Komodo strikes first at the TCEC Superfinal

[syzygy]

Bc2 wins in all lines and it was in SF pv moves earlier, somehow it played Kg4.

Well, if it was that simple and obvious, Stockfish would have played it !
Maybe you should play instead of Stockfish.

The winning line was in SF pv as early as move 60 including the bizarre 70 Ke6! as can be seen from the archives

60. g4 h5 61. c6 hxg4+ 62. Bxg4 Qd6 63. Bf5 Qf6 64. Bc2 Rb4 65. Be4 Rb2 66. Kg4 Qe6+ 67. Kg5 Qe7+ 68. Kf5 Qf7+ 69. Kxe5 Qg7+ 70. Ke6 Qg8+ 71. Kf5 Rb5+ 72. Kf4

Note that if Komodo would have seen all of that, it would have reported a +6.50 score or higher and the TCEC win rule would have kicked in before the bug did.

[syzygy]

Bc2 wins in all lines and it was in SF pv moves earlier, somehow it played Kg4.

Well, if it was that simple and obvious, Stockfish would have played it !
Maybe you should play instead of Stockfish.

The winning line was in SF pv as early as move 60 including the bizarre 70 Ke6! as can be seen from the archives

60. g4 h5 61. c6 hxg4+ 62. Bxg4 Qd6 63. Bf5 Qf6 64. Bc2 Rb4 65. Be4 Rb2 66. Kg4 Qe6+ 67. Kg5 Qe7+ 68. Kf5 Qf7+ 69. Kxe5 Qg7+ 70. Ke6 Qg8+ 71. Kf5 Rb5+ 72. Kf4

Note that if Komodo would have seen all of that, it would have reported a +6.50 score or higher and the TCEC win rule would have kicked in before the bug did.

Actually, Komodo did see all of that already at move 54:

54... Qd8 55. Qe3 Rb2 56. Bf3 Rb1+ 57. Kh2 Rb2+ 58. Kh3 Qf6 59. b7 @ 59... Qe6+ 60. g4 h5 61. c6 hxg4+ 62. Bxg4 Qd6 63. Bf5 Qf6 64. Bc2 Rb4 65. Be4 Rb2 66. Kg4 Qe6+ 67. Kg5 Qg8+ 68. Kf5 Qf7+ 69. Kxe5 Qg7+ 70. Ke6 Qg8+ 71. Kf5 Rb5+ 72. Kf4 (...)

but reporting only +4.50 which saved it.

[MikeB]

Here is the winning line..

[Event "TCEC SuperFinal 8"]
[Site "?"]
[Date "2015.11.11"]
[Round "?"]
[White "P2: 5 Stockfish"]
[Black "Komodo"]
[Result "*"]
[Annotator "IM Erik Kislik"]
[SetUp "1"]
[FEN "7k/1P6/2P2q2/4pB2/8/4Q2K/1r6/8 w - - 0 1"]
[PlyCount "59"]
[Source "ChessBase"]
[SourceDate "2013.11.20"]

