arasan8.epd

[Steve Maughan]

Jon,

drop
"Gulko-Saidy, American op Los Angeles 1987";

I think it would be a mistake to drop Gulko-Saidy it is 100% sound. Kurt Utzinger also shows that Shredder 10.1 has no problems with the position. I also like it because Rxd5 is the only move that clearly wins. Yes it's hard and will not be solved by most programs within a minute, even on fast hardware - but if the test suite is to be useful for a number of years there needs be positions that will show the strength on tomorrows tactical monsters.

Here is Glaurung's analysis on my 2 Ghz laptop:

[d]r4r1k/1bq2pp1/2p1p2p/1pPnP1BQ/4B2P/1p4P1/P4P2/3RR1K1 w - -

Engine: Glaurung 2.0.1 (512 MB)
by Tord Romstad

16.01 0:30 +1.19 1.Bxh6 f5 2.Bxd5 cxd5 3.Be3+ Kg8
4.axb3 Ra2 5.Re2 Ra3 6.Rb2 Rfa8
7.Bd4 Qf7 8.Qxf7+ Kxf7 9.Kg2 Bc6
10.f3 Ke7 (44.495.737) 1437

17.01 0:46 +1.21 1.Bxh6 f5 2.Bxd5 cxd5 3.Be3+ Kg8
4.axb3 Ra2 5.Bd4 Rfa8 6.Re2 R2a3
7.b4 Bc6 8.Kg2 Qf7 9.Qxf7+ Kxf7
10.f3 Rb3 11.Rb2 Rxb2+ 12.Bxb2 (67.708.256) 1466

17.02 1:44 +3.19 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 bxa2 5.Bxh6 a1Q 6.Bxg7+ Kxg7
7.Qh7+ Kf6 8.Rxa1 Rxa1+ 9.Kg2 Raa8
10.g4 d4+ 11.Qxb7 Kxg6 12.Qe4+ Kf6
13.Qxd4+ Ke7 14.Qd6+ (155.288.725) 1490

18.01 2:23 +3.41 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 bxa2 5.Bxh6 a1Q 6.Bxg7+ Kxg7
7.Qh7+ Kf6 8.Rxa1 Rxa1+ 9.Kg2 d4+
10.Qxb7 Kxg6 11.Qe4+ Kf7 12.Qxd4 Ra7
13.Kf3 Rfa8 (214.652.402) 1493

[jdart]

I can take another look - my concern was whether Rxd5 scored a lot better than Bxh6 (Glaurung seems to indicate it does - and also selects Rxd5 much faster than other programs).

--Jon

[Steve Maughan]

Jon,

Toga also find Rxd5 quickly and gives it a big plus. I quite sure it's sound - axb3 also seems to be a little better than Bxh6, both of which give white an advantage but not a clearly winning advantage.

[d]r4r1k/1bq2pp1/2p1p2p/1pPnP1BQ/4B2P/1p4P1/P4P2/3RR1K1 w - -

Engine: Toga II 1.3x4 (1024 MB)
by Thomas Gaksch and Fabien Letouzey

9/42 0:00 +0.89 1.Bxh6 f5 2.Bxd5 cxd5 3.Be3+ Kg8
4.axb3 Ra2 5.b4 Qf7 6.Qxf7+ Kxf7
7.Bg5 Rfa8 (563.731)

9/42 0:02 +1.51++ 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 Kg8 5.Bxh6 Qc3 6.Bd2 Qh8
7.Qxh8+ Kxh8 8.axb3 (1.757.684) 719

10/42 0:03 +1.47 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 Kg8 5.Bxh6 Qc3 6.Bd2 Qh8
7.Qxh8+ Kxh8 8.Bc3+ Kg8 9.axb3 (2.308.917) 729

11/45 0:06 +1.57 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 Kg8 5.Bxh6 Qc3 6.Bd2 Qh8 7.Qg4 e5
8.axb3 (4.622.323) 742

