Have you decided why Vas

[bob]

george, this is a little off topic, but can you tell me the formula for converting rybka evalaution to win percentage, as you did in your post

There is only one way to do this, and it is not going to be that accurate. You simply record the evaluation at each point in the game. And then record the game result. And then you repeat for lots of games. And then measure how well an evaluation of +XXX correlates with a game result of 1-0. And the variance will be huge. I watched a game between Rybka 3 and some hiarcs version on ICC a few nights ago where the operator said R3 had an evaluation of +2.3, and the game ended in a draw. Not the first nor last time that will happen. So clearly, Larry's numbers are the result of a curve-fit, with an immense error range. If you were to play Rybka against much weaker opponents, then +.3 might correlate almost 100% with winning chances, And you might find that -.2 correlates with 55% winning chances.

That is the kind of evaluation we would all like to have. But we are a _long_ way (as in decades) away from being able to predict outcomes that precisely, when you see programs reach +8 and draw, or -5 and draw, and all the other improbable events that happen in comp vs comp games.

Most programmers would say, and believe, that +eval means their winning chances are above 50%. Otherwise the evaluation would be useless. But most of us stop at that point and don't try to extrapolate winning chances for specific eval values, because we know that the answer could be quite accurate when looked at over thousands of games, but for a single game, it would be little more than a guess.

[bob]

I don't discuss with engine fanboys who feel personal attacked when someone points out certain flaws in a program.

Please tell me what is your ELO...you probably must be GM strenght to assume that Rybka and Toga are stupid in this position..

Why does that need to be so personal? To me, he appears to be a strong chessplayer. And a strong chessplayer would be interested in that particular aspect of computer chess. I could post dozens of positions where a human glances at the board and says "white is winning here" and a computer evaluation (no search) says "black is winning easily".

What strikes most good chessplayers that are not familiar with computers (and I know how this happens because I have spent way too many hours on the phone with GM Dzhindi) is that a program can play like a genius in many positions, and play like a complete bonehead in a few others. Where human GMs do not play like boneheads in any position except when they make one of their tactical oversights. But even then, they can quickly go back and make a solid assessment of who is winning, where a computer can't.

Good chess players simply don't see how one can be a genius in these positions, and an idiot in these positions, and correlate that to a "super-GM program"...

[George Tsavdaris]

george, this is a little off topic, but can you tell me the formula for converting rybka evalaution to win percentage, as you did in your post

There is only one way to do this, and it is not going to be that accurate. You simply record the evaluation at each point in the game. And then record the game result. And then you repeat for lots of games. And then measure how well an evaluation of +XXX correlates with a game result of 1-0.

Your method is what i would think myself too, but although L.Kaufman did not wrote many details about his method and i didn't understand it clearly, i think he did it with another way.
Here is how he describes he did it: LINK....

If you were to play Rybka against much weaker opponents, then +.3 might correlate almost 100% with winning chances, And you might find that -.2 correlates with 55% winning chances.

Larry obviously assumes perfect practical today play. In that case this means Rybka 3 for now.

A 0.02 obviously means 50.86% of winning when playing with another Rybka 3. If it plays with a weaker program obviously the only value the 50.86% has is that the probability of losing must be larger than 50.86%.

[George Tsavdaris]

george, this is a little off topic, but can you tell me the formula for converting rybka evalaution to win percentage, as you did in your post

I have derived some tables according to some Larry's statements about that:
Here are the tables and the method for calculating them.

[Dr.Ex]

You have no clue what you talk about here. What has this position to do with Botvinnik's Games?

I have. Have you ever studied Botvinnik's games on French.
This position is about equal. The best player will win.
Black has all the counterplay he typically has in such structures (pressing center with f6 and or g5).

Continue to make a fool out of yourself.
This is not a French type position at all. This is sort of a Caro Kann Advance type position where black played Nc6 instead of c6 and castled long. In the French the bishop is passiv on c8.
The position is objectively lost for black.