{Endgames in which your king is being chased for a very long time by the
opponent's queen are some of the most unpleasant to play and calculate for the
side being chased: on every move, you have to try to calculate how to reach
safety and find a safe haven. In some cases, it is psychologically
overwhelming. Optically, in many cases having no pawn cover for your king
makes you feel like there is no way to protect the king from endless checks.
To help deal with this psychological flaw, once every couple of months, I turn
on an engine with 6-man Tablebases and play through some difficult endgames I
played with an exposed king in the past. Each time I discover something new,
and defensive ideas and patterns click a little bit more in my mind. Sometimes
extremely visually unpleasant moves start to look natural, and confusing lines
start to seem digestible. Queen endings in particular are very difficult for
human players to handle well. The following endgame features one of the most
impressive and unbelievable king walks I have ever seen.} 1. Bc2 $1 {This is a
beautiful move, intending Qe4 with an immediate win. The bishop cannot be
captured because the rook is overloaded with defending b8. It is an optically
surprising move because it looks like it can be captured but it really cannot.}
Rb4 2. Be4 $1 {Now that Qe4 is not possible, the bishop comes back to e4 to
block the fourth rank to protect its king.} Rb2 3. Kg4 $1 Qe6+ 4. Kg5 Qe7+ 5.
Kf5 Qf7+ 6. Kxe5 {This is a very difficult plan to decide on, but what most
likely could be determined by process of elimination: amazingly, Black appears
to hold the position by all normal means and attempts by White, so the only
way forward is to take the pawn and try to avoid the checks with the queen and
rook. The bishop will be instrumental in blocking those checks and taking away
important checking squares.} Qg7+ 7. Ke6 $3 {This is a beautiful piece of
reciprocal logic: coming to f5 immediately does not work because Black gets
perpetual check. By coming to e6 first, White makes it harder for Black's
queen to conveniently check the king.} (7. Kf5 Rb5+ 8. Kf4 Qh6+ 9. Kf3 Qh3+ $11
) 7... Qg8+ 8. Kf5 Rb5+ (8... Qf7+ 9. Kg4 Qe6+ (9... Qg8+ 10. Qg5 Qe6+ 11. Qf5
Qg8+ 12. Kh3 Rb3+ 13. Bf3 $18) 10. Kg3 Qg8+ 11. Kf4 Qf7+ 12. Bf5 Rb4+ 13. Kg3
Rb3 14. Bd3 Qg7+ 15. Kh3 {and the checks stop.}) 9. Kf4 {The king now begins
the journey home.} Qf7+ (9... Qg5+ 10. Kf3 Qh5+ 11. Kf2 Qh4+ 12. Ke2 $18) 10.
Kg3 Qf6 {With the king cut off on the g-file, Black shoots for ...Rg5+. White
can luckily allow the king to pass though with accurate defense.} (10... Qg7+
11. Kf2 Qf6+ 12. Ke2 Rb2+ (12... Qb2+ 13. Qd2 {ends the checks.}) 13. Ke1 Qh4+
14. Kd1 Qh5+ 15. Kc1 {and Black must resign.}) 11. Qf3 $1 Rg5+ 12. Kf2 Qb2+ 13.
Qe2 Qd4+ 14. Kf3 Qf6+ 15. Ke3 Qc3+ 16. Qd3 Qe1+ 17. Kf3 Rg3+ {Black can win
the queen, but White will promote with check.} (17... Qg3+ 18. Ke2 {walks the
king to safety.}) 18. Kf4 Qf2+ 19. Bf3 {finally stops the checks and wins for
White. After traveling all the way to e6 and back, the king has walked from f2
to f3 to e3 to f3 to f4 and found safety. Each time, the king tried to find a
way to block checks from the front. Now that Black is checking from behind,
the king resides safely on f4.} Qh2 20. Qc3+ Rg7+ (20... Kh7 21. Qc2+ $1 {
White had other ways to win, but this is the simplest and most satisfying.}
Qxc2 22. Be4+ Qxe4+ 23. Kxe4 Rg8 24. c7 {is a funny finish to the game.}) 21.
Ke3 Qh6+ 22. Kd4 {Now White's king is safe, and can comfortably hide behind
the pawns on b7 and c6. No special calculation is required from here on out.}
Qd6+ 23. Kc4 Qf4+ 24. Kb5 Qf5+ 25. Ka6 Qb1 26. c7 Qa2+ 27. Kb5 (27. Kb6 Qb1+
28. Ka7 Qa2+ 29. Kb8 Qh2 30. Qa1 {This prevents checks on the a-file.} Kh7 31.
Ka8 Qxc7 32. b8=Q $18 {is another way to queen.}) 27... Qb1+ 28. Ka4 Qa2+ 29.
Kb4 Qb1+ 30. Ka3 $18 *

+1 excellent - one of the best analysis/lessons ever posted on TalkChess - thank you. And it pretty much shows without a doubt that SF had excellent winning chances - but perhaps it was simply too deep for even SF. I don't think for one second it proves there's a bug with SF. Some positions are still simply still too deep for current hardware. Makes for an interesting conversation to say the least.

[Uri Blass]

Here is the winning line..

