Some notes about Glaurung (long post)

[Peter Eizenhammer]

Hi, beeing a fan of Glaurung i would like to show some positions
where Glaurung(023) does not evaluate (and play) the best way.
The hope is, that this may be useful, destructive critizism is not what i intend,
but that should go without saying.
1)Wrong pawn (simple)
Glaurung does not know that bishop + wrong pawn is a draw, that should be easy to change and is really nasty to watch again and again, sorry.
FEN: [diag]3b4/8/6k1/7p/1P5P/8/8/5K2 w - - 0 44 [/diag]

Glaurung 0.2.3:
18 00:03 1.033.780 291.451 -3,53 44.Kg2 Kf5 45.Kf3 Ke6 46.b5 Kd7 47.b6 Kc8 48.b7+ Kb8 49.Kg3 Lf6 50.Kf4 Lxh4 51.Kf3 Le1 52.Kg2 Lc3 53.Kg3
19 00:06 1.643.600 285.843 -3,54 44.Kg2 Kf5 45.Kf3 Ke6 46.b5 Kd7 47.b6 Kc8 48.b7+ Kb8 49.Kg3 Lf6 50.Kf4 Lxh4 51.Kf3 Le1 52.Kg2 Lc3 53.Kg3 Ld4

Toga II 0.93:
22/29 00:06 4.370.997 742.418 0,00 44.Kg1 Lxh4 45.Kh1 Kf5 46.Kg2 Ke4 47.Kh1 Kf5

2)q+minor vs q = 1/2 (as a general rule, not always, of course)
Some may say this is a matter of taste, but in real games it can lead to completely wrong behaviour in endgames; easy to change
[diag]8/8/8/4Q3/1kb3K1/8/8/3q4 w - - 4 78 [/diag]
Glaurung 0.2.3:
9 00:03 -3,51 78.Kf4 Df1+ 79.Ke4 De2+ 80.Kf4 Df2+ 81.Ke4 Dc2+ 82.Ke3 Dc1+ 83.Ke4 De1+ 84.Kf5 Ld3+ 85.Ke6 Dd2

Toga II 0.93:
9/26 00:02 956.180 0 -0,93 78.Kf4 Df1+ 79.Kg3 Dg1+ 80.Kh4 Df2+ 81.Kg4 Le2+ 82.Kg5 Dg2+ 83.Kf6 Lc4 84.De7+ Kb5 85.Ke5

3)pawn ending, remote pawn
This one is more difficult, but Fritz8 shows, how it should be:
An easy win for black, not seen clearly by both Glaurung and Toga.
This kind of remote pawn is very important in real games. Some positions
let me think, that Glaurung does not always evaluate this high enough,
and it may be a problem of extensions, too.
[diag]8/3k4/8/3p1p2/p2P2p1/4P1P1/KR1r1PP1/8 b - - 6 49 [/diag]
Glaurung 0.2.3:
14 00:02 487.539 260.020 +0,54 49. ... Txb2+ 50.Kxb2 Kc6 51.Ka2 Kb6 52.Kb1 Kb5 53.Kc1 Kc4 54.Kb2 Kd3 55.Kb1 a3 56.Ka2 Ke2 57.Kxa3 Kxf2
17 00:07 1.856.544 248.035 0,00 49. ... Txb2+ 50.Kxb2 Kc6 51.Ka2 Kb6 52.Ka3 Kb5 53.f3 Ka5 54.fxg4 fxg4 55.Kb2 Kb4 56.Kc2 a3 57.Kb1 Kc4 58.Kc2 a2 59.Kb2

Toga II 0.93:
22/35 00:06 4.322.225 654.742 -1,39 49. ... Txb2+ 50.Kxb2 Kc6 51.Ka2 Kb5 52.Ka3 Ka5 53.f3 Kb5 54.Ka2 Kc4 55.Ka3 Kd3 56.fxg4 fxg4 57.Kxa4 Kxe3 58.Kb4 Kxd4 59.Kb3 Ke3 60.Kc2 d4 61.Kb3

