TCEC LIVE - Division I, Season 1 - All the top engines!

[Jeremy Bernstein]

Rybka 4 appears to analyze ...a2 as second choice for a short while, but then discards it entirely (...Kf7 takes its place). Houdini would have played ...a2. Naum 4.2 agrees with Houdini. HIARCS can't decide between ...Kf7 and ...Nxc6 for a while, although it comes around at around depth 22, discarding Nxc6 and evaluating both ...Kf7 and ...a2 at around 1.25. Rybka 3 reaches an identical conclusion as her big sister. Stockfish agrees with Rybka in this matter, as well.

I might point out that I analysed this with both IvanHoe and ComStock with all the TripleBases, and in each case backed up the score after 96...a2 to 0.00 (after playing forward to a draw), and the same with a +6 score after 96...Nxc6. Whether or not this is 100% reliable (due to transpositions, nullmove, and whatnot) is another issue. [With 96...Kf7, the scores hovered in the +2 range, though many of these were from RBP vs RN, which might well be drawn -- if I followed lines were pawns were not exchanged, it was still unclear, and I lost interest after knowing the results from the other two moves]. From the TCEC data, Rybka was playing at about 5 seconds per move from moves 96-100 (after using 46s at move 95 -- both had 11m left until Shredder used 8 minutes on 101. g5), and then suddenly took a 13-second think at move 102 and realised White was +3.

It really is amazing to me how many of these get decided in the later stages, though I think some of the Freestyle guys had made similar comments (that an error in the middlegame might change the score by -0.25 or -0.5, but in the endgame would simply lose).

It actually looks like Rybka's downfall began the move before. The expected move was ...a2, Rybka played ...Rb8. Houdini evaluates the position nearly a half-pawn in Shredder's favor at that point already if 96. f4. Critter is even more optimistic, giving Shredder about a half-pawn before ...Rb8 was played, and 1.3+ and rising afterward). DS12 (Mac) considers itself already a pawn ahead, regardless of what Black plays. Rybka's move simply makes matters worse.

Rybka had time, so I assume it just missed the point.

Martin, where does the evaluation displayed on the site come from? White? Or is there a 3rd engine providing evaluation commentary?

Jeremy

[BB+]

Martin, where does the evaluation displayed on the site come from? White? Or is there a 3rd engine providing evaluation commentary?

My understanding is that the TCEC site evals are from the engines themselves, while ChessBomb uses Stockfish. For instance, the TCEC data-dump from ChessGUI looks like:

90.Rxg5 { ev=0.10, d=23, pd=Rxf3, mt=00:00:19, tl=00:12:35, s=6919 kN/s, tb=0,  }
Nf4+ { ev=0.00, d=21, pd=Kxh4, mt=00:00:15, tl=00:11:07, s=817 kN/s, tb=321,  }
91.Kxh4 { ev=0.70, d=22, pd=Ne6, mt=00:00:34, tl=00:12:12, s=6968 kN/s, tb=0,  }
Ne6 { ev=0.00, d=21, pd=Rh5, mt=00:00:04, tl=00:11:14, s=765 kN/s, tb=0,  }
92.Rh5+ { ev=0.54, d=21, pd=Kg6, mt=00:00:18, tl=00:12:05, s=7237 kN/s, tb=0,  }
Kg6 { ev=0.00, d=22, pd=Be3, mt=00:00:03, tl=00:11:21, s=927 kN/s, tb=54,  }
93.Be3 { ev=0.41, d=21, pd=Rh8, mt=00:00:19, tl=00:11:58, s=7210 kN/s, tb=0,  }
Rfd8 { ev=0.00, d=22, pd=f4, mt=00:00:12, tl=00:11:20, s=777 kN/s, tb=29,  }
94.Rxd8 { ev=0.94, d=21, pd=Nxd8, mt=00:00:18, tl=00:11:50, s=7389 kN/s, tb=0,  }
Nxd8 { ev=0.00, d=20, pd=f4, mt=00:00:05, tl=00:11:27, s=693 kN/s, tb=186,  }
95.Rh8 { ev=0.87, d=24, pd=a2, mt=00:00:20, tl=00:11:41, s=7722 kN/s, tb=0,  }
Rb8 { ev=0.00, d=24, pd=f4, mt=00:00:46, tl=00:10:52, s=757 kN/s, tb=45027,  }
96.f4 { ev=1.49, d=23, pd=Nxc6, mt=00:00:19, tl=00:11:33, s=8100 kN/s, tb=0,  }
Nxc6 { ev=0.00, d=23, pd=f5, mt=00:00:06, tl=00:10:57, s=832 kN/s, tb=3,  }
97.f5+ { ev=1.93, d=26, pd=Kf7, mt=00:00:17, tl=00:11:28, s=8021 kN/s, tb=0,  }
Kf7 { ev=0.00, d=23, pd=Rxb8, mt=00:00:04, tl=00:11:03, s=836 kN/s, tb=22,  }
98.Rxb8 { ev=1.93, d=28, pd=Nxb8, mt=00:00:17, tl=00:11:22, s=7805 kN/s, tb=0,  }
Nxb8 { ev=0.00, d=23, pd=g5, mt=00:00:02, tl=00:11:13, s=833 kN/s, tb=0,  }
99.Bc1 { ev=2.47, d=28, pd=a2, mt=00:00:20, tl=00:11:13, s=9485 kN/s, tb=0,  }
a2 { ev=0.00, d=26, pd=Bb2, mt=00:00:02, tl=00:11:21, s=501 kN/s, tb=1320,  }

