Grob is draw?

[Jouni]

https://www.chessdb.cn/queryc_en/ has now 1.g4 with -204 evaluation. Should be loss. But but - after playing 10 first moves it's only -159. These moves: 1. g4 d5 2. e3 Nc6 3. d4 e5 4. Nc3 Be6 5. dxe5 Nxe5 6. h3 h5 7. Nf3 Nxf3+ 8. Qxf3 hxg4 9. hxg4 Bxg4 10. Qg2 Rxh1. Why?

[Eelco de Groot]

https://www.chessdb.cn/queryc_en/ has now 1.g4 with -204 evaluation. Should be loss. But but - after playing 10 first moves it's only -159. These moves: 1. g4 d5 2. e3 Nc6 3. d4 e5 4. Nc3 Be6 5. dxe5 Nxe5 6. h3 h5 7. Nf3 Nxf3+ 8. Qxf3 hxg4 9. hxg4 Bxg4 10. Qg2 Rxh1. Why?

I don't know how they compute those scores, does not have to be comparable with Stockfish output. For instance after your line, neither side has made any more mistakes, apart from the first move I think? So if -204 only reflects how bad that move is in comparison with standard openings, maybe it goes up if White does play optimally further on. I have no idea.

I get this in your line after a while, with one of Eduard's Stockfish variants:

[pgn]
[Event "?"]
[Site "?"]
[Date "2024.10.01"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]

1. g4 d5 2. e3 Nc6 3. d4 e5 4. Nc3 Be6 5. dxe5 Nxe5 6. h3
h5 7. Nf3 Nxf3+ 8. Qxf3 hxg4 9. hxg4 Bxg4 10. Qg2 Rxh1
11. Qxh1 Nf6 12. Bd2 Qd6 *
[/pgn]

[d]r3kb2/ppp2pp1/3q1n2/3p4/6b1/2N1P3/PPPB1P2/R3KB1Q w Qq -

Engine: Deep Ripper -1- (1024 MB)
gemaakt door EN and Stockfish developers (see AUTHORs

54 279:43 -1.54 13.Pb5 Db6 14.f3 a6 15.Pc3 Lf5
16.O-O-O O-O-O 17.Kb1 Te8 18.Lc1 Lb4
19.Pe2 Kb8 20.Pd4 Lc8 21.Pb3 g6
22.Dh4 Le7 23.Lg2 Ld7 24.Lf1 Dd6
25.Lg2 De6 26.Dh2 (7.622.325.323) 454

54 279:43 -1.57 13.Dg2 O-O-O 14.f3 Ld7 15.O-O-O g6
16.Pb5 Db6 17.a4 a6 18.a5 De6 19.Pd4 De5
20.Pb3 La4 21.Kb1 Lxb3 22.cxb3 Lg7
23.Ld3 Pe8 24.Lc1 Pd6 25.Th1 Kb8
26.Th7 (7.622.325.323) 454

54 279:43 -1.72 13.Lh3 Lxh3 14.Dxh3 Dd7 15.Df1 O-O-O
16.O-O-O g6 17.f3 a6 18.Kb1 Lc5
19.a3 De6 20.Te1 Th8 21.Pa4 Ld6
22.Lb4 Lxb4 23.axb4 Dh3 24.De2 Te8
25.b3 Dh4 26.Dd2 (7.622.325.323) 454

[Jouni]

These are SF evaluations. BTW this page https://github.com/robertnurnberg/grobt ... -file#1-g4 shows, that score has gone over 200, but win percentage is not improved in one year. About 98,5 % now.

[Jouni]