Analysis by Fritz 8:
1...Txb2+ 2.Kxb2 Kc6 3.Ka2 Kb6 4.f3 Kb5 5.Kb2
? (-0.61) Tiefe: 7/12 00:00:00 5kN
1...Txb2+!
-+ (-2.70) Tiefe: 16/37 00:00:02 1713kN
1...Txb2+!
-+ (-6.25) Tiefe: 19/33 00:00:09 7301kN
1...Txb2+ 2.Kxb2 Kc6 3.Kc2 Kb5 4.Kc3 Ka5 5.f3 Kb5 6.f4 Ka5
-+ (-7.22) Tiefe: 20/39 00:00:17 13659kN

4)King safety
Black king is in real trouble, not at once, but for sure.
Glaurung does not evaluate this high enough imo.
[diag]rn2kb1r/2qb3p/p4p1p/1p6/2pP4/2N2N1P/1P2BPP1/R2Q1RK1 w kq - 0 15[/diag]
Glaurung 0.2.3:
9 00:05 799.358 145.735 -0,03 15.Dd2 Lg7 16.De3+ Kd8 17.De4 Dc6 18.d5 Dd6 19.Tfe1 Tg8 20.Dxh7
11 00:26 3.797.209 143.204 +0,33 15.Dd2 Lg7 16.De3+ Kd8 17.De4 Dc6 18.d5 Dd6 19.Sd4 Te8 20.Dxh7 De7

Toga II 0.93:
9/27 00:07 2.262.618 368.542 +1,79 15.b3 Tg8 16.Dc2 Db7 17.Tae1 Kd8 18.Kh1 Lc6 19.bxc4 bxc4 20.Lxc4 Lxf3 21.gxf3 Dxf3+ 22.Kh2
11/36 00:39 14.031.317 363.745 +2,11 15.b3 Tg8 16.bxc4 Lxh3 17.Se1 Lb4 18.Sd5 Dd6 19.Sxb4 Dxb4 20.cxb5 Dd6 21.Lh5+ Ke7

5)pieces without effect, mobility?
White queen and rook are "blind", hard to see, how they could get any effect.
This cannot be evaluated by just a single criterion, so maybe kind of difficult.
But iot should be clear, that the doubled queen and rook cannot reach nothing here,
black shield is absolutely solid. As a result white has too little forces to
prevent black from winning on the queen side.
[diag]r4rk1/p2qb1p1/2n1b2p/1p1p1p1B/2pP4/2N1P1P1/PP1N1PPQ/R4K1R b - - 1 21 [/diag]
Glaurung 0.2.3:
17 07:29 69.739.838 155.365 +0,21 21. ... Sb4 22.Le2 Sc6 23.Ld1 Sb4 24.Tb1 Sd3 25.Lc2 Tac8 26.a3 Tfe8 27.Kg1 Ld6 28.Lxd3 cxd3 29.Tc1 Lf7

Toga II 0.93:
14/45 11:26 269.552.927 393.398 -1,48 21. ... a5 22.Sxb5 Db7 23.a4 Sa7 24.Ke1 Sxb5 25.axb5 Dxb5 26.Ta2 a4 27.Le2 Ld7 28.Dh5 Db6

6)activity, rooks on second, remote pawn
Crafty with an excellent performance here, black is close to winning.
Glaurung feels much too good, in a passive and quite hopeless position.
[diag]2r5/3b1pk1/p5p1/1p1Pp1p1/4P3/PR3P1P/2r1NKP1/7R b - - 2 36 [/diag]
Glaurung 0.2.3:
16 00:39 9.149.281 232.174 +0,04 36. ... Td2 37.Te1 Tcc2 38.g4 Kf6 39.Ke3 Ke7 40.Tbb1 Kd6 41.Tbd1 Txd1 42.Txd1 f6 43.Td3 a5 44.Sc3
19 04:36 62.485.529 226.178 -0,56 36. ... Td2 37.Te1 Tcc2 38.Ke3 b4 39.g4 Lb5 40.Sc1 f5 41.gxf5 gxf5 42.exf5 bxa3 43.Txa3 Txd5 44.Sd3 Tc4 45.Tb3 Kf6 46.Sb4 Tdd4