[Event "TCEC SuperFinal 8"]
[Site "?"]
[Date "2015.11.11"]
[Round "?"]
[White "P2: 5 Stockfish"]
[Black "Komodo"]
[Result "*"]
[Annotator "IM Erik Kislik"]
[SetUp "1"]
[FEN "7k/1P6/2P2q2/4pB2/8/4Q2K/1r6/8 w - - 0 1"]
[PlyCount "59"]
[Source "ChessBase"]
[SourceDate "2013.11.20"]

{Endgames in which your king is being chased for a very long time by the
opponent's queen are some of the most unpleasant to play and calculate for the
side being chased: on every move, you have to try to calculate how to reach
safety and find a safe haven. In some cases, it is psychologically
overwhelming. Optically, in many cases having no pawn cover for your king
makes you feel like there is no way to protect the king from endless checks.
To help deal with this psychological flaw, once every couple of months, I turn
on an engine with 6-man Tablebases and play through some difficult endgames I
played with an exposed king in the past. Each time I discover something new,
and defensive ideas and patterns click a little bit more in my mind. Sometimes
extremely visually unpleasant moves start to look natural, and confusing lines
start to seem digestible. Queen endings in particular are very difficult for
human players to handle well. The following endgame features one of the most
impressive and unbelievable king walks I have ever seen.} 1. Bc2 $1 {This is a
beautiful move, intending Qe4 with an immediate win. The bishop cannot be
captured because the rook is overloaded with defending b8. It is an optically
surprising move because it looks like it can be captured but it really cannot.}
Rb4 2. Be4 $1 {Now that Qe4 is not possible, the bishop comes back to e4 to
block the fourth rank to protect its king.} Rb2 3. Kg4 $1 Qe6+ 4. Kg5 Qe7+ 5.
Kf5 Qf7+ 6. Kxe5 {This is a very difficult plan to decide on, but what most
likely could be determined by process of elimination: amazingly, Black appears
to hold the position by all normal means and attempts by White, so the only
way forward is to take the pawn and try to avoid the checks with the queen and
rook. The bishop will be instrumental in blocking those checks and taking away
important checking squares.} Qg7+ 7. Ke6 $3 {This is a beautiful piece of
reciprocal logic: coming to f5 immediately does not work because Black gets
perpetual check. By coming to e6 first, White makes it harder for Black's
queen to conveniently check the king.} (7. Kf5 Rb5+ 8. Kf4 Qh6+ 9. Kf3 Qh3+ $11
) 7... Qg8+ 8. Kf5 Rb5+ (8... Qf7+ 9. Kg4 Qe6+ (9... Qg8+ 10. Qg5 Qe6+ 11. Qf5
Qg8+ 12. Kh3 Rb3+ 13. Bf3 $18) 10. Kg3 Qg8+ 11. Kf4 Qf7+ 12. Bf5 Rb4+ 13. Kg3
Rb3 14. Bd3 Qg7+ 15. Kh3 {and the checks stop.}) 9. Kf4 {The king now begins
the journey home.} Qf7+ (9... Qg5+ 10. Kf3 Qh5+ 11. Kf2 Qh4+ 12. Ke2 $18) 10.
Kg3 Qf6 {With the king cut off on the g-file, Black shoots for ...Rg5+. White
can luckily allow the king to pass though with accurate defense.} (10... Qg7+
11. Kf2 Qf6+ 12. Ke2 Rb2+ (12... Qb2+ 13. Qd2 {ends the checks.}) 13. Ke1 Qh4+
14. Kd1 Qh5+ 15. Kc1 {and Black must resign.}) 11. Qf3 $1 Rg5+ 12. Kf2 Qb2+ 13.
Qe2 Qd4+ 14. Kf3 Qf6+ 15. Ke3 Qc3+ 16. Qd3 Qe1+ 17. Kf3 Rg3+ {Black can win
the queen, but White will promote with check.} (17... Qg3+ 18. Ke2 {walks the
king to safety.}) 18. Kf4 Qf2+ 19. Bf3 {finally stops the checks and wins for
White. After traveling all the way to e6 and back, the king has walked from f2
to f3 to e3 to f3 to f4 and found safety. Each time, the king tried to find a
way to block checks from the front. Now that Black is checking from behind,
the king resides safely on f4.} Qh2 20. Qc3+ Rg7+ (20... Kh7 21. Qc2+ $1 {
White had other ways to win, but this is the simplest and most satisfying.}
Qxc2 22. Be4+ Qxe4+ 23. Kxe4 Rg8 24. c7 {is a funny finish to the game.}) 21.
Ke3 Qh6+ 22. Kd4 {Now White's king is safe, and can comfortably hide behind
the pawns on b7 and c6. No special calculation is required from here on out.}
Qd6+ 23. Kc4 Qf4+ 24. Kb5 Qf5+ 25. Ka6 Qb1 26. c7 Qa2+ 27. Kb5 (27. Kb6 Qb1+
28. Ka7 Qa2+ 29. Kb8 Qh2 30. Qa1 {This prevents checks on the a-file.} Kh7 31.
Ka8 Qxc7 32. b8=Q $18 {is another way to queen.}) 27... Qb1+ 28. Ka4 Qa2+ 29.
Kb4 Qb1+ 30. Ka3 $18 *

