[Event "Chessgames Humans vs. Machines"]
[Site "Yahoo.com"]
[Date "2011.5.21"]
[Round ""]
[White "STOCKFISH"]
[Black "Schiller, Hosea & Wall"]
[Result "1/2-1/2"]
[Eco "C80"]
[Annotator "Schiller"]
[Source ""]

Comments on Stockfish - Humans game, 5/21/11:

Game score:

1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Ba4 Nf6 5.0–0 Nxe4 6.d4 b5 7.Bb3 d5 8.dxe5 Be6 9.Nbd2 Nc5 10.c3 d4 11.Bxe6 Nxe6 12.cxd4 Ncxd4 13.Ne4 Be7 14.Be3 Qd5 15.Nxd4 Qxe4 16.Nxe6 fxe6 17.Qh5+ g6 18.Qh3 Qxe5 19.Rfe1 Qf5 20.Qxf5 gxf5 21.Bd4 0–0 22.Rxe6 Rfe8 23.Rc6 Bd6 24.Kf1 Re4 25.Rd1 Rae8 26.g3 Bxg3 27.hxg3 Rd8 28.Rdc1 Rexd4 29.Rxa6 Rd1+ 30.Rxd1 Rxd1+ 31.Ke2 Rc1 32.Kd2 Rf1 33.Ke3 Re1+ 34.Kf4 Re2 35.f3 Rxb2 36.Kxf5 Kf8 37.g4 b4 38.Ra7 Rc2 39.Ra4 Rc4 40.f4 Kf7 41.Ra5 h6 42.a3 bxa3 43.Rxa3 Rc5+ 44.Ke4 Rc4+ 45.Kf3 Rc6 1/2-1/2

Stockfish seems to always think it has a healthy advantage as white when playing a Ruy Lopez. This game was no exception. From moves 10-25, the eval averaged about 0.60. From moves 26-40, the eval averaged close to 1.00, peaking at 1.09 at move 34. From moves 41-44, the eval averaged about 0.70. Then the eval crashed to 0.00 at move 45 and I accepted the draw offer from the human team. I have seen this behavior several times in this tournament - Stockfish teases me with what looks like a strong advantage, and then eventually says "never mind - it's a draw". Stockfish is the only tournament participant at this point with five draws out of five games.

Once again, the human team played solid chess and played very well together.

The most interesting period of the game started with white's 24th move. Here is a deep-ply analysis of the position:

After 23...Bd6:

click for larger view

Analysis by Stockfish 2.0.1 JA 64bit (31-ply):

1. ± (0.72): 24.g3 f4 25.Rd1 f3 26.Rc3 c5 27.Bf6 Re6 28.Bg5 Be5 29.Rxc5 Bxb2 30.Be3 Rae8 31.h4 Re4 32.Rg5+ Bg7 33.Rd7 R4e7 34.Rd3 Re5 35.Kh2 Rxg5 36.hxg5 Re4 37.Ra3

2. ² (0.60): 24.Kf1 Rad8 25.Bf6 Ra8 26.g3 Kf7 27.Bg5 Re4 28.Rd1 Ra4 29.a3 Ke6 30.Be3 Re4 31.Kg2 a5 32.Kf3 Rd8 33.Bg5 Rd7 34.h3 a4 35.Bf4 b4 36.h4 h5 37.Bg5 bxa3 38.bxa3 Rg7 39.Bh6

3. = (0.24): 24.Rac1 Re4 25.Bc3 b4 26.Bd2 Re2 27.Rd1 Re5 28.Rcc1 Rd5 29.g3 Re8 30.Kf1 Kf7 31.b3 a5 32.Be3 Rxd1+ 33.Rxd1 a4 34.Rd4 axb3 35.axb3 Bc5 36.Rd7+ Re7 37.Rxe7+ Bxe7 38.Ke2 Ke6 39.h3 h5

4. = (0.20): 24.Be3 f4 25.Bd2 Re4 26.Ba5 Re2 27.Rb1 Rae8 28.Kf1 f3 29.gxf3 Bxh2 30.Bxc7 Bxc7 31.Rxc7 Rd2 32.Kg2 Ree2 33.Ra7 Rxf2+ 34.Kg3 Rxb2 35.Rxb2 Rxb2 36.Rxa6 Kf7 37.Kg4 Rf2 38.f4 h5+ 39.Kf5 h4

24. Kf1 seemed to catch the human team off guard. This touched off a 20-minute debate for their next move. Here is a deep-ply analysis of black's 24th move:

After 24. Kf1:

click for larger view

Analysis by Stockfish 2.0.1 JA 64bit (31-ply):