[Jeremy Bernstein]

Martin, where does the evaluation displayed on the site come from? White? Or is there a 3rd engine providing evaluation commentary?

My understanding is that the TCEC site evals are from the engines themselves, while ChessBomb uses Stockfish. For instance, the TCEC data-dump from ChessGUI looks like:

90.Rxg5 { ev=0.10, d=23, pd=Rxf3, mt=00:00:19, tl=00:12:35, s=6919 kN/s, tb=0,  }
Nf4+ { ev=0.00, d=21, pd=Kxh4, mt=00:00:15, tl=00:11:07, s=817 kN/s, tb=321,  }
91.Kxh4 { ev=0.70, d=22, pd=Ne6, mt=00:00:34, tl=00:12:12, s=6968 kN/s, tb=0,  }
Ne6 { ev=0.00, d=21, pd=Rh5, mt=00:00:04, tl=00:11:14, s=765 kN/s, tb=0,  }
92.Rh5+ { ev=0.54, d=21, pd=Kg6, mt=00:00:18, tl=00:12:05, s=7237 kN/s, tb=0,  }
Kg6 { ev=0.00, d=22, pd=Be3, mt=00:00:03, tl=00:11:21, s=927 kN/s, tb=54,  }
93.Be3 { ev=0.41, d=21, pd=Rh8, mt=00:00:19, tl=00:11:58, s=7210 kN/s, tb=0,  }
Rfd8 { ev=0.00, d=22, pd=f4, mt=00:00:12, tl=00:11:20, s=777 kN/s, tb=29,  }
94.Rxd8 { ev=0.94, d=21, pd=Nxd8, mt=00:00:18, tl=00:11:50, s=7389 kN/s, tb=0,  }
Nxd8 { ev=0.00, d=20, pd=f4, mt=00:00:05, tl=00:11:27, s=693 kN/s, tb=186,  }
95.Rh8 { ev=0.87, d=24, pd=a2, mt=00:00:20, tl=00:11:41, s=7722 kN/s, tb=0,  }
Rb8 { ev=0.00, d=24, pd=f4, mt=00:00:46, tl=00:10:52, s=757 kN/s, tb=45027,  }
96.f4 { ev=1.49, d=23, pd=Nxc6, mt=00:00:19, tl=00:11:33, s=8100 kN/s, tb=0,  }
Nxc6 { ev=0.00, d=23, pd=f5, mt=00:00:06, tl=00:10:57, s=832 kN/s, tb=3,  }
97.f5+ { ev=1.93, d=26, pd=Kf7, mt=00:00:17, tl=00:11:28, s=8021 kN/s, tb=0,  }
Kf7 { ev=0.00, d=23, pd=Rxb8, mt=00:00:04, tl=00:11:03, s=836 kN/s, tb=22,  }
98.Rxb8 { ev=1.93, d=28, pd=Nxb8, mt=00:00:17, tl=00:11:22, s=7805 kN/s, tb=0,  }
Nxb8 { ev=0.00, d=23, pd=g5, mt=00:00:02, tl=00:11:13, s=833 kN/s, tb=0,  }
99.Bc1 { ev=2.47, d=28, pd=a2, mt=00:00:20, tl=00:11:13, s=9485 kN/s, tb=0,  }
a2 { ev=0.00, d=26, pd=Bb2, mt=00:00:02, tl=00:11:21, s=501 kN/s, tb=1320,  }

Sorry, I'm left-handed and apparently blind. I saw the eval on the left-hand side and missed it on the right. Now everything makes sense. Sorry for the noise.

