Black to play and Draw, which engine can solve this Puzzle ?

[Modern Times]

Strange that Stockfish quickly settles on a 0.00 score, but then later loses it. It is only possible for engines to see this quickly when they recognize the 'rook-pawn plus wrong bishop' pattern (possibly through an EGT). Not only pure, but also with additional pawns in the same file.

It is strange that Stockfish even when it reach this position it does NOT immediately recognize that it is a dead draw, by applying the Opposition principle, since it does NOT have the Black Square Bishop which is needed to support the promotion of its h pawn to h8 = Queen, since with the White Square Bishop it is impossible to win with two pawns and the wrong color Bishop

[fen]8/8/8/6k1/8/5B1b/6PP/5K2 w - - 0 1 [/fen]
FEN: 8/8/8/6k1/8/5B1b/6PP/5K2 w - - 0 1

Immediately recognises a draw here:

Engine: Stockfish 15 64-bit (4096 MB)
by the Stockfish developers (see AUTHORS f
Found 510 tablebases
Found 510 tablebases
NNUE evaluation using nn-6877cd24400e.nnue enabled
44/67 0:00 0.00 1.gxh3 Kh6 2.Be2 Kg5 3.Bg4 Kh6 4.Kg1 Kg6 5.Kg2 Kg7 6.Bd1 Kh8 7.Bc2 Kg8 8.Kg3 Kf8 9.Kg4 Kg7 10.Kg5 Kg8 11.Kf6 Kh8 12.Kg6 Kg8 13.Bb3+ Kh8 14.Kg5 (25.332.406) 49768 TB:14
45/70 0:04 0.00 1.gxh3 Kh6 2.Be2 Kg5 3.Bg4 Kh6 4.Kg1 Kg6 5.Kg2 Kg7 6.Bd1 Kh8 7.Bc2 Kg8 8.Kg3 Kf8 9.Kg4 Kg7 10.Kg5 Kg8 11.Kf6 Kh8 12.Kg6 Kg8 13.Bb3+ Kh8 14.Kg5 (216.159.699) 46247 TB:14
46/72 1:08 0.00 1.gxh3 Kh6 2.Be2 Kg5 3.Bg4 Kh6 4.Kg1 Kh7 5.Bf5+ Kh8 6.Bc2 Kg7 7.h4 Kh8 8.Be4 Kg7 9.Bf5 Kf7 10.Bc2 Kg7 11.Kg2 Kh6 12.Bd3 Kg7 13.Bb1 Kf6 14.Bh7 (3.217.954.735) 46713 TB:14
47/72 1:11 0.00 1.gxh3 Kh6 2.Be2 Kg5 3.Bg4 Kh6 4.Kg1 Kh7 5.Bf5+ Kh8 6.Bc2 Kg7 7.h4 Kh8 8.Be4 Kg7 9.Bf5 Kf7 10.Bc2 Kg7 11.Kg2 Kh6 12.Bd3 Kg7 13.Bb1 Kf6 14.Bh7 (3.370.994.604) 46821 TB:14
best move: g2xh3 time: 1:12.125 min n/s: 46.821.320 nodes: 3.370.994.604 TB: 14

[Uri Blass]

It is clear that berserk without tablebases lost knowledge
[fen]5N1k/7P/8/8/8/8/8/5K2 b - - 0 1[/fen]

Berserk-9-x64-avx2-pext:
time -1 start 104420953 alloc 0 max 2147483647 depth 127 timeset 0 searchmoves 0
1/1 00:00 28k 28,130k -256.04 Kh8-g7

