Sorry about bringing up the Belka thing again but I thought there was an interesting evaluation, of which I'm not sure whether it's an example of a bad Belka bug or a really brilliant difference in eval possibly also displayed by Rybka?
I have not tested this with Rybka so maybe it is interesting to see what recent versions of Rybka think of this position.
The position came up as a result of the discovery that there was a flaw in my recent tests of Glaurung where I was using the -excellent!- Rybka.bkt book made by Harry schnapp for the tests. I discovered that because I had left several sliders in the GUI book options in the default position for use with the Shredder books and had also included the requirement that every line could be played only if there were at least two games with this line, that not following here Harry's original instructions, this had the unfortunate effect of making the book so very narrow that a large proportion of my testgames consisted just of a single variation in the Sicilian Najdorf
Yes you may laugh, but it was rather a pity of all my diligent testing of Glaurung settings! It remains to be seen what will be the outcome of new tests if I will be using either fixed starting positions or the Rybka.bkt again, but this time with a broader choice of repertoire.
I was also looking at the particular variation that is as follows:
1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Bc4
Nbd7 7. Bg5 e6 8. O-O Qb6 9. Bb3 Be7 10. a4 Nc5 11. a5 Qc7
12. f3 h6 13. Be3 Nxb3 14. Nxb3 Bd7 15. Bb6 Qc8 16. f4 Bc6 {Last bookmove}
I wanted to see how much work there is involved if you wanted to add a few moves at the end of a bookline. Unfortunately there is no way to add single moves easily in some sort of menu, I think that goes for most major GUIs, lines have to be added as individual games.
I ended up by having SCID making a new PGN where I can edit some moves without disrupting analyzing engines, and analyzing moves by several engines parallel in two copies of the Shredder interface.
The "main line" of my slightly modified Rybka.bkt now looks as follows in this pgn using one of the testgames against Belka, output by SCID:
[Event "12 Minutes/Game + 3 Seconds/Move"]
[Site "Engine Match"]
[Date "2007.12.11"]
[Round "2"]
[White "Glaurung 2.0.1 Mjolnir 001g2"]
[Black "Belka 1.8.11"]
[Result "0-1"]
1.e4 {book 0s}
1...c5 {book 0s}
2.Nf3 {book 0s}
2...d6 {book 0s}
3.d4 {book 0s}
3...cxd4 {book 0s}
4.Nxd4 {book 0s}
4...Nf6 {book 0s}
5.Nc3 {book 0s}
5...a6 {book 0s}
6.Bc4 {book 0s}
6...Nbd {book 0s}
7.Bg5 {book 0s}
7...e6 {book 0s}
8.O-O {book 0s}
8...Qb6 {book 0s}
9.Bb3 {book 0s}
9...Be7{book 0s}
10.a4 {book 0s}
10...Nc5 {book 0s}
11.a5 {book 0s}
11...Qc7 {book 0s}
12.f3 {book 0s}
12...h6 {book 0s}
13.Be3 {book 0s}
13...Nxb3 {book 0s}
14.Nxb3 {book 0s}
14...Bd7 {book 0s}
15.Bb6 {book 0s}
15...Qc8 {book 0s}
16.f4 {book 0s}
16...Bc6 {book 0s}
17.Qe2
( 17.Qd3 {20.03 38:13 -0.15 17.Qe2 O-O 18.Bd4 Re8 19.Rf3 Qb8
20.Rd1 e5 21.fxe5 dxe5 22.Bc5 Bxc5+ 23.Nxc5 Qa7 24.b4 Rad8 25.Rfd3
Rxd3 26.cxd3 Rd8 27.Qf2 Qb8 28.Rf1 Qc7 Glaurung; 21.01 70:58 -0.17
17.Qe2 O-O 18.Rad1 Re8 19.Rf3 d5 20.e5 Nd7 21.Bd4 Nf8 22.Bb6 Ng6
23.Nd4 Bb4 24.Nxc6 bxc6 25.Na2 Be7 26.Nc3 Rb8 27.Qd3 c5 Glaurung}
17...O-O {-0.26/14 34s} 18.Nd4 {-0.17/15 38s} 18...d5 {-0.04/14
20s} 19.exd5 {-0.17/16 38s} 19...Bxd5 {-0.06/15 31s} 20.Ra4 {
-0.29/15 52s} 20...Re8 {+0.20/15 2:19m} 21.Kh1 {-0.47/15 56s}
21...Nd7 {+0.27/15 29s} 22.f5 {-0.52/15 19s} 22...Nxb6 {+0.20/15
21s} 23.axb6 {-0.35/16 21s} 23...Qc5 {+0.35/14 45s} 24.fxe6 {
-0.41/14 26s} 24...fxe6 {+0.45/14 27s} 25.Qe2 {-0.54/14 40s} 25...