1. ² (0.60): 24...f4 25.a3 Kf7 26.g3 Re6 27.Rd1 Ke7 28.Rc3 Rf8 29.Bc5 Kd7 30.b4 Rg6 31.Rd5 Re6 32.Kg2 Rg6 33.Rcd3 h6 34.Rh5 Ke6 35.Rd1 Rf5

2. ² (0.68): 24...a5 25.Rd1 f4 26.Bc5 Be5 27.b3 Rad8 28.Rxd8 Rxd8 29.Ke2 Kf7 30.g3 Rd5 31.Kf3 fxg3 32.hxg3 Rd2 33.Be3 Rxa2 34.Rc5 Bd6 35.Rxb5 a4 36.bxa4 Rxa4 37.Rh5 Kg6 38.Rh6+ Kg7 39.g4 Ra3 40.Rh1

3. ± (0.72): 24...Red8 25.Bc3 a5 26.g3 Kf7 27.Re1 Re8 28.Rd1 Re4 29.f4 a4 30.a3 Ke7 31.Rd2 Ra7 32.Bd4 Ra8 33.Bb6 Rc4 34.Rxc4 bxc4 35.Bd4 Kd7 36.Ke2 Kc6 37.Rd1 Rb8 38.Kd2 Re8 39.Kc3

4. ± (0.76): 24...Rad8 25.Bf6 Rd7 26.Rxa6 Bxh2 27.Bc3 Bd6 28.a3 Bc5 29.Ra5 c6 30.a4 Rb8 31.axb5 cxb5 32.g3 Rd5 33.Ra8 Rxa8 34.Rxa8+ Kf7 35.Rb8 Be7 36.Rh8 Kg6 37.Rg8+ Kh6 38.Ke2 Bg5 39.Re8 Kh5 40.f4

5. ± (0.88): 24...Kf7 25.Bc5 Be5 26.Re1 f4 27.g3 Bxb2 28.Rxc7+ Kg6 29.Rxe8 Rxe8 30.gxf4 Kf5 31.Be3 Re4 32.Rc5+ Kf6 33.Rc6+ Re6 34.Rc7 Re7 35.Rc2 Ba3 36.Ke2 Re4 37.Rc6+ Re6 38.Rc7 Bd6 39.Rxh7 Bxf4

6. ± (0.88): 24...Re4 25.Rd1 Bxh2 26.Bc5 Be5 27.b3 Rg4 28.Re1 Re8 29.g3 Kf7 30.Rxa6 f4 31.gxf4 Rxf4 32.Rh6 Rf5 33.Rxh7+ Kg6 34.Rd7 Ra8 35.Re2 Kf6 36.Rd5 Bc3 37.Be7+ Kg6 38.Rd7 c5 39.Rc7 Bd4 40.Rc2

Another interesting decision for the human team occurred on their 25th move:

After 25. Rd1:

click for larger view

Analysis by Stockfish 2.0.1 JA 64bit (30-ply):

1. ± (0.80): 25...a5 26.g3 a4 27.f3 Ree8 28.Bc5 Be5 29.Bb4 Rad8 30.Rxd8 Rxd8 31.Rc5 Bxb2 32.Rxb5 Rd1+ 33.Ke2 Ra1 34.Rxf5 Rxa2 35.Kd3 Ba3 36.Bxa3 Rxa3+ 37.Ke4 Ra2 38.Rc5 Re2+ 39.Kf5 Rxh2 40.Rxc7 a3 41.Ra7

2. ± (0.88): 25...Rd8 26.Bc3 b4 27.Bf6 Rde8 28.Rxa6 Rc4 29.g3 f4 30.Ra5 Kf7 31.Bd4 Re4 32.Rd5 Ke6 33.Bg7 Kf7 34.Bh8 Rc2 35.Bd4 Ke6 36.Rh5 Be5 37.Bxe5 Rxe5 38.Rxe5+ Kxe5 39.Re1+ Kd4 40.Re2

3. ± (1.01): 25...Bxh2 26.Bc5 Be5 27.b3 Rg4 28.Re1 Re8 29.g3 Kf7 30.Rxa6 f4 31.gxf4 Rxf4 32.Rh6 Rf5 33.Rxh7+ Kg6 34.Rd7 Ra8 35.a3 Bf4 36.Re6+ Kh5 37.Rc6 Bd6 38.Bxd6 cxd6 39.Rcxd6 Rxa3 40.Rd5 Kg6 41.Rxf5