With the same position

Berserk-8.5.1-x64-avx2-pext:
time -1 start 104529015 alloc 0 max 0 depth 99 timeset 0 searchmoves 0
1/12 00:00 6k 5,538k -0.89 Kh8-g7
2/13 00:00 8k 8,314k -0.88 Kh8-g7 Kf1-f2
3/17 00:00 10k 10,077k -0.88 Kh8-g7 Kf1-g1 Kg7-h8
4/18 00:00 12k 11,965k -0.88 Kh8-g7 Kf1-g1 Kg7-h8 Kg1-g2
5/20 00:00 14k 13,784k -0.88 Kh8-g7
6/20 00:00 16k 15,659k -0.16 Kh8-g7
7/21 00:00 18k 17,968k -0.66 Kh8-g7 Kf1-g1 Kg7-h8 Kg1-g2 Kh8-g7 Kg2-f2 Kg7-h8
8/29 00:00 21k 21,426k -0.96 Kh8-g7 Kf1-f2 Kg7-h8 Kf2-e3 Kh8-g7 Ke3-f3 Kg7-h8 Kf3-g3 Kh8-g7

[peter]

Strange that Stockfish quickly settles on a 0.00 score, but then later loses it. It is only possible for engines to see this quickly when they recognize the 'rook-pawn plus wrong bishop' pattern (possibly through an EGT). Not only pure, but also with additional pawns in the same file.

Who says, SF loses this?
30 threads of 16x3.5GHz CPU, 3070ti GPU, SF Syzygy Probe Limit=0, GUI- tbs off, LC0 with 6men, so SF plays practically against tbs from 3rd move onwards:
[pgn][Event "AMD, Blitz 10m+5s"]
[Site "DESKTOP-9QB5AU4"]
[Date "2022.06.18"]
[Round "1"]
[White "Lc0 v0.30.0-dev+git.77b5659"]
[Black "Stockfish 070622"]
[Result "1/2-1/2"]
[Annotator "0.01;2.37"]
[SetUp "1"]
[FEN "r7/8/8/5bk1/8/5B2/5RPP/6K1 b - - 0 1"]
[PlyCount "52"]
[TimeControl "600+5"]

1... Ra1+ {2.37/51 37} 2. Rf1 {0.01/3 20} Rxf1+ {2.37/55 51} 3. Kxf1 {0.00/1 0}
Bh3 {2.37/60 31} 4. gxh3 {0.00/0 0} Kg6 {2.37/54 59} 5. h4 {0.00/0 0 (Be4+)}
Kg7 {2.37/49 58} 6. h5 {0.00/0 0 (Kg2)} Kh7 {2.02/37 19} 7. Ke2 {0.00/0 0 (Kg2)
} Kg7 {2.54/52 67} 8. Be4 {0.00/0 0 (Bg4)} Kh6 {2.37/44 25} 9. Bg6 {0.00/0 0}
Kg7 {2.17/40 27} 10. Kd3 {0.00/0 0 (Kf3)} Kf6 {2.50/48 45} 11. Kc4 {0.00/0 0
(Ke3)} Kg7 {2.38/49 14} 12. Kc5 {0.00/0 0 (Kd4)} Kf6 {2.21/42 13} 13. Kc4 {0.
00/0 0 (Kd4)} Kg7 {2.36/34 26} 14. Kd4 {0.00/0 0 (Kd5)} Kg8 {2.23/42 13} 15. h6
{0.00/0 0 (h3)} Kh8 {2.20/47 15} 16. Be4 {0.00/0 0 (Bd3)} Kg8 {2.20/25 1} 17.
h7+ {0.00/0 0 (Bg6)} Kg7 {0.83/36 16} 18. Bg6 {0.00/0 0 (Kd5)} Kh8 {0.80/37 13}
19. Kc3 {0.00/0 0 (Kc4)} Kg7 {0.37/20 1} 20. Kd4 {0.00/0 0} Kh8 {0.84/44 116}
21. Kc5 {0.00/0 0 (Ke3)} Kg7 {0.80/22 1} 22. Kb4 {0.00/0 0} Kh8 {0.84/40 11}
23. Ka4 {0.00/0 0} Kg7 {0.84/31 1} 24. Ka3 {0.00/0 0} Kh8 {0.84/34 49} 25. Ka4
{0.00/0 0 (h4)} Kg7 {0.41/21 1} 26. Ka3 {0.00/0 0 (Be4)} Kh8 {1.07/40 18} 27.
Ka4 {0.00/0 0 Draw accepted} 1/2-1/2[/pgn]

[hgm]

Strange that Stockfish quickly settles on a 0.00 score, but then later loses it. It is only possible for engines to see this quickly when they recognize the 'rook-pawn plus wrong bishop' pattern (possibly through an EGT). Not only pure, but also with additional pawns in the same file.