Qxb6 {+0.62/12 3s} 26.Nxd5 {-0.88/14 19s} 26...exd5 {+0.63/12 5s}
27.Nf5 {-0.72/14 19s} 27...Bf6 {+0.78/13 19s} 28.Qg4 {-0.76/14 17s
} 28...Kh7 {+0.93/13 16s} 29.Rb4 {-0.70/14 16s} 29...Qc6 {+1.08/14
16s} 30.Qh5 {-0.82/13 14s} 30...Re4 {+1.56/14 16s} 31.Rxe4 {
-1.72/15 19s} 31...dxe4 {+1.50/14 13s} 32.b3 {-2.00/15 21s} 32...
Re8 {+1.64/14 28s} 33.Qe2 {-1.94/14 22s} 33...e3 {+1.73/15 16s}
34.c4 {-2.13/15 19s} 34...g6 {+1.72/15 20s} 35.Ng3 {-2.37/15 14s}
35...Kg7 {+1.70/14 21s} 36.Qg4 {-2.19/14 14s} 36...Re5 {+1.79/13
9s} 37.Qe2 {-2.33/14 15s} 37...h5 {+1.79/14 8s} 38.Qd3 {-2.29/16
12s} 38...h4 {+1.86/15 11s} 39.Ne2 {-2.35/17 12s} 39...h3 {
+1.82/14 6s} 40.Rf3 {-2.35/14 11s} 40...hxg2+ {+1.82/14 7s} 41.
Kxg2 {-2.35/14 0s} 41...Kh7 {+1.87/14 18s} 42.Ng3 {-2.37/15 38s}
42...b6 {+1.87/14 10s} 43.Qe2 {-2.21/13 7s} 43...Bg5 {+1.91/13 7s}
44.Qf1 {-2.23/14 20s} 44...Be7 {+1.99/13 11s} 45.Qe2 {-2.25/13 7s}
45...Bc5 {+2.07/14 15s} 46.Qd1 {-2.45/13 14s} 46...a5 {+2.14/13 6s
} 47.Qf1 {-2.39/13 8s} 47...Re6 {+2.40/13 10s} 48.Qe2 {-2.76/14 9s
} 48...Rd6 {+3.26/14 5s} 49.Nf1 {-4.07/14 6s} 49...Qe4 {+3.52/14
6s} 50.h4 {-4.19/14 7s} 50...Kg8 {+3.54/13 7s} 51.Kh3 {-4.62/14
10s} 51...Qe6+ {+3.93/13 5s} 52.Kh2 {-4.88/15 15s} 52...Rd4 {
+4.27/14 16s} 53.Rxe3 {-4.86/14 7s} 53...Rxh4+ {+4.27/13 2s} 54.