12/51 0:25 +2.73++ 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 bxa2 5.Bxh6 a1Q 6.Bxg7+ Kxg7
7.Qh7+ Kf6 8.Rxa1 Rxa1+ 9.Kg2 d4+
10.Qxb7 Kxg6 11.Qe4+ Kf7 12.Qxd4 Ra7
13.c6 Ra2 (18.840.681) 737

13/51 0:35 +2.82 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 bxa2 5.Bxh6 a1Q 6.Bxg7+ Kxg7
7.Qh7+ Kf6 8.Rxa1 Rxa1+ 9.Kg2 d4+
10.Qxb7 Kxg6 11.Qe4+ Kf7 12.Qxd4 Ra2
13.Qd7+ Kf6 14.Qxb5 (26.527.385) 739

14/58 1:05 +3.11 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 bxa2 5.Bxh6 a1Q 6.Bxg7+ Kxg7
7.Qh7+ Kf6 8.Rxa1 Rxa1+ 9.Kg2 d4+
10.Qxb7 Kxg6 11.Qe7 Rf7 12.Qxe6+ Kg7
13.c6 Rc1 14.Qd5 (48.573.190) 744

15/58 1:53 +3.22 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 bxa2 5.Bxh6 a1Q 6.Bxg7+ Kxg7
7.Qh7+ Kf6 8.Rxa1 Rxa1+ 9.Kg2 d4+
10.Qxb7 Kxg6 11.Qe7 Rf7 12.Qxe6+ Kg7
13.h5 Rfa7 14.Qg4+ (84.571.553) 748

[Guetti]

Hiarcs 11.2 agrees with Toga and Shredder:

depth=12 time=5.21 node=4842327 speed=929786 score=+0.50 pv="axb3"

depth=12 time=6.83 node=6434386 speed=941939 score=+1.11 pv="axb3 Kg8 Bxh6 f5 Bxd5 cxd5 Qg6 Qf7 Qxf7+ Kxf7 Bg5 Ra2 b4 Rfa8 h5"

depth=13 time=16.26 node=15442650 speed=949849 score=+1.35 pv="axb3 f5 exf6 Nxf6 Bxf6"

depth=13 time=16.26 node=15442662 speed=949850 score=+2.71 pv="Rxd5"

depth=13 time=16.88 node=16068318 speed=952027 score=+2.71 pv="Rxd5 f5"

depth=14 time=19.63 node=18965941 speed=965974 score=+2.46 pv="Rxd5"

depth=14 time=29.64 node=28655021 speed=966866 score=+2.46 pv="Rxd5 f5 exf6 cxd5 fxg7+ Kxg7 Bxh6+"

depth=15 time=52.09 node=50808374 speed=975283 score=+2.68 pv="Rxd5 f5 exf6 cxd5 fxg7+ Qxg7 Bg6 bxa2 Bxh6 a1=Q Bxg7+ Kxg7 Qh7+ Kf6 Rxa1 Rxa1+ Kg2 d4+ Qxb7 Kxg6 Qe4+ Kf7 Qxd4 Ra7 c6"

depth=16 time=95.03 node=90748953 speed=954960 score=+2.93 pv="Rxd5"

depth=16 time=154.06 node=149160389 speed=968190 score=+3.16 pv="Rxd5 f5 exf6 cxd5 fxg7+ Qxg7 Bg6 bxa2 Bxh6 a1=Q Bxg7+ Kxg7 Qh7+ Kf6 Rxa1 Rxa1+ Kg2 d4+ Qxb7 Kxg6 Qe4+ Kf7 Qxd4 Ra7 c6 Raa8"

[jdart]

Ok, I will put back in Saidy-Gulko, replacing this one, to keep it at an even 300:

r5k1/b1pr1pp1/2p4p/4p2q/Q1P5/2B1P1P1/P2NbP1P/1R2R1K1 b - - bm Rxd2; id "arasan5.28";

(This position is pretty easy).

--Jon

[Eelco de Groot]

Ok, I will put back in Saidy-Gulko, replacing this one, to keep it at an even 300:

r5k1/b1pr1pp1/2p4p/4p2q/Q1P5/2B1P1P1/P2NbP1P/1R2R1K1 b - - bm Rxd2; id "arasan5.28";

(This position is pretty easy).