Who says, SF loses this?
30 threads of 16x3.5GHz CPU, 3070ti GPU, SF Syzygy Probe Limit=0, GUI- tbs off, LC0 with 6men, so SF plays practically against tbs from 3rd move onwards:
[pgn][Event "AMD, Blitz 10m+5s"]
[Site "DESKTOP-9QB5AU4"]
[Date "2022.06.18"]
[Round "1"]
[White "Lc0 v0.30.0-dev+git.77b5659"]
[Black "Stockfish 070622"]
[Result "1/2-1/2"]
[Annotator "0.01;2.37"]
[SetUp "1"]
[FEN "r7/8/8/5bk1/8/5B2/5RPP/6K1 b - - 0 1"]
[PlyCount "52"]
[TimeControl "600+5"]

1... Ra1+ {2.37/51 37} 2. Rf1 {0.01/3 20} Rxf1+ {2.37/55 51} 3. Kxf1 {0.00/1 0}
Bh3 {2.37/60 31} 4. gxh3 {0.00/0 0} Kg6 {2.37/54 59} 5. h4 {0.00/0 0 (Be4+)}
Kg7 {2.37/49 58} 6. h5 {0.00/0 0 (Kg2)} Kh7 {2.02/37 19} 7. Ke2 {0.00/0 0 (Kg2)
} Kg7 {2.54/52 67} 8. Be4 {0.00/0 0 (Bg4)} Kh6 {2.37/44 25} 9. Bg6 {0.00/0 0}
Kg7 {2.17/40 27} 10. Kd3 {0.00/0 0 (Kf3)} Kf6 {2.50/48 45} 11. Kc4 {0.00/0 0
(Ke3)} Kg7 {2.38/49 14} 12. Kc5 {0.00/0 0 (Kd4)} Kf6 {2.21/42 13} 13. Kc4 {0.
00/0 0 (Kd4)} Kg7 {2.36/34 26} 14. Kd4 {0.00/0 0 (Kd5)} Kg8 {2.23/42 13} 15. h6
{0.00/0 0 (h3)} Kh8 {2.20/47 15} 16. Be4 {0.00/0 0 (Bd3)} Kg8 {2.20/25 1} 17.
h7+ {0.00/0 0 (Bg6)} Kg7 {0.83/36 16} 18. Bg6 {0.00/0 0 (Kd5)} Kh8 {0.80/37 13}
19. Kc3 {0.00/0 0 (Kc4)} Kg7 {0.37/20 1} 20. Kd4 {0.00/0 0} Kh8 {0.84/44 116}
21. Kc5 {0.00/0 0 (Ke3)} Kg7 {0.80/22 1} 22. Kb4 {0.00/0 0} Kh8 {0.84/40 11}
23. Ka4 {0.00/0 0} Kg7 {0.84/31 1} 24. Ka3 {0.00/0 0} Kh8 {0.84/34 49} 25. Ka4
{0.00/0 0 (h4)} Kg7 {0.41/21 1} 26. Ka3 {0.00/0 0 (Be4)} Kh8 {1.07/40 18} 27.
Ka4 {0.00/0 0 Draw accepted} 1/2-1/2[/pgn]

'Loses it' = loses the 0.00 score and reports some non-zero value.

[hgm]

Strange that Stockfish quickly settles on a 0.00 score, but then later loses it. It is only possible for engines to see this quickly when they recognize the 'rook-pawn plus wrong bishop' pattern (possibly through an EGT). Not only pure, but also with additional pawns in the same file.