Kg2 {-5.60/15 8s} 54...Rg4+ {+4.49/14 5s} 55.Kh2 {-5.98/15 6s}
55...Bxe3 {+4.49/14 5s} )
( )
17...O-O 18.Rad1 Nd7 19.Be3 Qc7 20.Rf3 *
[d]r4rk1/1pqnbpp1/p1bpp2p/P7/4PP2/1NN1BR2/1PP1Q1PP/3R2K1 b - -
Now in this position the programs that I tried all want to play 20... Nc5 21. Nxc5 dxc5 This is not so bad as a main line for White, I thought a possible move next could be 22.Rg3 with the possibilty of an attack on the enemy King. The programs don't really handle that very well but this position does get a better eval in time and it looks like a normal human plan to me to move the Rook to g3, possibly the queen can go to h5 later. A possibly big downside however is that the pawn on a5 must fall, - unless the Rook moves from d1 to a1 again, but this Rook is needed for the attack on the Kingside!, if not, a5 falls and Black can the march on the Kingside with its pawns on the a and b file. Are the programs underestimating this?
Belka was the only program that came up with a different variation that involved forcing Black to exchange its last knight on b3 and using this possibilty to get the Bishop back to b6 where it can chase the black queen away again and have pawn a5 defended again. This looks all eminently logical and the only drawback is the doubled pawn after the exchange of the Knights on b3. The last of these pawns on b2 is however a very important base for the Knight on c3 and the front runner on b3 is very useful when it moves up to b4. After an exchange bxa5 bxa5 the doubled pawn is then dissolved so in fact there is no weakness attached to the doubled pawn. Is Belka the only program capable of evaluting this doubled pawn correctly or am I missing something?
If I follow Belka's variation further the eval is at the moment in this position after 21. Bd4 Nxb3 22. Bb6 Qd7 23. cxb3
[d] r4rk1/1p1qbpp1/pBbpp2p/P7/4PP2/1PN2R2/1P2Q1PP/3R2K1 b - -
Three best moves multivariation analysis, the eval has gone down somewhat but is still positive for White:
r4rk1/1p1qbpp1/pBbpp2p/P7/4PP2/1PN2R2/1P2Q1PP/3R2K1 b - -
Engine: Belka 1.8.11 (64 MB)
by Yuri Osipov, Igor Korshunov
18 69:29 -0.12 23...f5 24.Bd4 fxe4 25.Nxe4 Bxe4
26.Qxe4 d5 27.Qg6 Bf6 28.Bc5 Rfc8
29.b4 Bxb2 30.Rh3 Qf7 31.Rxh6 Qxg6
32.Rxg6 Kf7 33.Rg3 Rh8 34.Re1 (1.124.037.064) 269
18 46:31 -0.30 23...Bd8 24.Bc5 Bc7 25.Rg3 Qe7
26.Bb4 Qf6 27.Qe3 Rfd8 28.Ne2 Qxb2
29.Bc3 Qc2 30.Rxg7+ Kf8 31.Rd3 Bxa5
32.Bf6 Qb1+ 33.Nc1 (749.573.766) 268
18 74:04 -0.42 23...Rac8 24.Bd4 Qd8 25.Rg3 g6
26.Qg4 Bd7 27.Bb6 Qe8 28.Qh5 Kh7