4. ± (1.01): 25...f4 26.f3 Ree8 27.Bf2 Be5 28.b3 Red8 29.Re1 Rd5 30.Rc5 Rxc5 31.Bxc5 Bc3 32.Rd1 Re8 33.g4 h5 34.gxh5 Re5 35.Rd8+ Kf7 36.Rd7+ Ke8 37.Rxc7 Rxh5 38.Kg2 Rd5 39.Bf2 b4

5. ± (1.05): 25...Rae8 26.g3 Bxg3 27.hxg3 Rd8 28.Rxa6 Rexd4 29.Rc1 f4 30.gxf4 Rxf4 31.Rxc7 Rd2 32.Ra8+ Rf8 33.Rxf8+ Kxf8 34.b4 Rxa2 35.Rb7 Rb2 36.Rxb5 Ke7 37.Kg2 Rb3 38.Rb7+ Kd6 39.b5 h5 40.Rb6+ Ke7 41.f3

6. ± (1.05): 25...Bf4 26.f3 Re7 27.g3 Rd7 28.gxf4 Rad8 29.Kf2 Rxd4 30.Rxd4 Rxd4 31.Ke3 Ra4 32.a3 b4 33.Rc4 c5 34.axb4 cxb4 35.b3 Ra2 36.Rxb4 Rxh2 37.Kd4 Rh3 38.Ke5 Rxf3 39.Kxf5 Rg3 40.Ke5

It appears that Stockfish made a mistake with 41. Ra5. At deeper ply, here is its analysis:

After 40...Kf7:

click for larger view

Analysis by Stockfish 2.0.1 JA 64bit (31-ply):

1. ± (1.01): 41.Ke5 h6 42.a3 c5 43.Ra7+ Kf8 44.Ra6 Kg7 45.Kf5 Rd4 46.Ra5 Rd5+ 47.Ke4 Rd4+ 48.Ke5 Rc4 49.axb4 cxb4 50.Ra7+ Kg8 51.Rb7

2. ² (0.56): 41.g5 Kg7 42.a3 Rc5+ 43.Ke4 bxa3 44.Rxa3 Rc6 45.f5 h6 46.g6 Rc1 47.Kd5 Re1 48.Ra8 Rd1+ 49.Ke5 Re1+ 50.Kd4 Rd1+ 51.Ke3 Rf1 52.Ra5 Kf6 53.Ke4 Re1+ 54.Kf3 Rg1

3. ² (0.40): 41.a3 Rc5+ 42.Ke4 bxa3 43.Rxa3 Rc4+ 44.Kf3 Rc6 45.Kg3 h6 46.Ra8 Kf6 47.Rf8+ Kg7 48.Rc8 Rc3+ 49.Kf2 h5 50.gxh5 Rc5 51.Kf3 Rxh5 52.Rxc7+ Kg8 53.Rc6 Kf7 54.Kg4 Rh1 55.Kf5 Ra1 56.Rc7+ Kf8

4. = (0.24): 41.Ra6 Rc5+ 42.Ke4 Rc4+ 43.Ke5 Rc5+ 44.Kd4 Rc2 45.Ra5 h6 46.Ra4 c5+ 47.Kd5 Kf6 48.Ra6+ Kf7 49.Rxh6 Rxa2 50.Rh7+ Kf6 51.g5+ Kf5 52.Rf7+ Kg6 53.Rf6+ Kh5 54.Rb6 Kg4 55.Ke6 Kxf4 56.g6 Rg2

In fact, a deeper ply analysis after 41. Ra5 h6, shows that the eval has already collapsed to 0.00:

After 41...h6:

click for larger view

Analysis by Stockfish 2.0.1 JA 64bit (33-ply):

1. = (0.00): 42.a3 bxa3 43.Rxa3 Rc5+ 44.Ke4 Kf6 45.Ra6+ c6 46.Ra1 Rc4+ 47.Kf3 Rc3+ 48.Kg2 Rc2+ 49.Kg3 Rc3+ 50.Kh4 Rc2 51.Rb1 c5 52.Rb6+ Kg7 53.f5 c4 54.Kg3 c3 55.Rg6+ Kh7 56.Rc6 Kg7 57.Rc7+ Kf6

2. = (0.00): 42.Ra6 Rc2 43.Rxh6 Rxa2 44.Rh7+ Kg8 45.Rxc7 Ra5+ 46.Ke6 Rb5 47.Rc2 b3 48.Rb2 Kg7 49.f5 Kh6 50.f6 Rb6+ 51.Ke5 Kg6 52.g5 Rb5+ 53.Ke6 Rb6+ 54.Ke5