Crafty-1.2 Cito:
12 00:11 5.946.163 517.057 -1,02 36. ... Td2 37.g4 f5 38.Te1 Tcc2 39.a4 Kf6 40.a5 Ta2 41.Tc3 Txa5 42.Ke3
16 04:01 130.933.686 543.293 -1,58 36. ... Td2 37.g4 f5 38.Te1 Tcc2 39.a4 Kf6 40.a5 Ta2 41.Tbb1 Txa5 42.Ta1 Txa1 43.Txa1 Lc8 44.Ke3 Tc2

Analysis by Fritz 8:
1...Kf8 2.Ke3 Ke7 3.g3 Kd6 4.Kd3 b4 5.Ke3 Lb5
? (-0.79) Tiefe: 7/9 00:00:00
1...Td2 2.Te3 f5 3.Te1 f4 4.Tb3 Tcc2 5.a4 Kf6
-+ (-1.68) Tiefe: 17/43 00:04:44 250412kN

7)activity, mobility
Again Glaurung does not feel, that the black pieces are very superior in mobility and effect.
The position is a black dream, Glaurung is too materialistic, as it seems.
[diag]2b1r1k1/p3r1qp/2pb1p2/p2p1ppQ/P2P4/1PN1P1N1/2P2PPP/RR4K1 b - - 1 23 [/diag]
Glaurung 0.2.3:
15 00:34 5.575.555 161.901 0,00 23. ... f4 24.exf4 Lxf4 25.Sge2 Ld6 26.Sg3 Df7 27.Dh6 Dg7 28.Dh5
17 04:03 40.626.706 166.641 -0,16 23. ... f4 24.exf4 Lb4 25.Df3 gxf4 26.Sh5 Dg5 27.Sa2 Ld6 28.g3 Lf5 29.h4 Dh6 30.Tc1 La3 31.Tf1 Te2

Crafty-1.2 Cito:
14 03:58 117.167.602 492.300 -0,97 23. ... Dg6 24.Dxg6+ hxg6 25.Sf1 f4 26.Tc1 La6 27.Sd1 Lxf1 28.Kxf1 fxe3 29.Sxe3 Te4 30.c4 Txd4 31.cxd5 cxd5 32.g3

Analysis by Fritz 8:
1...Dg6 2.Dd1 f4 3.Sf1 fxe3 4.Sxe3 f5 5.h3 f4 6.Sg4 Lf5 7.Tc1 h5
? (-1.21) Tiefe: 14/34 00:00:30 26645kN
1...Dg6 2.Dd1 f4 3.Sf1 fxe3 4.Sxe3 f5 5.h3 f4 6.Sg4 Lf5 7.Tc1 h5
? (-1.33) Tiefe: 15/37 00:01:12 64048kN

8)dead piece
A hard one, u might disagree here, but:
bh2 is dead. I would be afraid of real trouble in this pos, but Gl feels quite good.
Again about mobility? I dont know, but no engine should tell that white is better here.
[diag]2br1r1k/2q1b3/p1n4p/1pPp2p1/3P1p2/P1N2N1P/3Q1PPB/1R2R1K1 w - g6 0 32[/diag]
Glaurung 0.2.3:
14 01:18 12.595.873 160.776 +0,38 32.Sa2 Lf5 33.Tb3 Lf6 34.Sb4 Sa5 35.Tc3 Dg7 36.Td1 Le4 37.Db2 Sc4 38.Da1 a5

Toga II 0.93:
13/30 01:13 28.065.398 386.729 -0,15 32.Se5 Sxe5 33.Txe5 Lf6 34.Tee1 Lf5 35.Tb4 Da5 36.Dc1 Tf7 37.h4 Te7 38.Txe7 Lxe7 39.hxg5 hxg5

Analysis by Fritz 8:
1.Ta1!
? (-0.42) Tiefe: 9/26 00:00:02 1075kN
1.Sa2 Lf5 2.Tbd1 Lf6 3.Sb4 Le4 4.Dc3 Db7 5.Td2 Kg7 6.Sxc6 Dxc6 7.Se5 Lxe5
? (-0.57) Tiefe: 17/44 00:42:02 2175174kN

Ok, this has to do for now.
Peter

[Tord Romstad]

Hi Peter!