[BB+]

Another late-game surprise in Stockfish-Naum. After an inordinate amount of shuffling with 0.00 scores from White (but Naum desiring more), the prying open of the position gives White an advantage due to a passed e-pawn and the weak Black pawns (and the light-squared Bishop is finally not so hobbled). Again I'm not so sure whether "increment" chess really played a part here, but the moves are coming much faster around move 100 than previously.

[Martin Thoresen]

Jeremy,

Yes BB+ is correct, the eval is from the playing engines themselves.

Best,
Martin

[BB+]

So close to termination... on the last possible move (10th ply) of a 0.00 sequence that would end Division I, HIARCS jumps to 0.52. This is also a fight for survival, as HIARCS needs to win to avoid relegation. [It seems the position is dead-drawn though White is up 2 pawns, and even w/o TripleBases I do not think IvanHoe would lose this -- unless some mis-guided attempt at a perpetual (all draws are alike, right?) pushes the value of a passed pawn past the horizon -- and now it's increment chess to boot, so everything is possble perhaps]. EDIT: Egad, it's the Rook and a+c pawns ending again in some lines! But I don't think IvanHoe has the requisite 6-piece TB knowledge...

[BB+]

HIARCS is giving +1.35 in a 5-piece TB draw. Maybe it's in "swindle" mode... [Given all the other rules about draws/resignation, I'm surprised there isn't simply GUI adjudication, at least for 5 piece].

Final standings appear to be:

Houdini   +5 -0 =9  24
Rybka     +6 -3 =5  23
Stockfish +4 -2 =8  20
Critter   +3 -3 =8  17
IvanHoe   +3 -3 =8  17
Shredder  +1 -4 =9  12
Naum      +2 -6 =6  12
HIARCS    +0 -3 =11 11

I might point out that HIARCS had by far the least volatile openings (absolute value of ~0.11 only, with the mean about twice that).

[BB+]

Oh great... A perfect ending... with a Bishop underpromotion!!