It is strange that Stockfish even when it reach this position it does NOT immediately recognize that it is a dead draw, by applying the Opposition principle, since it does NOT have the Black Square Bishop which is needed to support the promotion of its h pawn to h8 = Queen, since with the White Square Bishop it is impossible to win with two pawns and the wrong color Bishop

[fen]8/8/8/6k1/8/5B1b/6PP/5K2 w - - 0 1 [/fen]
FEN: 8/8/8/6k1/8/5B1b/6PP/5K2 w - - 0 1

Immediately recognises a draw here:

Engine: Stockfish 15 64-bit (4096 MB)
by the Stockfish developers (see AUTHORS f
Found 510 tablebases
Found 510 tablebases
NNUE evaluation using nn-6877cd24400e.nnue enabled
44/67 0:00 0.00 1.gxh3 Kh6 2.Be2 Kg5 3.Bg4 Kh6 4.Kg1 Kg6 5.Kg2 Kg7 6.Bd1 Kh8 7.Bc2 Kg8 8.Kg3 Kf8 9.Kg4 Kg7 10.Kg5 Kg8 11.Kf6 Kh8 12.Kg6 Kg8 13.Bb3+ Kh8 14.Kg5 (25.332.406) 49768 TB:14
45/70 0:04 0.00 1.gxh3 Kh6 2.Be2 Kg5 3.Bg4 Kh6 4.Kg1 Kg6 5.Kg2 Kg7 6.Bd1 Kh8 7.Bc2 Kg8 8.Kg3 Kf8 9.Kg4 Kg7 10.Kg5 Kg8 11.Kf6 Kh8 12.Kg6 Kg8 13.Bb3+ Kh8 14.Kg5 (216.159.699) 46247 TB:14
46/72 1:08 0.00 1.gxh3 Kh6 2.Be2 Kg5 3.Bg4 Kh6 4.Kg1 Kh7 5.Bf5+ Kh8 6.Bc2 Kg7 7.h4 Kh8 8.Be4 Kg7 9.Bf5 Kf7 10.Bc2 Kg7 11.Kg2 Kh6 12.Bd3 Kg7 13.Bb1 Kf6 14.Bh7 (3.217.954.735) 46713 TB:14
47/72 1:11 0.00 1.gxh3 Kh6 2.Be2 Kg5 3.Bg4 Kh6 4.Kg1 Kh7 5.Bf5+ Kh8 6.Bc2 Kg7 7.h4 Kh8 8.Be4 Kg7 9.Bf5 Kf7 10.Bc2 Kg7 11.Kg2 Kh6 12.Bd3 Kg7 13.Bb1 Kf6 14.Bh7 (3.370.994.604) 46821 TB:14
best move: g2xh3 time: 1:12.125 min n/s: 46.821.320 nodes: 3.370.994.604 TB: 14

That is because you use EGT, and not by virtue of the engine's evaluation.

[Modern Times]

That is because you use EGT, and not by virtue of the engine's evaluation.

OK, well doesn't every serious tester or analyst use 5-men at a minimum ? Standard practice these days. Their use is so widespread that I think authors are entitled to assume their presence. Having said that, Stefan Pohl is one exception according to his website.

[Chessqueen]

That is because you use EGT, and not by virtue of the engine's evaluation.

OK, well doesn't every serious tester or analyst use 5-men at a minimum ? Standard practice these days. Their use is so widespread that I think authors are entitled to assume their presence. Having said that, Stefan Pohl is one exception according to his website.

My personal opinion is that if you really want to test the strength of engines it should be done with no more than 6 pieces EGTB. I will explain why now, so far most testers have used at least 5-men at a minimum, but once we get our hand on the 8-pieces EGTB, a lot of strong engines or should I say the top 6 will lose Elo, because of one simple reason, there would be lot and lots of engines in the range of 2700 to 3000 that will get to use 8 pieces endgame and all that those weaker engines have to do is play a few more moves from those long opening up to 12 or 16 moves, and Stockfish, or Komodo Dragon, and Lela etc... will only be able to get draws versus those weaker engines. Therefore, all the Knowledge that Programmers spend years to implement into their engine would be for nothing, since without 6-pieces, 7 or 8 pieces EGTB the more knowledgeable engines should little by little squeeze a win from the weaker engines if the weaker engines, and it is really not that hard for engines in the range between 2700 thru 3000 to get a balance position against stockfish or Komodo after the opening that get those weaker engine up to 12 moves or most of the times, so with a few more moves from the weaker engines they should get a balance position or slightly worse that guarantee them a draw against the best and engines with more knowledge to get a balance position or slightly worse.