I smashed Rybka 3 from this position on the playchess server in a 3 minute game but had to agree to a draw in a totally won position or I would have lost the game on time.

[Event "Wertungspartie, 3m + 0s"]
[Site "Maschinenraum"]
[Date "2008.07.25"]
[Round "?"]
[White "Mischa"]
[Black "Kapaun"]
[Result "1/2-1/2"]
[ECO "B00"]
[WhiteElo "2120"]
[BlackElo "2591"]
[Annotator "8.07;0.04"]
[PlyCount "82"]
[EventDate "2008.07.25"]
[TimeControl "180"]

{Rybka 3 32-bit server: 13.4 ply; 97kN/s Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.
40GHz 2400MHz, (4 threads)} 1. e4 {1} Nc6 {0.04/12 4} 2. d4 {(Sc3) 2} d5 {
0.04/13 5} 3. e5 {1} Bf5 {-0.01/14 5} 4. c3 {(Sf3) 4} Qd7 {-0.01/11 4} 5. Bb5 {
(Sf3) 9} f6 {-0.25/12 3} 6. f4 {(Sf3) 11} a6 {-0.32/11 2} 7. Ba4 {(Ld3) 4} Nh6
{-0.42/12 7} 8. Nf3 {4} e6 {-0.30/13 2} 9. O-O {11} Be7 {-0.23/13 2} 10. Be3 {
(Sa3) 4} O-O-O {-0.44/11 3} 11. b4 {(h3) 5} Kb8 {-0.50/10 2} 12. b5 {(h3) 8}
axb5 {-0.50/8 0} 13. Bxb5 {2} Be4 {-0.22/11 4} 14. Nbd2 {8} Ng4 {0.17/12 9} 15.
Bf2 {(De2) 6} b6 {0.10/9 3} 16. Qe2 {(Da4) 11} Nxf2 {0.18/10 18} 17. Rxf2 {2}
Bxf3 {0.91/10 4} 18. Nxf3 {5} Ba3 {1.10/12 4} 19. Rb1 {(De3) 11} Rhf8 {
1.57/10 8} 20. Rb3 {(c4) 2} Be7 {1.32/8 1} 21. a4 {1} g5 {2.23/10 8} 22. a5 {
(exf6) 3} g4 {1.12/8 1} 23. Ne1 {3} Rg8 {1.55/10 8} 24. Qa2 {(Sd3) 5} Nb4 {
3.52/8 2} 25. Rxb4 {11} Qc8 {6.83/11 0} 26. Rb3 {(axb6) 6} fxe5 {4.53/9 2} 27.
fxe5 {(Lc6) 2} Rg7 {5.36/8 5} 28. Bd3 {(axb6) 2} g3 {6.14/10 9} 29. hxg3 {5}
Qb7 {2} 30. axb6 {8.07/9 0} c6 {8.05/9 2} 31. Qa7+ {(La6) 3} Qxa7 {1} 32. bxa7+
{5.50/11 0} Ka8 {5.72/12 2} 33. Ra2 {(Tfb2) 3} Bg5 {5.16/10 2} 34. Nf3 {(Tb6) 3
} Be3+ {5.02/10 1} 35. Kf1 {1} Rxa7 {5.12/12 1} 36. Rxa7+ {2} Kxa7 {5.32/14 0}
37. Ke2 {1} Bh6 {5.31/12 0} 38. Ra3+ {(Lxh7) 1} Kb7 {4.96/11 1} 39. Rb3+ {
(Lxh7) 0} Ka7 {0.00/9 0} 40. Ra3+ {(Lxh7) 0} Kb7 {0.00/26 1} 41. Rb3+ {(Lxh7) 0
} Ka7 {(Lag: Av=0.15s, max=0.4s) 0.00/27 1} 1/2-1/2

[George Tsavdaris]

You have no clue what you talk about here. What has this position to do with Botvinnik's Games?