Grob is black win. But Stockfish is not perfect yet. It cannot win all black games. This happen in one of 70 test games.
[pgn]
[Event "Grob"]
[Site "?"]
[Date "2024.10.03"]
[Round "1"]
[White "ShashChess 35.1"]
[Black "Stockfish 17"]
[Result "1/2-1/2"]
[ECO "A00"]
[Annotator "-1.20;-1.33"]
[PlyCount "163"]
1. g4 d5 {-1.33/29 10} 2.
e3 {-1.20/25 12 (c4)} Nc6 {-1.39/31 15 (h5)} 3. d4 {-1.22/23 3} e5 {-1.40/28 3}
4. Nc3 {-1.25/26 6} Be6 {-1.37/28 4} 5. dxe5 {-1.30/27 14} Nxe5 {-1.44/26 4} 6.
h3 {-1.36/29 10} h5 {-1.41/30 5} 7. Nf3 {-1.33/29 10} Nxf3+ {-1.51/29 5} 8.
Qxf3 {-1.36/27 3} hxg4 {-1.44/31 6} 9. hxg4 {-1.37/27 6} Bxg4 {-1.43/30 6} 10.
Qg2 {-1.47/28 14} Rxh1 {-1.41/30 5} 11. Qxh1 {-1.32/28 4} Nf6 {-1.44/30 5} 12.
Bd2 {-1.33/28 4} Qd6 {-1.53/30 7} 13. Qg2 {-1.38/23 3} O-O-O {-1.46/28 5} 14.
f3 {-1.37/27 5} Bd7 {-1.51/29 7} 15. O-O-O {-1.42/28 6} g6 {-1.42/29 6} 16. Kb1
{-1.38/28 10} Qe6 {-1.49/32 11} 17. Ne2 {-1.32/30 21} Bc5 {-1.53/28 6} 18. Nf4
{-1.38/24 3} Qe7 {-1.58/30 6} 19. Re1 {-1.46/30 25} Bb4 {-1.56/30 6} 20. a3 {
-1.46/25 4} Bxd2 {-1.58/30 5} 21. Qxd2 {-1.49/26 4} Qd6 {-1.63/32 23} 22. Qd4 {
-1.40/25 4} Kb8 {-1.63/29 5} 23. Nd3 {-1.46/27 4} Re8 {-1.63/30 5} 24. Nc5 {-1.
46/25 3} Bc8 {-1.66/27 5} 25. Qb4 {-1.43/29 8} Qg3 {-1.63/29 8} 26. Ka2 {-1.44/
26 10 (Bb5)} Ka8 {-1.66/28 10} 27. Bd3 {-1.58/28 28} Qd6 {-1.67/27 5 (c6)} 28.
Nb3 {-1.40/23 3 (Kb1)} Qxb4 {-1.77/30 8 (Qe7)} 29. axb4 {-1.24/29 3} a6 {-1.74/
30 5} 30. Nd4 {-1.30/29 4} Kb8 {-1.81/29 5} 31. c4 {-1.21/32 7 (Kb3)} c6 {-1.
31/65 18} 32. b3 {-1.20/31 4} Kc7 {-1.30/35 4 (Bd7)} 33. Kb2 {-1.19/32 3} Kd6 {
-1.30/44 5 (Rh8)} 34. Kc3 {-1.22/32 5} Bd7 {-1.32/36 6 (Rh8)} 35. Bf1 {-1.23/
34 10 (Kd2)} Rh8 {-1.27/43 7} 36. Bd3 {-1.14/30 2 (Be2)} Rh2 {-1.24/41 5 (Rh3)}
37. e4 {-0.85/33 12 (Rb1)} dxe4 {-0.91/35 11} 38. fxe4 {-0.92/27 1} Ke7 {-1.15/
41 26} 39. e5 {-0.85/29 2} Ng4 {-1.20/25 1} 40. e6 {-1.05/30 5} fxe6 {-1.21/29
2} 41. Bxg6 {-1.12/26 3} Kf6 {-1.10/37 7 (e5)} 42. Bd3 {-1.07/28 2} e5 {-1.07/
28 1} 43. Nf3 {-1.04/29 2} Rh3 {-1.06/29 1} 44. Ng1 {-1.03/27 3} Re3 {-1.19/31
5} 45. Rf1+ {-1.12/28 3} Ke7 {-1.17/27 1} 46. Kd2 {-1.20/30 16} Be6 {-1.20/25 2
} 47. Ne2 {-0.95/26 4} Rh3 {-1.21/26 2} 48. Rg1 {-0.85/22 2 (Ng1)} Rh2 {-0.92/
30 12} 49. Kc3 {-0.78/20 1} Nf2 {-0.94/23 1} 50. Bb1 {-0.97/26 8} Bg4 {-0.97/
27 2} 51. Ng3 {-0.88/21 1} Nd1+ {-1.12/28 6 (Ke6)} 52. Kd3 {-0.94/23 2} Nb2+ {
-0.86/27 4 (Ke6)} 53. Kc3 {-1.02/24 1} Nd1+ {-0.83/19 0} 54. Kd3 {-0.97/24 1}
Nf2+ {-0.77/28 5} 55. Kc3 {-0.93/21 1} Ke6 {-0.73/22 1} 56. Bc2 {-0.78/26 7}
Nh3 {-0.75/21 1} 57. Rh1 {-0.99/23 2} Rxh1 {-0.75/21 1} 58. Nxh1 {-0.72/21 1}
Nf4 {-0.72/28 4 (Ng5)} 59. Ng3 {-0.69/20 2} b6 {-0.76/27 1} 60. Kd2 {-0.58/23 2
} Ke7 {-0.75/37 2} 61. Be4 {-0.60/22 2} Kd7 {-0.80/28 2} 62. Nh1 {-0.44/27 6}
Nh3 {-0.77/28 2} 63. Ke3 {-0.50/25 1 (Bc2)} Kd6 {-0.57/34 6} 64. Ng3 {-0.53/22
1} Ng5 {-0.59/24 2} 65. Bc2 {-0.44/24 2} Be6 {-0.45/29 10} 66. Ne2 {-0.43/23 3}
Bf7 {-0.37/27 4 (c5)} 67. Nc3 {-0.35/20 2} Ne6 {-0.28/23 3 (b5)} 68. Bb1 {-0.
32/25 3} Bg8 {-0.24/28 4} 69. Na4 {-0.25/26 2} b5 {-0.19/22 2} 70. Nc3 {-0.19/
26 3} Ke7 {-0.19/25 2} 71. Ke4 {-0.18/25 2} Kf6 {-0.16/21 1} 72. Ke3 {-0.22/28
2} Ke7 {-0.23/32 3 (Bf7)} 73. Ke4 {-0.13/30 2} Kf6 {-0.17/25 1} 74. Ke3 {-0.14/
32 2} Bf7 {-0.17/30 2} 75. Ne4+ {-0.12/33 5 (Ne2)} Ke7 {-0.17/26 2} 76. Bd3 {
-0.10/32 2} Bh5 {-0.07/36 4} 77. Kd2 {-0.11/30 2 (Bf1)} Bg6 {-0.12/33 6 (Bf3)}
78. Nf2 {-0.11/34 8} Bxd3 {-0.08/30 1} 79. Kxd3 {0.00/32 1} Kf6 {-0.01/48 3}
80. Ne4+ {0.00/38 2} Kf5 {0.00/43 2} 81. Ng3+ {0.00/40 2} Kf4 {0.00/47 2} 82.
Ne2+ {0.00/42 2 Draw accepted} 1/2-1/2