--Jon

I'm still >70% convinced that the Glaurung Mjolnir settings are actually stronger than the Defaults. It may be processor dependant but I think here on my computer there is a good chance that the settings are stronger. In some of these Arasan 8 testpostions the defaults, or settings related to the defaults, will have much better solution times though. So what does that mean? That for some positions there is actually a high chance of an inverse relation solution time/speed <-> playing strength? It is not impossible I think, when sometimes some search instabilities will bring the solution time down but hurt in actual playing strength. For instance in this Saidy - Gulko position the default is a lot faster than the alternative Mjolnir settings. In Shredder - Rybka the same. But I do not trust all the variations that appear, and I also know that the aggressive pruning that Tord has put in Glaurung will sometimes lead to a slightly unbalanced PV.

On the other hand I also made a new setting that is going back to Selectivity 6 AND is using a Maximum Razoring Depth of 4. It is slower in Rybka- Shredder but better than Mjolnir 001b at least in this Saidy - Gulko position. And the variations it is giving out seem pretty steady as well, close to Mjolnir 001b. But how would it do in games? It would need a lot of testing. Working Title: Mjolnir 001g



[d]r4r1k/1bq2pp1/2p1p2p/1pPnP1BQ/4B2P/1p4P1/P4P2/3RR1K1 w - -

Engine: Glaurung 2.0.1 Mjolnir 001g2 (256 MB, Athlon 2009 MHz)
by Tord Romstad

9.00 0:00 +0.72 1.Bxh6 f5 2.Bxd5 cxd5 3.axb3 gxh6
4.Qxh6+ Kg8 5.Qxe6+ Kg7 6.Qd6 Qf7 (166.230) 664

10.00 0:00 +0.82 1.Bxh6 f5 2.Bxd5 cxd5 3.axb3 gxh6
4.Qxh6+ Kg8 5.Qxe6+ Rf7 6.Qd6 Qc8
7.e6 (352.062) 683

11.00 0:01 +0.94 1.Bxh6 f5 2.Bxd5 cxd5 3.axb3 gxh6
4.Qxh6+ Kg8 5.Qxe6+ Rf7 6.Qd6 Qc8
7.e6 Rg7 (750.423) 686

12.01 0:01 +0.82 1.Bxh6 f5 2.Bxd5 cxd5 3.axb3 d4
4.Rxd4 Qc6 5.Bc1+ Kg8 6.f3 Qxc5
7.Bb2 Ra2 8.Re2 (1.334.099) 694

13.01 0:04 +0.37 1.Bxh6 f5 2.Bxd5 cxd5 3.axb3 gxh6
4.Qxh6+ Qh7 5.Qxe6 Ra6 6.c6 Bxc6
7.Rxd5 Re8 8.Qf6+ Kg8 9.Qg5+ Kf7 (3.336.457) 697

13.05 0:09 +0.72 1.axb3 Kg8 2.Rd4 g6 3.Bxg6 fxg6
4.Qxg6+ Qg7 5.Qxe6+ Kh7 6.Bd2 Qg6
7.Qd7+ Qf7 8.e6 Qxd7 9.exd7 (6.770.385) 695

14.01 0:13 +0.52 1.axb3 Kg8 2.Rd4 g6 3.Qxh6 Qxe5
4.Rdd1 Qh8 5.Qxh8+ Kxh8 6.Bh6 Rfe8
7.Bg5 Kg8 8.Kg2 Kg7 (9.623.841) 698

14.18 0:44 +0.78 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 bxa2 5.Bxh6 a1Q 6.Bxg7+ Kxg7
7.Qh7+ Kf6 8.Rxa1 Rxa1+ 9.Kg2 Bc6
10.Qh6 Rfa8 11.Qg5+ Kg7 (31.266.232) 709