29.Qe2 e5 30.fxe5 Bh4 31.Rgd3 Qxe5
32.Rxd6 (1.197.312.642) 269
Analysis by Glaurung in parallel, with an eval favoring Black:
r4rk1/1p1qbpp1/pBbpp2p/P7/4PP2/1PN2R2/1P2Q1PP/3R2K1 b - -
Engine: Glaurung 2.0.1 Mjolnir 001h (64 MB)
by Tord Romstad
20 121:03 +0.17 23...Rac8 24.Rg3 g6 25.Rh3 Kh7
26.Bd4 Rfe8 27.b4 Bf8 28.Rhd3 Qe7
29.Bb6 Kg8 30.b5 axb5 31.Rxd6 Qf6
32.R6d4 Qxf4 33.Nxb5 (1.255.350.840) 172
Fruit Mandarin also sees an advantage for White but not choosing 21. Bd4 in the first position
[d]r4rk1/1pqnbpp1/p1bpp2p/P7/4PP2/1NN1BR2/1PP1Q1PP/3R2K1 b - -
Engine: Fruit 2.3.1 Mandarin (64 MB)
by Fabien Letouzey, France
13.01 0:44 -0.02 20...Nc5 21.Nxc5 dxc5 22.f5 Rad8
23.Rxd8 Rxd8 24.fxe6 fxe6 25.Rf1 Qe5
26.Bf4 Qd4+ 27.Kh1 Bf6 28.Be3 Qb4
29.Qg4 Qxb2 30.Qxe6+ Kh8 (6.602.553) 148
14.01 1:18 +0.02 20...Nc5 21.Nxc5 dxc5 22.f5 Rad8
23.Rxd8 Rxd8 24.fxe6 fxe6 25.Rf1 Kh8
26.Qh5 Rf8 27.Rf7 Rxf7 28.Qxf7 Qd6
29.Bf4 e5 30.Be3 (11.856.679) 150
15.01 3:04 -0.38-- 20...Nc5 21.Nxc5 dxc5 22.f5 Rad8
23.Rdf1 Qxa5 24.Rg3 Bh4 25.Rxg7+ Kxg7
26.Qh5 Bf6 27.Qxh6+ Kg8 28.Qxf6 Rc8
29.fxe6 fxe6 (30.055.896) 162
15.02 4:30 -0.29 20...Rac8 21.Bd4 e5 22.Be3 f5
23.fxe5 Bxe4 24.Rg3 g5 25.e6 Ne5
26.Bd4 Bf6 27.Nxe4 fxe4 28.Rc3 Qh7
29.Rxc8 Rxc8 30.Bxe5 Bxe5 (43.244.182) 159
16.01 13:02 -0.29 20...Rac8 21.Bd4 e5 22.Be3 f5
23.fxe5 Bxe4 24.Rg3 g5 25.e6 Ne5
26.Bd4 Bf6 27.Nxe4 fxe4 28.Rc3 Qh7
29.Rxc8 Rxc8 30.Bxe5 Bxe5 (119.070.436) 152
17.01 39:07 -0.45 20...Rac8 21.Rg3 Kh8 22.Qg4 Rg8
23.Qh5 Rcf8 24.Bd4 Nf6 25.Qe2 e5
26.Bb6 Qb8 27.Rf3 Qc8 28.Kh1 Qg4
29.fxe5 Nxe4 30.Nxe4 Bxe4 (351.339.259) 149
17.02 45:18 -0.24 20...Nc5 21.Nxc5 dxc5 22.f5 Bd6
23.Kh1 Rad8 24.f6 Be5 25.fxg7 Rxd1+
26.Nxd1 Rd8 27.Bxh6 Bxe4 28.Rf1 Bg6
29.Qg4 Bf5 30.Qh5 Bxg7 31.Qg5 f6 (408.925.865) 150
18.01 62:17 -0.17 20...Nc5 21.Nxc5 dxc5 22.f5 Bd6
23.Kh1 Rad8 24.f6 Be5 25.fxg7 Rxd1+
26.Nxd1 Rd8 27.Bxh6 Bxe4 28.Rf1 Bg6
29.Nc3 Bf5 30.Qh5 Bxg7 31.Bxg7 Kxg7
32.Qg5+ Kh7 (565.048.303) 151
19.01 96:26 -0.28 20...Nc5 21.Nxc5 dxc5 22.f5 Bd6
23.Kh1 Rfd8 24.fxe6 fxe6 25.Qc4 Qe7
26.h3 Kh7 27.Kg1 Bb5 28.Qa2 Be5
29.Rxd8 Rxd8 30.Nxb5 Rd1+ 31.Rf1 Rxf1+
32.Kxf1 axb5 33.Ke2 g6 (884.252.554) 152
So I was wondering
- a) do other programs choose the variation 21. Bd4 Nxb3 22. Bb6 Qd7 23. cxb3 and
- b) am I evaluating this new 'main line' of the Rybka Modified.bkt correctly as better for White?
- c) is this caused by imperfect eval of doubled pawns? Or also underestimating importance of preserving a5?
23.Qh5 Kf8 24.Qg4 g6 25.f5 Bg5