+1 excellent - one of the best analysis/lessons ever posted on TalkChess - thank you. And it pretty much shows without a doubt that SF had excellent winning chances - but perhaps it was simply too deep for even SF. I don't think for one second it proves there's a bug with SF. Some positions are still simply still too deep for current hardware. Makes for an interesting conversation to say the least.

You could be right in case that stockfish really did not see the win but seeing the win earlier and later forgetting it is a bug and I read that the bug(smp bug) was already fixed before the game but too late because only after the tournament started.

http://abrok.eu/stockfish/

Author: lucasart
Date: Tue Nov 10 21:41:42 2015 +0000
Timestamp: 1447191702

Ensure that rootDepth < DEPTH_MAX

Indeed, if we use a depth >= DEPTH_MAX, we start having negative depth in the
TT (due to int8_t cast).

No functional change in single thread mode

Resolves #490

[M ANSARI]

Bc2 wins in all lines and it was in SF pv moves earlier, somehow it played Kg4.

Well, if it was that simple and obvious, Stockfish would have played it !
Maybe you should play instead of Stockfish.

The winning line was in SF pv as early as move 60 including the bizarre 70 Ke6! as can be seen from the archives

60. g4 h5 61. c6 hxg4+ 62. Bxg4 Qd6 63. Bf5 Qf6 64. Bc2 Rb4 65. Be4 Rb2 66. Kg4 Qe6+ 67. Kg5 Qe7+ 68. Kf5 Qf7+ 69. Kxe5 Qg7+ 70. Ke6 Qg8+ 71. Kf5 Rb5+ 72. Kf4

Note that if Komodo would have seen all of that, it would have reported a +6.50 score or higher and the TCEC win rule would have kicked in before the bug did.

Yes the +6.5 score needs a re-think. I don't see the value of not letting that go to say when both sides see mate. If one engine has a +6.5 score then in general the other side will be mated quite quickly. It is unfortunate that SF lost a half point in this truly amazing game and I really think a human has zero chances of winning that position as there were many "only" moves and that included allowing an X ray check and seeing a very long and exhaustive king march that would go up and down the board several times. I do think that this type of position is trivial for the latest engines as they have no problem going quickly through all the possible checks and am not surprised that this is a bug of some sort. Give credit to Komodo though for a truly amazing level of resistance and for understanding the position and pouncing on the opportunity to equalize! I think that this game more than any other ... truly shows how far the level of chess by engines has surpassed the level of the very best humans. I think Kasparov and Carlsen would look at this game in awe!

[syzygy]

And it pretty much shows without a doubt that SF had excellent winning chances - but perhaps it was simply too deep for even SF.

SF saw the win, there is no doubt about that whatsoever.

The old YBW version returns a TB win score after a while. The current lazy smp version has a too unstable search to return a TB win score, but it does go up to +70 and higher. (The reason it went from +26 to +8 before it blundered is that the search became unstable with scores going up and down like crazy.)

I don't think for one second it proves there's a bug with SF.

If SF sees the winning continuation for 5 moves or so with the right score, and then suddenly picks the wrong move with a draw score, then something is very definitely wrong.