I have. Have you ever studied Botvinnik's games on French.
This position is about equal. The best player will win.
Black has all the counterplay he typically has in such structures (pressing center with f6 and or g5).

Continue to make a fool out of yourself.
This is not a French type position at all. This is sort of a Caro Kann Advance type position where black played Nc6 instead of c6 and castled long. In the French the bishop is passiv on c8.
The position is objectively lost for black.

I smashed Rybka 3 from this position on the playchess server in a 3 minute game but had to agree to a draw in a totally won position or I would have lost the game on time.

You can't say that you smashed Rybka 3 since the end result is only a draw. Time is a factor of the game so you can't ignore it.

For example if i spend too much time to find good moves while my opponent doesn't and i reach a good position against him but fail to win and only draw because of time pressure, then i can't claim i was unlucky or that i crushed my opponent. I should have been more careful in time handling.
Also my opponent can claim too that if he decided to play more slowly him too he would not reach a worse position.

[Event "Wertungspartie, 3m + 0s"]
[Site "Maschinenraum"]
[Date "2008.07.25"]
[Round "?"]
[White "Mischa"]
[Black "Kapaun"]
[Result "1/2-1/2"]
[ECO "B00"]
[WhiteElo "2120"]
[BlackElo "2591"]
[Annotator "8.07;0.04"]
[PlyCount "82"]
[EventDate "2008.07.25"]
[TimeControl "180"]

{Rybka 3 32-bit server: 13.4 ply; 97kN/s Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.
40GHz 2400MHz, (4 threads)} 1. e4 {1} Nc6 {0.04/12 4} 2. d4 {(Sc3) 2} d5 {
0.04/13 5} 3. e5 {1} Bf5 {-0.01/14 5} 4. c3 {(Sf3) 4} Qd7 {-0.01/11 4} 5. Bb5 {
(Sf3) 9} f6 {-0.25/12 3} 6. f4 {(Sf3) 11} a6 {-0.32/11 2} 7. Ba4 {(Ld3) 4} Nh6
{-0.42/12 7} 8. Nf3 {4} e6 {-0.30/13 2} 9. O-O {11} Be7 {-0.23/13 2} 10. Be3 {
(Sa3) 4} O-O-O {-0.44/11 3} 11. b4 {(h3) 5} Kb8 {-0.50/10 2} 12. b5 {(h3) 8}
axb5 {-0.50/8 0} 13. Bxb5 {2} Be4 {-0.22/11 4} 14. Nbd2 {8} Ng4 {0.17/12 9} 15.
Bf2 {(De2) 6} b6 {0.10/9 3} 16. Qe2 {(Da4) 11} Nxf2 {0.18/10 18} 17. Rxf2 {2}
Bxf3 {0.91/10 4} 18. Nxf3 {5} Ba3 {1.10/12 4} 19. Rb1 {(De3) 11} Rhf8 {
1.57/10 8} 20. Rb3 {(c4) 2} Be7 {1.32/8 1} 21. a4 {1} g5 {2.23/10 8} 22. a5 {
(exf6) 3} g4 {1.12/8 1} 23. Ne1 {3} Rg8 {1.55/10 8} 24. Qa2 {(Sd3) 5} Nb4 {
3.52/8 2} 25. Rxb4 {11} Qc8 {6.83/11 0} 26. Rb3 {(axb6) 6} fxe5 {4.53/9 2} 27.
fxe5 {(Lc6) 2} Rg7 {5.36/8 5} 28. Bd3 {(axb6) 2} g3 {6.14/10 9} 29. hxg3 {5}
Qb7 {2} 30. axb6 {8.07/9 0} c6 {8.05/9 2} 31. Qa7+ {(La6) 3} Qxa7 {1} 32. bxa7+
{5.50/11 0} Ka8 {5.72/12 2} 33. Ra2 {(Tfb2) 3} Bg5 {5.16/10 2} 34. Nf3 {(Tb6) 3
} Be3+ {5.02/10 1} 35. Kf1 {1} Rxa7 {5.12/12 1} 36. Rxa7+ {2} Kxa7 {5.32/14 0}
37. Ke2 {1} Bh6 {5.31/12 0} 38. Ra3+ {(Lxh7) 1} Kb7 {4.96/11 1} 39. Rb3+ {
(Lxh7) 0} Ka7 {0.00/9 0} 40. Ra3+ {(Lxh7) 0} Kb7 {0.00/26 1} 41. Rb3+ {(Lxh7) 0
} Ka7 {(Lag: Av=0.15s, max=0.4s) 0.00/27 1} 1/2-1/2