Hi, beeing a fan of Glaurung i would like to show some positions
where Glaurung(023) does not evaluate (and play) the best way.
The hope is, that this may be useful, destructive critizism is not what i intend,
but that should go without saying.

Yes, it really goes without saying. This form of detailed feedback is very rare and extremely valuable when trying to improve an engine. Thanks a lot for your very thorough work!

You are right that there are numerous serious holes and weaknesses in Glaurung's evaluation function, and I think this is the most important reason why Glaurung is so far behind the best amateur engines. At the moment, Glaurung is a rather slow program with a primitive and badly tuned evaluation function, but with a relatively efficient, stable and bug-free search.

1)Wrong pawn (simple)
Glaurung does not know that bishop + wrong pawn is a draw, that should be easy to change and is really nasty to watch again and again, sorry.(

A minor correction: Glaurung does know that KBP vs K with bishop + wrong rook pawn is a draw. It does not, however, understand that it is still a draw if the defending side has any pieces in addition to the king. Of course I agree that this should be fixed. You are right that this is relatively easy to change, but it is not so easy to change in an aesthetically pleasing way. In order to make a program play well in basic endgames, you have to add code for a lot of special cases like bishop + wrong rook pawn. This adds considerably to the size and complexity of the program, and is difficult to combine with my goal of keeping Glaurung's source code compact, simple and clean.

Nevertheless, I certainly will add knowledge about bishop + wrong rook pawn draws, as well as many other important endgame draws and wins, at some stage. This work has already started. Glaurung 0.2.3 has much more such specific endgame knowledge than 0.2.1. You may have noticed that it plays KPK endgames perfectly (by using a bitbase), that it recognizes some KRKP draws, and that it understands the third-rank defence and a couple of other common drawing techniques in KRPKR endgames. I am not entirely happy about the current look of this part of my code, however. I will probably look for a cleaner way to do it before I add lots of new special cases.

2)q+minor vs q = 1/2 (as a general rule, not always, of course)

Glaurung is already supposed to know this. Apparently I have a bug. Thanks for letting me know.

3)pawn ending, remote pawn
This one is more difficult, but Fritz8 shows, how it should be:
An easy win for black, not seen clearly by both Glaurung and Toga.
This kind of remote pawn is very important in real games. Some positions
let me think, that Glaurung does not always evaluate this high enough,
and it may be a problem of extensions, too.

I think it is purely an evaluation problem. Glaurung does not know about the concept of remote passed pawns at all. In fact it has no special knowledge about pawn endgames at all, except for the KPK bitbase and a very rudimentary unstoppable passed pawn detection. You are right, there is definitely scope for improvement here.

4)King safety
Black king is in real trouble, not at once, but for sure.
Glaurung does not evaluate this high enough imo.
[diag]rn2kb1r/2qb3p/p4p1p/1p6/2pP4/2N2N1P/1P2BPP1/R2Q1RK1 w kq - 0 15[/diag]

I agree. I find this to be a somewhat difficult problem. When the king is not yet under direct attack, it is not easy to formulate any good, general rules for whether a solid king shelter is important. But it seems that most other programs evaluate such positions better than Glaurung, so I suppose I should be able to improve this part of my eval.

5)pieces without effect, mobility?
White queen and rook are "blind", hard to see, how they could get any effect.
This cannot be evaluated by just a single criterion, so maybe kind of difficult.
But iot should be clear, that the doubled queen and rook cannot reach nothing here,
black shield is absolutely solid. As a result white has too little forces to
prevent black from winning on the queen side.
[diag]r4rk1/p2qb1p1/2n1b2p/1p1p1p1B/2pP4/2N1P1P1/PP1N1PPQ/R4K1R b - - 1 21 [/diag]

Glaurung actually likes white's position because it believes that white has a dangerous kingside attack. The bishop on h5, the queen on h2 and the rook on h1 are all considered as attacking pieces (the bishop by attacking squares near the black king, the queen and rook by being placed on an open file in front of the black king). A possible improvement in the king safety eval might be to reduce the bonus for queens and rooks on half-open files when the defending pawn on the file is protected by another pawn.