[/pgn]

[jkominek]

These are SF evaluations. BTW this page https://github.com/robertnurnberg/grobt ... -file#1-g4 shows, that score has gone over 200, but win percentage is not improved in one year. About 98,5 % now.

This is a fascinating project. I like what the author has done in creating grobtrack.

For the historical plot of the PV's leaf node evaluation (top graph at the web link) the sudden sharp jump upwards in the first week of September aligns to the release of Stockfish 17. So I'd posit that noobpwnftw upgraded the evaluation engine shortly after its release, and the upgrade accounts for the sudden change of centipawn evals. But a retuned win-draw-loss model accompanies the new release. This accounts for the win percentages being stable through the transition. As a metric, win percentage has its own issue of stability, as the function that maps centipawns to win percentage depends on the depth of the PV, and the best line length does fluctuate day to day. Still useful information though.

So, the Grob is a win for black? The cdb evaluation at time of writing is -193. What about the delayed Grob? I've just now taken it out to (a mere) 32 ply. cdb reports -219.

[jefk]

there are endgame positions (eg. opposite bishops or fortress) where
you still can achieve a draw with a minus two eval.
But from the starting position with 1.g4 there are sufficient options for
Black to avoid such drawish positions and thus I believe it's a win for Black.
Apparently at the end of in the middle game (practical endgames) there
are a lot of variations, and it also can take some time before the eval
goes from minus two to eg. - 2.5 or later -3. A separate project,
independent but possibly similar to the cn.db could generate more
options/variations in a faster way, and thus imo probably could achieve
such evals as -3 within a reasonable period (lets say a few years); after
which it's only a matter of time to declare the Grob as a loss for White.

Meanwhile normal chess (eg. with 1.Nf3! 1.e4, d4, or 1.c4?!) is a draw
with perfect play as is already evident from the cn.db (and some logical
reasoning and -computer- or - correspondence- chess experience).