You were playing alone or as a centaur with an engine's assistance?

[Dr.Ex]

You have no clue what you talk about here. What has this position to do with Botvinnik's Games?

I have. Have you ever studied Botvinnik's games on French.
This position is about equal. The best player will win.
Black has all the counterplay he typically has in such structures (pressing center with f6 and or g5).

Continue to make a fool out of yourself.
This is not a French type position at all. This is sort of a Caro Kann Advance type position where black played Nc6 instead of c6 and castled long. In the French the bishop is passiv on c8.
The position is objectively lost for black.

I smashed Rybka 3 from this position on the playchess server in a 3 minute game but had to agree to a draw in a totally won position or I would have lost the game on time.

You can't say that you smashed Rybka 3 since the end result is only a draw. Time is a factor of the game so you can't ignore it.

Of course I can say that. The end result is irrelevant.


For example if i spend too much time to find good moves while my opponent doesn't and i reach a good position against him but fail to win and only draw because of time pressure, then i can't claim i was unlucky or that i crushed my opponent.

I have not claimed that I was unlucky.

I should have been more careful in time handling.
Also my opponent can claim too that if he decided to play more slowly him too he would not reach a worse position.

[Event "Wertungspartie, 3m + 0s"]
[Site "Maschinenraum"]
[Date "2008.07.25"]
[Round "?"]
[White "Mischa"]
[Black "Kapaun"]
[Result "1/2-1/2"]
[ECO "B00"]
[WhiteElo "2120"]
[BlackElo "2591"]
[Annotator "8.07;0.04"]
[PlyCount "82"]
[EventDate "2008.07.25"]
[TimeControl "180"]