15.01 0:53 +1.58 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 bxa2 5.Bxh6 a1Q 6.Bxg7+ Kxg7
7.Qh7+ Kf6 8.Rxa1 Rxa1+ 9.Kg2 Ra7
10.Qh6 Rc8 11.Qg5+ Kg7 12.Bh5+ Kh7
13.Qg6+ Kh8 14.Qxe6 (38.269.954) 711

16.01 1:19 +2.80 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 Kg8 5.Bxh6 Qc3 6.Bd2 Qh8 7.Qg4 Rf6
8.axb3 Qg7 9.Rxe6 Ra1+ 10.Be1 (57.235.291) 717

17.01 1:46 +2.41 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 Kg8 5.Bxh6 Qc3 6.Bd2 Qh8
7.Qxh8+ Kxh8 8.axb3 d4 9.Rxe6 Bd5
10.Re1 Bxb3 11.Re7 (76.972.799) 722

18.01 4:19 +3.41 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 bxa2 5.Bxh6 a1Q 6.Bxg7+ Kxg7
7.Qh7+ Kf6 8.Rxa1 Rxa1+ 9.Kg2 d4+
10.Qxb7 Kxg6 11.Qe4+ Kf7 12.Qxd4 Ra7
13.Kf3 Rfa8 (190.995.284) 735

19.01 9:05 +3.64 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 bxa2 5.Bxh6 a1Q 6.Bxg7+ Kxg7
7.Qh7+ Kf6 8.Rxa1 Rxa1+ 9.Kg2 d4+
10.Qxb7 Kxg6 11.Qe4+ Kf7 12.Qxd4 Ra7
13.Qb4 Rb7 14.c6 (407.868.522) 748

20.01 19:50 +4.11 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 Rf6 5.Bxf6 Qxf6 6.Qxh6+ Kg8
7.axb3 Bc8 8.c6 Ra2 9.f4 Qd4+ 10.Kh1 Qa7
11.Qg5 Qf2 12.Bh5+ Kh8 13.Qd8+ Kg7
14.Qe7+ (887.459.857) 745

Eelco

[Eelco de Groot]

P.S. at a depth of 21 and 22 plies:


21.01 41:49 +4.31 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 Rf5 5.Bxf5 exf5 6.axb3 d4
7.Bxh6 Qh7 8.Re8+ Rxe8 9.Qxe8+ Qg8
10.Qd7 Bd5 11.Bg7+ Kh7 12.Bxd4+ Qf7
13.Qxb5 Bxb3 14.c6 (1.884.727.107) 750

22.01 71:17 +4.78 1.Rxd5 f5 2.exf6 cxd5 3.fxg7+ Qxg7
4.Bg6 Rf5 5.Bxf5 exf5 6.Bxh6 Qh7
7.axb3 d4 8.Re8+ Rxe8 9.Qxe8+ Qg8
10.Qe5+ Kh7 11.Qc7+ Kxh6 12.Qb6+ Qg6
13.Qxb7 Qe6 14.Qxb5 (3.216.584.782) 751


This was the corresponding result for the Shredder - Rybka position, Mjolnir 001g needs 19 ply to find Bh6. Finding Rae1 Qe7 after no more than 2 minutes, and no switching between Qf2 and Rae1 in the variations before getting Bh6 is possibly just as important as finding Bh6 though.

Shredder Rybka:


[d]3q1rk1/pr1b1p1p/1bp2p2/2ppP3/8/2P1BN2/PPQ3PP/R4RK1 w - -

Engine: Glaurung 2.0.1 Mjolnir 001g2 (256 MB)
by Tord Romstad

7.00 0:00 -0.76 1.Rae1 Re8 2.exf6 Qxf6 3.Nd4 Qg6
4.Nf5 Be6 (88.307) 226

8.00 0:00 -0.11 1.Rae1 f5 2.Qf2 c4 3.Qg3+ Kh8 4.Bxb6 Qxb6+
5.Rf2 c5 (128.765) 284

9.00 0:00 -0.60 1.Rae1 Qe7 2.exf6 Qxf6 3.Nd4 Qg6
4.Nf5 Bxf5 5.Rxf5 Re8 6.Qd3 (204.849) 364