[Hai]

Bc2 wins in all lines and it was in SF pv moves earlier, somehow it played Kg4.

Well, if it was that simple and obvious, Stockfish would have played it !
Maybe you should play instead of Stockfish.

The winning line was in SF pv as early as move 60 including the bizarre 70 Ke6! as can be seen from the archives

60. g4 h5 61. c6 hxg4+ 62. Bxg4 Qd6 63. Bf5 Qf6 64. Bc2 Rb4 65. Be4 Rb2 66. Kg4 Qe6+ 67. Kg5 Qe7+ 68. Kf5 Qf7+ 69. Kxe5 Qg7+ 70. Ke6 Qg8+ 71. Kf5 Rb5+ 72. Kf4

Note that if Komodo would have seen all of that, it would have reported a +6.50 score or higher and the TCEC win rule would have kicked in before the bug did.

Yes the +6.5 score needs a re-think. I don't see the value of not letting that go to say when both sides see mate. If one engine has a +6.5 score then in general the other side will be mated quite quickly. It is unfortunate that SF lost a half point in this truly amazing game and I really think a human has zero chances of winning that position as there were many "only" moves and that included allowing an X ray check and seeing a very long and exhaustive king march that would go up and down the board several times. I do think that this type of position is trivial for the latest engines as they have no problem going quickly through all the possible checks and am not surprised that this is a bug of some sort. Give credit to Komodo though for a truly amazing level of resistance and for understanding the position and pouncing on the opportunity to equalize! I think that this game more than any other ... truly shows how far the level of chess by engines has surpassed the level of the very best humans. I think Kasparov and Carlsen would look at this game in awe!

TCEC rule should be changed into:
Both engines must evaluate for 10 moves with +30.00 or more to give a 1-0 result.

[AdminX]

Bc2 wins in all lines and it was in SF pv moves earlier, somehow it played Kg4.

Well, if it was that simple and obvious, Stockfish would have played it !
Maybe you should play instead of Stockfish.

The winning line was in SF pv as early as move 60 including the bizarre 70 Ke6! as can be seen from the archives

60. g4 h5 61. c6 hxg4+ 62. Bxg4 Qd6 63. Bf5 Qf6 64. Bc2 Rb4 65. Be4 Rb2 66. Kg4 Qe6+ 67. Kg5 Qe7+ 68. Kf5 Qf7+ 69. Kxe5 Qg7+ 70. Ke6 Qg8+ 71. Kf5 Rb5+ 72. Kf4

Note that if Komodo would have seen all of that, it would have reported a +6.50 score or higher and the TCEC win rule would have kicked in before the bug did.

Yes the +6.5 score needs a re-think. I don't see the value of not letting that go to say when both sides see mate. If one engine has a +6.5 score then in general the other side will be mated quite quickly.

I disagree, I think it should be left the way it is. I think doing what you suggest would just encourage developers to lower their evals to beat the TCEC rule on this.

[Uri Blass]

And it pretty much shows without a doubt that SF had excellent winning chances - but perhaps it was simply too deep for even SF.

SF saw the win, there is no doubt about that whatsoever.

The old YBW version returns a TB win score after a while. The current lazy smp version has a too unstable search to return a TB win score, but it does go up to +70 and higher. (The reason it went from +26 to +8 before it blundered is that the search became unstable with scores going up and down like crazy.)

I don't think for one second it proves there's a bug with SF.

If SF sees the winning continuation for 5 moves or so with the right score, and then suddenly picks the wrong move with a draw score, then something is very definitely wrong.

I think that the main problem in testing with fishtest is that the system adjudicate games and not save important games.

It is better to have no adjudication and keep pgn of the games with the biggest evaluation score that the side with the high evaluation did not win because these games can be productive for fixing bugs in stockfish.

[Hai]

It's simple Stockfish hadn't won game 22 because he evaluated Kh3 as a draw by 3-Fold Rep but it was only a 2-Fold Rep.
But Stockfish evaluates 2-Fold Rep = 3-Fold Rep.
And that's why Stockfish played Kf3??.

As a result you need to fix 2-Fold Rep and evaluate this correctly.