{Rybka 3 32-bit server: 13.4 ply; 97kN/s Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.
40GHz 2400MHz, (4 threads)} 1. e4 {1} Nc6 {0.04/12 4} 2. d4 {(Sc3) 2} d5 {
0.04/13 5} 3. e5 {1} Bf5 {-0.01/14 5} 4. c3 {(Sf3) 4} Qd7 {-0.01/11 4} 5. Bb5 {
(Sf3) 9} f6 {-0.25/12 3} 6. f4 {(Sf3) 11} a6 {-0.32/11 2} 7. Ba4 {(Ld3) 4} Nh6
{-0.42/12 7} 8. Nf3 {4} e6 {-0.30/13 2} 9. O-O {11} Be7 {-0.23/13 2} 10. Be3 {
(Sa3) 4} O-O-O {-0.44/11 3} 11. b4 {(h3) 5} Kb8 {-0.50/10 2} 12. b5 {(h3) 8}
axb5 {-0.50/8 0} 13. Bxb5 {2} Be4 {-0.22/11 4} 14. Nbd2 {8} Ng4 {0.17/12 9} 15.
Bf2 {(De2) 6} b6 {0.10/9 3} 16. Qe2 {(Da4) 11} Nxf2 {0.18/10 18} 17. Rxf2 {2}
Bxf3 {0.91/10 4} 18. Nxf3 {5} Ba3 {1.10/12 4} 19. Rb1 {(De3) 11} Rhf8 {
1.57/10 8} 20. Rb3 {(c4) 2} Be7 {1.32/8 1} 21. a4 {1} g5 {2.23/10 8} 22. a5 {
(exf6) 3} g4 {1.12/8 1} 23. Ne1 {3} Rg8 {1.55/10 8} 24. Qa2 {(Sd3) 5} Nb4 {
3.52/8 2} 25. Rxb4 {11} Qc8 {6.83/11 0} 26. Rb3 {(axb6) 6} fxe5 {4.53/9 2} 27.
fxe5 {(Lc6) 2} Rg7 {5.36/8 5} 28. Bd3 {(axb6) 2} g3 {6.14/10 9} 29. hxg3 {5}
Qb7 {2} 30. axb6 {8.07/9 0} c6 {8.05/9 2} 31. Qa7+ {(La6) 3} Qxa7 {1} 32. bxa7+
{5.50/11 0} Ka8 {5.72/12 2} 33. Ra2 {(Tfb2) 3} Bg5 {5.16/10 2} 34. Nf3 {(Tb6) 3
} Be3+ {5.02/10 1} 35. Kf1 {1} Rxa7 {5.12/12 1} 36. Rxa7+ {2} Kxa7 {5.32/14 0}
37. Ke2 {1} Bh6 {5.31/12 0} 38. Ra3+ {(Lxh7) 1} Kb7 {4.96/11 1} 39. Rb3+ {
(Lxh7) 0} Ka7 {0.00/9 0} 40. Ra3+ {(Lxh7) 0} Kb7 {0.00/26 1} 41. Rb3+ {(Lxh7) 0
} Ka7 {(Lag: Av=0.15s, max=0.4s) 0.00/27 1} 1/2-1/2

You were playing alone or as a centaur with an engine's assistance?

[bob]

george, this is a little off topic, but can you tell me the formula for converting rybka evalaution to win percentage, as you did in your post

There is only one way to do this, and it is not going to be that accurate. You simply record the evaluation at each point in the game. And then record the game result. And then you repeat for lots of games. And then measure how well an evaluation of +XXX correlates with a game result of 1-0.

Your method is what i would think myself too, but although L.Kaufman did not wrote many details about his method and i didn't understand it clearly, i think he did it with another way.
Here is how he describes he did it: LINK....

If you were to play Rybka against much weaker opponents, then +.3 might correlate almost 100% with winning chances, And you might find that -.2 correlates with 55% winning chances.

Larry obviously assumes perfect practical today play. In that case this means Rybka 3 for now.

A 0.02 obviously means 50.86% of winning when playing with another Rybka 3. If it plays with a weaker program obviously the only value the 50.86% has is that the probability of losing must be larger than 50.86%.

Reading that, it sounds like he did sort of what I suggested. Perhaps in a different way. Pick a position with a given evaluation, then run a monte-carlo match and look at the winning or losing percentage and then claim that that score corresponds to that winning percentage. However, I don't like the monte-carlo approach here using very fast games, to project the winning percentage for a score/evaluation produced in a longer game. I think the length of the time control has a significant influence on winning percentage for all programs (longer -> more draws for example).

[Uri Blass]

[d] 2kr3r/1ppqb1pp/p1n1pp1n/3pPb2/BP1P1P2/2P1BN2/P5PP/RN1Q1RK1 b - b3 0 11

How long does it like to play Kb8 in this position? I mean, come on. Isn't it stupid to walk directly into an attack?