10.00 0:01 -0.43 1.Rae1 Rb8 2.exf6 Qxf6 3.Nd4 Qg6
4.Nf5 Bxf5 5.Qxf5 c4 6.Bxb6 Rxb6
7.Re2 (562.247) 529

11.01 0:01 -0.13 1.Rae1 Rb8 2.Qf2 c4 3.Bxb6 Rxb6
4.Qg3+ Kh8 5.exf6 Qxf6 6.Ng5 Bf5
7.Qh3 Qxg5 8.Rxf5 (1.139.580) 607

12.01 0:03 -0.07 1.Rae1 f5 2.Qf2 c4 3.Qg3+ Kh8 4.Bxb6 Qxb6+
5.Rf2 c5 6.Rd1 Rg8 7.Ng5 Qg6 (2.723.376) 683

13.01 0:07 -0.13 1.Rae1 f5 2.Qd2 f6 3.Bf2 fxe5 4.Nxe5 Qf6
5.Qd3 Kh8 6.Qc2 Rg8 7.Re2 Re8 8.Nxd7 Rxe2
9.Qxe2 Rxd7 (4.992.371) 702

14.01 0:13 0.00 1.Rae1 f5 2.Bg5 c4+ 3.Kh1 Qb8 4.Bf6 Bd8
5.Qc1 f4 6.Qxf4 Bxf6 7.Qxf6 Rxb2
8.Qg5+ Kh8 9.Qf6+ Kg8 10.Qg5+ (9.331.330) 716

15.01 0:17 0.00 1.Rae1 f5 2.Bg5 c4+ 3.Kh1 Qb8 4.Bf6 Bd8
5.Qc1 f4 6.Qxf4 Bxf6 7.Qxf6 Rxb2
8.Qg5+ Kh8 9.Qf6+ Kg8 10.Qg5+ (12.458.367) 724

16.01 0:57 +0.01 1.Rae1 c4 2.exf6 Qxf6 3.Bxb6 axb6
4.Ne5 Qd6 5.Qf2 Qc5 6.Qxc5 bxc5
7.Rf2 Kg7 8.a4 f6 9.Nxd7 Rxd7 10.a5 Kf7 (43.386.347) 758

17.01 2:01 +0.13 1.Rae1 Qe7 2.Qf2 Qe6 3.Qg3+ Qg4
4.Qxg4+ Bxg4 5.exf6 Rc8 6.Ne5 Be6
7.Rf3 Bc7 8.Nxc6 Bxh2+ 9.Kxh2 Rxc6
10.Rf2 Rc8 11.Kg3 (93.167.047) 763

18.01 2:56 +0.21 1.Rae1 Qe7 2.Qf2 Qe6 3.Qg3+ Qg4
4.Qxg4+ Bxg4 5.exf6 Rc8 6.Ne5 Be6
7.Rf3 Bc7 8.Nxc6 Bxh2+ 9.Kxh2 Rxc6
10.Rf2 Rcb6 11.Bxc5 Rxb2 12.Rxb2 Rxb2 (135.054.260) 763

19.01 6:14 +0.23 1.Rae1 Qe7 2.Qf2 Qe6 3.Qg3+ Qg4
4.Qxg4+ Bxg4 5.exf6 Rc8 6.Ne5 Be6
7.Rf3 Bc7 8.Nxc6 Bxh2+ 9.Kxh2 Rxc6
10.Rf2 Rb8 11.Kg3 Re8 12.Rd2 (287.823.896) 767

19.02 19:33 +1.21 1.Bh6 f5 2.Bxf8 Kxf8 3.Rae1 Kg8
4.Qd2 Be6 5.Qh6 Qf8 6.Qh5 c4+ 7.Kh1 Bd8
8.Ng5 Bxg5 9.Qxg5+ Qg7 10.Qd2 c5
11.Kg1 (900.808.649) 767



Eelco

[Eelco de Groot]

Running some matches now with Glaurung using the Mjolnir 001g2 settings. I will probably post the results in the Tournaments Forum. I will have to test the Glaurung settings against Belka again because that was the first opponent for Glaurung Mjolnir 001b and Glaurung 2.0.1 Default and I want to have the comparison match. Also the games are usually interesting enough in spite of the fact that Belka is a clone, I hope nobody has great difficulty with it if I post some results later. Glaurung Mjolnir 001g2 was very lucky in the first games keeping them to a draw although with its back against the wall a few times. Very tense!