3. = (0.00): 42.Ke5 Rc6 43.Rb5 Rc4 44.Rd5 Rc2 45.Rd7+ Kf8 46.f5 Rxa2 47.Rxc7 Ra5+ 48.Ke6 Ra6+ 49.Kd5 Rb6 50.Rc2 b3 51.Rb2 Kf7 52.Kc5 Rb8 53.Rb1 Kf6 54.Kc4 h5 55.gxh5 Kxf5 56.Rxb3 Rxb3 57.Kxb3 Kf6

4. = (0.00): 42.Rb5 Ke7 43.Re5+ Kf7 44.Ra5 Rc6 45.Rb5 Rc4 46.Ke5 Ke7 47.Ra5 Rc2 48.Ra4 Rb2 49.Kd5 Rd2+ 50.Ke4 Re2+ 51.Kf3 Rb2 52.Ra6 c5 53.Rxh6 Rxa2 54.g5 c4 55.Rb6 b3 56.f5 b2 57.f6+ Kf7

However, 41. Ke5 seems to work out no better for white. Sliding forward, after <41. Ke5 h5 42. gxh5 Rc5+ 43. Ke4 Rxh5 44. Rxb4 Ra5>, the eval again crashed to 0.00.

<1. e4 e5?!>

Mankind entered this game trailing Aylerkupp / Rybka by half a point, so a more dynamic opening, such as the Owen, might have been called for.

<2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Nxe4>

Initially suggested by yours truly, purely as a joke! Amazingly enough (or perhaps not...) <Open Defence> actually went for it.

<6. d4 b5 7. Bb3 d5 8. dxe5 Be6 9. Nbd2>

<Eric Schiller>, who reappeared after being absent for a while, already disliked Black's position here somewhat.

<9...Nc5>

<Open Defence> suggested 9...Be7, but Schiller strongly supported the text. Clocks: 1:40 - 1:28

<10. c3 d4 11. Bxe6 Nxe6 12. cxd4 Ncxd4 13. Ne4 Be7 14. Be3 Qd5>

Home prep by <OD>, and a ChessGames database novelty.

<15. Nxd4 Qxe4 16. Nxe6 fxe6 17. Qh5+>

White here had several interesting alternatives, including 17.Rc1 and 17.Re1.

<17...g6 18. Qh3 Qxe5 19. Rfe1 Qf5 20. Qxf5 gxf5 21. Bd4 O-O 22. Rxe6 Rfe8>

<Open Defence> was still in prep...

<23. Rc6 Bd6 24. Kf1!?>

A surprise to Houdini and Hosea alike. In retrospect it was probably a good move, though not quite enough to confuse the human team into losing.

<24...Re4>

A very long think - curiously it took 8 minutes into the discussion for a serious alternative to be suggested! At that point, <Open Defence> started championing for 24...Kf7. To me, the stated reasoning of "avoiding Bc5" seemed rather curious, as I felt the move would merely make Bc5 more effective! (It should, however, still be good enough for a draw.) Mankind eventually spent some 20 minutes on this move, easily the carbon units' longest think in the tournament. Clocks: 1:15 - 0:52

<25. Rd1>

Houdini's evals here rose markedly at deep depths, but Black remains in draw territory.

<25...Rae8 26. g3 Bxg3>

This rather simple ploy was initially missed by mankind's finest, but eventually spotted by <wwall>. The rook endgame should be drawn with best play. Clocks: 1:10 - 0:44

(Missing the tactic wouldn't have been any instant disaster, as 26...a5 followed by 27...a4 appears to be a legit alternative.)

<27. hxg3 Rd8 28. Rdc1>

Both 28.Rxc7 and 28.Rxa6 are very playable alternatives. Indeed, the strongest absolute continuation was probably 28.Rxa6 Rexd4 29.Rxd4! Rxd4 30.Ke2, but even then White appears unlikely to win.

<28... Rexd4 29. Rxa6 Rd1+ 30. Rxd1 Rxd1+>

Black spent 43 seconds on this move...

<31. Ke2 Rc1 32. Kd2 Rf1 33. Ke3 Re1+ 34. Kf4 Re2 35. f3 Rxb2 36. Kxf5 Kf8 37. g4 b4 38. Ra7 Rc2 39. Ra4 Rc4 40. f4 Kf7>

Clocks: 0:48 - 0:26

<41. Ra5>

In post-game analysis Stockfish turned to 41.Ke5, which is indeed a better move but still only draws.

<41...h6 42. a3 bxa3>

<wwall>'s 42...c5 was also playable. <Open Defence> thought 43.Ra7+ looked dangerous... I'd be dangerous with a shotgun!