Here is some analysis
Rybka on slow hardware admits after enough time that white is better here

Mischa - Kapaun, Wertungspartie, 3m + 0s 2008
2kr3r/1ppqb1pp/p1n1pp1n/3pPb2/BP1P1P2/2P1BN2/P5PP/RN1Q1RK1 b - b3 0 1

Analysis by Rybka 3 1-cpu 32-bit :

11...Nh6-g4
³ (-0.52) Depth: 2 00:00:00
11...Nh6-g4
³ (-0.55) Depth: 3 00:00:00
11...Nh6-g4
³ (-0.57) Depth: 4 00:00:00
11...Kc8-b8 12.h2-h3
³ (-0.38) Depth: 5 00:00:00
11...Kc8-b8 12.Qd1-d2 Rh8-f8
³ (-0.50) Depth: 6 00:00:00 15kN
11...Kc8-b8 12.Qd1-d2 Bf5-e4 13.Nb1-a3
³ (-0.49) Depth: 7 00:00:01 22kN
11...Kc8-b8 12.Nb1-a3 Nh6-g4 13.Qd1-d2 Bf5-e4
³ (-0.34) Depth: 8 00:00:01 36kN
11...Kc8-b8 12.Nb1-a3 Nh6-g4 13.Qd1-d2 a6-a5 14.Na3-c2 Bf5xc2
³ (-0.36) Depth: 9 00:00:04 93kN
11...Kc8-b8 12.Nb1-a3 Rh8-f8 13.b4-b5 a6xb5 14.Na3xb5 Nh6-g4
³ (-0.36) Depth: 10 00:00:08 176kN
11...Kc8-b8 12.Nb1-a3 Nh6-g4 13.Qd1-d2 a6-a5 14.Na3-c2 Bf5xc2 15.Ba4xc2 a5xb4 16.Ra1-b1
³ (-0.35) Depth: 11 00:00:12 273kN
11...Kc8-b8 12.Nb1-a3 Nh6-g4 13.Be3-c1 Bf5-e4 14.Na3-c2 h7-h5 15.h2-h3 Ng4-h6 16.Nc2-e3
³ (-0.37) Depth: 12 00:00:24 522kN
11...Kc8-b8 12.b4-b5 a6xb5 13.Ba4xb5
= (-0.19) Depth: 13 00:00:55 1113kN
11...Bf5-e4 12.Nb1-d2 Be4-d3 13.Rf1-e1 Nh6-f5 14.Ra1-c1 Kc8-b8 15.Be3-f2 h7-h5 16.Nd2-b3 Bd3-e4 17.Nb3-a5
= (-0.07) Depth: 14 00:07:18 8748kN
11...Bf5-e4 12.Nb1-d2 Be4-d3 13.Rf1-e1 Kc8-b8 14.Nd2-b3 Bd3-e4 15.Nb3-a5 b7-b5
= (-0.04) Depth: 15 00:08:09 9819kN
11...Bf5-e4 12.Nb1-d2 Be4-d3 13.Rf1-e1 Nh6-f5 14.Be3-f2 Kc8-b8 15.Nd2-b3 Bd3-e4 16.Nb3-d2
= (0.00) Depth: 16 00:10:40 12849kN
11...Nh6-g4 12.Be3-c1 Qd7-e8 13.h2-h3 Ng4-h6 14.Nb1-d2 Bf5-d3 15.Rf1-e1 Rh8-g8 16.Re1-e3 Bd3-g6 17.e5xf6 Be7xf6 18.Qd1-e1 Rd8-d6 19.Bc1-a3 b7-b5 20.Ba4-b3
= (0.14) Depth: 17 00:33:13 36883kN

(so k, 13.08.2008)

[Marc MP]

You have no clue what you talk about here. What has this position to do with Botvinnik's Games?

I have. Have you ever studied Botvinnik's games on French.
This position is about equal. The best player will win.
Black has all the counterplay he typically has in such structures (pressing center with f6 and or g5).

Continue to make a fool out of yourself.
This is not a French type position at all. This is sort of a Caro Kann Advance type position where black played Nc6 instead of c6 and castled long. In the French the bishop is passiv on c8.
The position is objectively lost for black.

I smashed Rybka 3 from this position on the playchess server in a 3 minute game but had to agree to a draw in a totally won position or I would have lost the game on time.