NOTE it would be like if Carsen face GM Pragg 16 years old, and GM Pragg is booked up to 18 moves at least, after a few more moves GM Pragg will obtain a position that it is very hard for GM Carlsen to get an edge, but if they would allow GM Pragg to use EGTB like in the cases of engine vs engine
GM Pragg can walk easy win at least a draw. Which is NOT fair for GM Carlsen or Stockfish in the case of Engine vs Engine to allow 8-pieces EGTB in the future.

[Ajedrecista]

Hello:

I had seen this problem before, so I know the solution, but I did not know that the problem was that old! Googling the position retrieves a problem of 1623 by Gioachino Greco. It was difficult to find his manuscripts, but I succeeded with one of them, thanks to the title given at Wikipedia article about Greco.

Thank you very much to the Cleveland Public Library Digital Gallery, which hosts a copy of Trattato del Nobilissimo et Militare Essercitio de Scacchi nel Quale si Contengono Molti Bellissimi Tratti et la Vera Scienza di Esso Gioco. Composto da Gioachino Greco Calabrese. I was lucky enough to find the problem without diagrammes and in a foreign language to me (Italian). It starts at the end of this page:

https://cplorg.contentdm.oclc.org/digit ... 20/id/9597

The setup of the white pieces starts at:

B. Re alla casa di suo C.
B. Ro di suo Re alla 2ª casa di A. di suo Re.

And continues through the next page:

B. A. di suo Re alla 3ª sua
B. P. di C. di suo Re alla sua casa.
B. P. di Ro. di suo Re alla sua casa.
____________________________________
N. Re alla 4ª casa di suo C.
N. A. di sua D. alla 4ª casa di A. di suo Re.
N. Ro. di sua D. alla sua casa.

'Black will draw against white':

Il nero farà patta al bianco.

After the first moves, the key move ..., Bh3 is written as it follows:

N. A. di sua D. alla 3ª casa di Ro. di Re con.

Then: gxh3 (PxB), black king goes to h8 and draw:

B. P. di C. di suo Re p. A. con.
N. Re alla casa di suo Ro. et farà patta al bianco.

Some typos are expected because I do not know Italian language.







Regards from Spain.

Ajedrecista.

[Modern Times]

OK, well doesn't every serious tester or analyst use 5-men at a minimum ? Standard practice these days. Their use is so widespread that I think authors are entitled to assume their presence. Having said that, Stefan Pohl is one exception according to his website.

My personal opinion is that if you really want to test the strength of engines it should be done with no more than 6 pieces EGTB.

Agreed, but do any serious testers not use egtb at all, and if not why not ? And it is inconceivable that someone who wanted to analyse a position would not use egtb at all.

[peter]

Hello:

I had seen this problem before, so I know the solution, but I did not know that the problem was that old! Googling the position retrieves a problem of 1623 by Gioachino Greco. It was difficult to find his manuscripts, but I succeeded with one of them, thanks to the title given at Wikipedia article about Greco.

Thank you very much to the Cleveland Public Library Digital Gallery, which hosts a copy of Trattato del Nobilissimo et Militare Essercitio de Scacchi nel Quale si Contengono Molti Bellissimi Tratti et la Vera Scienza di Esso Gioco. Composto da Gioachino Greco Calabrese. I was lucky enough to find the problem without diagrammes and in a foreign language to me (Italian). It starts at the end of this page:

https://cplorg.contentdm.oclc.org/digit ... 20/id/9597

Great, thanks a lot regards