And it even shows forth a bug in the system (probably with ChessGUI and the 50-move rule?):
[White "HIARCS"]
[Black "IvanHoe"]
[Result "1/2-1/2"]
1. e4 c6 2. d4 d5 3. exd5 cxd5 4. c4 Nf6 5. Nf3 e6 6. Nc3 Bb4 7. a3 Bxc3+ 8. bxc3 O-O 9. Qc2 h6 10. Bd3 dxc4 11. Bxc4 b6 12. Bf4 Bb7 13. Be2 Nc6 14. O-O Rc8 15. Rac1 Ne7 16. Ne5 Ne4 17. Qd3 f6 18. Nc4 g5 19. Be3 Qd5 20. f3 Ba6 21. Qc2 Ng3 22. hxg3 Bxc4 23. f4 Bxe2 24. Qxe2 Nf5 25. g4 Ng3 26. Qc2 Qe4 27. Rfe1 Qxc2 28. Rxc2 Ne4 29. c4 Rfd8 30. g3 gxf4 31. gxf4 f5 32. gxf5 exf5 33. a4 Kf7 34. c5 Rg8+ 35. Kf1 Ke6 36. Rb2 bxc5 37. dxc5 Rg7 38. Rc2 Kd7 39. Rb2 Kc6 40. Rb5 Ng3+ 41. Kf2 h5 42. Ra5 Ne4+ 43. Kf1 Rcg8 44. Ra6+ Kb7 45. Re6 Ng3+ 46. Kf2 Kc8 47. Kf3 Ne4 48. Re5 Rg3+ 49. Ke2 Rg2+ 50. Kd3 Rd8+ 51. Kc4 h4 52. Rxf5 h3 53. Rf7 h2 54. Rh7 Nf6 55. Rxa7 Rc2+ 56. Kb5 Rd3 57. Ra8+ Kc7 58. Rh8 Rb3+ 59. Ka5 Ng4 60. Bd4 Rd3 61. Be5+ Nxe5 62. Rxe5 Rd8 63. Rh7+ Rd7 64. Rhh5 Rf7 65. f5 Rf2 66. Rh8 R7xf5 67. Rxf5 Rxf5 68. Kb5 Rf2 69. Rh7+ Kb8 70. Ka6 Rb2 71. c6 Rc2 72. Kb6 Rb2+ 73. Ka5 Kc8 74. Ka6 Kb8 75. a5 Rg2 76. Kb5 Rc2 77. Rh6 Kc7 78. a6 Rb2+ 79. Kc5 h1=R 80. Rxh1 Rc2+ 81. Kb5 Rb2+ 82. Ka5 Kxc6 83. Rc1+ Kd7 84. a7 Ra2+ 85. Kb6 Rb2+ 86. Ka6 Ra2+ 87. Kb7 Rb2+ 88. Ka8 Ra2 89. Rc3 Ra1 90. Kb7 Rb1+ 91. Ka6 Ra1+ 92. Kb6 Rb1+ 93. Kc5 Ra1 94. Rf3 Kc7 95. Rf7+ Kc8 96. Rg7 Rc1+ 97. Kd6 Rd1+ 98. Kc6 Rc1+ 99. Kd5 Ra1 100. Rh7 Rd1+ 101. Kc6 Rc1+ 102. Kd6 Rd1+ 103. Ke5 Ra1 104. Rg7 Re1+ 105. Kf6 Ra1 106. Ke6 Re1+ 107. Kf5 Ra1 108. Re7 Rf1+ 109. Ke5 Ra1 110. Kd5 Rd1+ 111. Ke4 Ra1 112. Kd4 Rd1+ 113. Kc5 Ra1 114. Kb6 Rb1+ 115. Ka6 Ra1+ 116. Kb5 Rb1+ 117. Kc4 Ra1 118. Kb4 Rb1+ 119. Ka4 Ra1+ 120. Kb3 Ra5 121. Rf7 Ra1 122. Rh7 Ra5 123. Kb4 Ra1 124. Kb5 Rb1+ 125. Kc5 Ra1 126. Rf7 Rc1+ 127. Kd6 Rd1+ 128. Kc6 Rc1+ 129. Kb5 Ra1 130. Kb4 Rb1+ 131. Kc4 Ra1 132. Kd3 Rd1+ 133. Kc3 Ra1 134. a8=B ½-½So why is this a draw? The 50-move rule did not take effect (it was 99 ply, not 100 ply), there was nonzero scores within 5 moves (and a pawn-push to boot), so...

EDIT: The reported message is "{ Game Nr. 56 : Hiarcs 13.2 draws against Ivanhoe B47cB by 50-move rule GameDuration = 05:29:06, Finalposition B1k5/5R2/8/8/8/2K5/8/r7 b - - 100 134 }" -- but this is bogus, as after a8=B the 50-move counter is reset to zero. Looks like a bug in the corner-case in ChessGUI.

Also, for the pedants, IvanHoe took 4th place by 0.5 Sonneborn-Berger tie-break points. Naum is demoted under any scoring, but Shredder/HIARCS actually had equal S-B tie-breaks, and so if the 3/1/0 scoring were not in effect, would be tied for the other relegation spot, though then greater number of wins gives Shredder the edge.

[Martin Thoresen]

Division I has now finished, thanks to everyone following
and congratulations to Robert & Houdini for winning.

Houdini and Rybka will meet in the Elite Match which will start
after a maintenance break, while Hiarcs and Naum will play
in Division II for the next Season.

Final standings:

Code: Select all

N Engine           Pts     SB    Ho Ry St Cr Iv Sh Na Hi 

1 Houdini 1.5      24.0  147.00     1= == == == 11 1= 1=
2 Rybka 4.0        23.0  128.00  0=    0= =1 11 10 1= =1
3 Stockfish 2.0.1  20.0  128.00  == 1=    0= =0 == 11 =1
4 Ivanhoe B47cB    17.0  115.00  == =0 1=    =0 =1 01 ==
5 Critter 0.9      17.0  114.50  == 00 =1 =1    == 01 ==
6 Shredder 12.0    12.0   91.50  00 01 == =0 ==    == ==
7 Naum 4.2         12.0   80.50  0= 0= 00 10 10 ==    ==
8 Hiarcs 13.2      11.0   91.50  0= =0 =0 == == == ==   

[Dave Mitchell]

All the programs (including the interface and your website), got a good chance to show their stuff, no doubt.

Congratulations on the tournament, Martin, and thanks.