[Event "Wertungspartie, 3m + 0s"]
[Site "Maschinenraum"]
[Date "2008.07.25"]
[Round "?"]
[White "Mischa"]
[Black "Kapaun"]
[Result "1/2-1/2"]
[ECO "B00"]
[WhiteElo "2120"]
[BlackElo "2591"]
[Annotator "8.07;0.04"]
[PlyCount "82"]
[EventDate "2008.07.25"]
[TimeControl "180"]

{Rybka 3 32-bit server: 13.4 ply; 97kN/s Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.
40GHz 2400MHz, (4 threads)} 1. e4 {1} Nc6 {0.04/12 4} 2. d4 {(Sc3) 2} d5 {
0.04/13 5} 3. e5 {1} Bf5 {-0.01/14 5} 4. c3 {(Sf3) 4} Qd7 {-0.01/11 4} 5. Bb5 {
(Sf3) 9} f6 {-0.25/12 3} 6. f4 {(Sf3) 11} a6 {-0.32/11 2} 7. Ba4 {(Ld3) 4} Nh6
{-0.42/12 7} 8. Nf3 {4} e6 {-0.30/13 2} 9. O-O {11} Be7 {-0.23/13 2} 10. Be3 {
(Sa3) 4} O-O-O {-0.44/11 3} 11. b4 {(h3) 5} Kb8 {-0.50/10 2} 12. b5 {(h3) 8}
axb5 {-0.50/8 0} 13. Bxb5 {2} Be4 {-0.22/11 4} 14. Nbd2 {8} Ng4 {0.17/12 9} 15.
Bf2 {(De2) 6} b6 {0.10/9 3} 16. Qe2 {(Da4) 11} Nxf2 {0.18/10 18} 17. Rxf2 {2}
Bxf3 {0.91/10 4} 18. Nxf3 {5} Ba3 {1.10/12 4} 19. Rb1 {(De3) 11} Rhf8 {
1.57/10 8} 20. Rb3 {(c4) 2} Be7 {1.32/8 1} 21. a4 {1} g5 {2.23/10 8} 22. a5 {
(exf6) 3} g4 {1.12/8 1} 23. Ne1 {3} Rg8 {1.55/10 8} 24. Qa2 {(Sd3) 5} Nb4 {
3.52/8 2} 25. Rxb4 {11} Qc8 {6.83/11 0} 26. Rb3 {(axb6) 6} fxe5 {4.53/9 2} 27.
fxe5 {(Lc6) 2} Rg7 {5.36/8 5} 28. Bd3 {(axb6) 2} g3 {6.14/10 9} 29. hxg3 {5}
Qb7 {2} 30. axb6 {8.07/9 0} c6 {8.05/9 2} 31. Qa7+ {(La6) 3} Qxa7 {1} 32. bxa7+
{5.50/11 0} Ka8 {5.72/12 2} 33. Ra2 {(Tfb2) 3} Bg5 {5.16/10 2} 34. Nf3 {(Tb6) 3
} Be3+ {5.02/10 1} 35. Kf1 {1} Rxa7 {5.12/12 1} 36. Rxa7+ {2} Kxa7 {5.32/14 0}
37. Ke2 {1} Bh6 {5.31/12 0} 38. Ra3+ {(Lxh7) 1} Kb7 {4.96/11 1} 39. Rb3+ {
(Lxh7) 0} Ka7 {0.00/9 0} 40. Ra3+ {(Lxh7) 0} Kb7 {0.00/26 1} 41. Rb3+ {(Lxh7) 0
} Ka7 {(Lag: Av=0.15s, max=0.4s) 0.00/27 1} 1/2-1/2

[d]r3k2r/1ppqb1pp/p1n1pp1n/3pPb2/B2P1P2/2P1BN2/PP4PP/RN1Q1RK1 b kq - 3 1

It looks to me that 10... 0-0-0 is a bad move here. Black just castles into the attack. Black made this worse with 11... Kb8 ?, but the position is already bad after castling long. Castling short would have been a much better idea.