Belka 1.8.11 - Glaurung 2.0.1 Mjolnir 001g2 first games at twelve minutes + 3 seconds Fischer bonus, slightly different from the first matches but I don't think the timecontrol is the deciding factor If you remember the result from Glaurung 2.0.1. Default at 8 minutes + 2 seconds I think you will agree this is not looking so bad:

Belka 1.8.11 - Glaurung 2.0.1 Mjolnir 001g2

1-1 (+0 =2 -0)

1½-½ (+1 =1 -0)

1-1 (+0 =2 -0)

Eelco

[BBauer]

Toga has no problem whith this position

Code: Select all

FEN: r4r1k/1bq2pp1/2p1p2p/1pPnP1BQ/4B2P/1p4P1/P4P2/3RR1K1 w - - 0 1 

Toga II 1.4 beta4a:
 10/40+	00:11	   9.627.335	1.677.246	+2,31	Rd1xd5 f7f5 e5xf6/ep c6xd5 f6xg7+ Qc7xg7 Be4g6 Kh8g8 Bg5xh6 Qg7c7 a2xb3 Rf8f6 Qh5g5 Qc7e7 Bg6f5+ Kg8h8
 11/46	00:17	  13.851.381	1.637.650	+2,47	Rd1xd5 f7f5 e5xf6/ep c6xd5 f6xg7+ Qc7xg7 Be4g6 b3xa2 Bg5xh6 a2a1Q Bh6xg7+ Kh8xg7 Qh5h7+ Kg7f6 Re1xa1 Ra8xa1+ Kg1g2 d5d4+ Qh7xb7 Kf6xg6 Qb7e4+ Kg6f7 Qe4xd4 Ra1a7 c5c6 Ra7a2
 12/46	00:24	  19.314.494	1.628.913	+2,58	Rd1xd5 f7f5 e5xf6/ep c6xd5 f6xg7+ Qc7xg7 Be4g6 b3xa2 Bg5xh6 a2a1Q Bh6xg7+ Kh8xg7 Qh5h7+ Kg7f6 Re1xa1 Ra8xa1+ Kg1g2 d5d4+ Qh7xb7 Kf6xg6 Qb7e4+ Kg6f7 Qe4xd4 Ra1a2 Qd4d7+ Kf7f6 Qd7xb5
 13/58+	00:49	  39.359.306	1.617.564	+3,09	Rd1xd5 f7f6 e5xf6 c6xd5 f6xg7+ Qc7xg7 Be4g6 b3xa2 Bg5xh6 a2a1Q Bh6xg7+ Kh8xg7 Qh5h7+ Kg7f6 Re1xa1 Ra8xa1+ Kg1g2 d5d4+ Qh7xb7 Kf6xg6 Qb7e7 Ra1a8 Qe7xe6+ Kg6g7 Qe6e5+ Kg7g8 c5c6 d4d3 Qe5d5+ Rf8f7 Qd5xd3

kind regards
Bernhard

[jdart]

I'm still >70% convinced that the Glaurung Mjolnir settings are actually stronger than the Defaults. It may be processor dependant but I think here on my computer there is a good chance that the settings are stronger. In some of these Arasan 8 testpostions the defaults, or settings related to the defaults, will have much better solution times though.

I don't think strength is very related to test suite performance. There is some correlation, because Arasan 7.4 (for example) does a lot worse than recent versions on this test, and it is surely weaker. But in general it is only one indicator and you have to look at other tests and game results as well.

Another use of test suites is to find anomalies. I have found a few bugs by observing that some position that should be easy was difficult to find, or other weirdnesses like unusual scores.