6)activity, rooks on second, remote pawn
Crafty with an excellent performance here, black is close to winning.
Glaurung feels much too good, in a passive and quite hopeless position.
[diag]2r5/3b1pk1/p5p1/1p1Pp1p1/4P3/PR3P1P/2r1NKP1/7R b - - 2 36 [/diag]

I wonder why Glaurung is so optimistic here. Perhaps it overevaluates white's passed pawn on d5. This position is definitely worth a closer examination.

7)activity, mobility
Again Glaurung does not feel, that the black pieces are very superior in mobility and effect.
The position is a black dream, Glaurung is too materialistic, as it seems.
[diag]2b1r1k1/p3r1qp/2pb1p2/p2p1ppQ/P2P4/1PN1P1N1/2P2PPP/RR4K1 b - - 1 23 [/diag]

Glaurung probably overestimates the importance of black's pawn structure defects.
Glaurung 0.2.3:
15 00:34 5.575.555 161.901 0,00 23. ... f4 24.exf4 Lxf4 25.Sge2 Ld6 26.Sg3 Df7 27.Dh6 Dg7 28.Dh5
17 04:03 40.626.706 166.641 -0,16 23. ... f4 24.exf4 Lb4 25.Df3 gxf4 26.Sh5 Dg5 27.Sa2 Ld6 28.g3 Lf5 29.h4 Dh6 30.Tc1 La3 31.Tf1 Te2

Crafty-1.2 Cito:
14 03:58 117.167.602 492.300 -0,97 23. ... Dg6 24.Dxg6+ hxg6 25.Sf1 f4 26.Tc1 La6 27.Sd1 Lxf1 28.Kxf1 fxe3 29.Sxe3 Te4 30.c4 Txd4 31.cxd5 cxd5 32.g3

Analysis by Fritz 8:
1...Dg6 2.Dd1 f4 3.Sf1 fxe3 4.Sxe3 f5 5.h3 f4 6.Sg4 Lf5 7.Tc1 h5
? (-1.21) Tiefe: 14/34 00:00:30 26645kN
1...Dg6 2.Dd1 f4 3.Sf1 fxe3 4.Sxe3 f5 5.h3 f4 6.Sg4 Lf5 7.Tc1 h5
? (-1.33) Tiefe: 15/37 00:01:12 64048kN

8)dead piece
A hard one, u might disagree here, but:
bh2 is dead. I would be afraid of real trouble in this pos, but Gl feels quite good.
Again about mobility? I dont know, but no engine should tell that white is better here.
[diag]2br1r1k/2q1b3/p1n4p/1pPp2p1/3P1p2/P1N2N1P/3Q1PPB/1R2R1K1 w - g6 0 32[/diag]

I don't disagree, but such positions tends to be difficult to evaluate correctly. It is not hard to imagine a slightly different position where white could liberate the bishop by playing h4 at the right moment, and where black's exposed king is more important than the temporarily reduced mobility of the white bishop.

You mention mobility in several of your positions. In fact, Glaurung currently does not evaluate mobility at all (except for some very simple code for bishops). I have had some experimental versions which did evaluate mobility, but they never worked well. Their play looked more sound and intelligent than the versions without mobility, but surprisingly the results were worse. I still think I should add some sort of mobility evaluation to Glaurung, I just haven't found the right way to do it yet.

Once again, thank you very much for your work! I have saved all your positions for use in future testing.

Tord

[Richard Pijl]

2)q+minor vs q = 1/2 (as a general rule, not always, of course)
Some may say this is a matter of taste, but in real games it can lead to completely wrong behaviour in endgames; easy to change

I've played with this one, and it is dangerous to assume this without conditions. Using the Shredder endgame oracle I've found that there are many positions where the stronger side still wins (up to 50% of the cases with fixed king/king configurations). Only when the defending side is to move and not in check this percentage is decreasing significantly (close to 0), which is not the case in the sample position.
Richard.

