GmdY/bzAhc7KW
In this position sf thinks 32.d8=Q is checkmate, but it doesn't see that the bishop can take the rook and blow up both the rook and the queen checking the king, and then the position is completely winning for black.
P.S.
Something of note is that while skoudone and I were playing around with multi-variant stockfish earlier this year, we discovered that after reverting this patch github.com/ddugovic/Stockfish/commit/a9344db3a9722a6ac900a2f12987b6d8b3428d42 modern atomic sf improved quite a bit, but it was not completely fixed. So while this patch did bug atomic sf, it isn't solely responsible for all of the bugs.
This topic has been archived and can no longer be replied to.