[Peter Eizenhammer]

Hello, some more positions for Glaurung:
(These positions are all described from a human player's point of view. I know that engines have a different view
and am not so naive to think: change some bytes and its ok. Just to clarify that... )

1) passer, active pieces, equal?!
3r4/p2r1k2/P1RPq1p1/5p1p/3RpQ2/6PP/5P1K/8 w - h6 0 56
[diag]3r4/p2r1k2/P1RPq1p1/5p1p/3RpQ2/6PP/5P1K/8 w - h6 0 56[/diag]

White pawn d6 is strong (ok, he may get lost, we dont know exactly),
and white initiative should be there, no matter how big the advantage is.
Glaury sees an equal position (and may find something after long thinking,
but here it is about evaluation).

Glaurung 0.2.3:
6 00:00 13.280 428.387 -0,09 56.h4 Kg7 57.Kg1 Kg8 58.Td2 Kg7
7 00:00 45.933 367.464 -0,08 56.h4 Kg8 57.Dg5 De5 58.De3 Db5 59.Dc3
14 01:22 14.485.852 176.119 +0,01 56.Tb4 e3 57.Dxe3 Dxe3 58.fxe3 Ke6 59.Tb7 Txb7 60.axb7 Tb8 61.h4 Kd7 62.Tc7+ Kxd6 63.Tf7 a5


2) Active pieces
Black is ahead in development, has strong bishops, rook on open file,
an advanced strong pawn c4, creating a field d3 for knight.
In addition there is a pawn majority on queenside, which gives chances for a passer there . A first glance should give more than just a little advantage, later analysis may show, that things are not so easy, but black is certainly clearly better.
r2r2k1/pb3pp1/1p2qb1p/4nN2/2p1P3/P7/1PQ1BPPP/R1B1R1K1 w - - 0 21
[diag]r2r2k1/pb3pp1/1p2qb1p/4nN2/2p1P3/P7/1PQ1BPPP/R1B1R1K1 w - - 0 21[/diag]

Analysis by Glaurung 0.2.3:
21.f3 Sd3 22.Td1 Tac8 23.Tb1 Sxc1 24.Tbxc1 Txd1+ 25.Txd1
= (-0.25) Tiefe: 8 00:00:01 124kN
21.f3 Sd3 22.Td1 La6 23.Tb1 Sxc1 24.Txd8+ Txd8 25.Txc1 Dc6 26.Kh1 b5 27.f4 Lb7 28.Sg3
? (-0.31) Tiefe: 14 00:02:33 17755kN

Compare Fritz8:
Analysis by Fritz 8:
21.Se3 Sd3 22.Td1 Tac8 23.f3 Sxc1 24.Taxc1 b5 25.Lf1
-+ (-1.50) Tiefe: 8/21 00:00:00 371kN
21.Lf4!
? (-1.17) Tiefe: 9/23 00:00:05 811kN
21.f3 Sd3 22.Td1 Tac8 23.Tb1 Tc5 24.g4 De5 25.Le3 Lg5 26.b4 Tc6 27.Lxd3
? (-1.10) Tiefe: 14/39 00:03:10 96598kN

3) Open file, active pieces in good coordination -> strong attack
In some cases Glaurung gives too high value to rook and queen on open file,
even when there is no connection with the rest of the pieces, so that the attack has to fail:
Here it is the other way round, white has an excellent position with active pieces, open g file and the connectivity between all the pieces is excellent. The last criterion of connectivity may not exist at all,
but it is certainly the reason, why sometimes two rooks on open file are bad, and sometimes a single rook can be crushing, if he gets support of additional pieces, just as here.
2rqrnk1/1p2b1pp/p1p1p3/3pB2n/3P1P2/2NBP2P/PPQ5/2KR3R w - - 0 1
[diag]2rqrnk1/1p2b1pp/p1p1p3/3pB2n/3P1P2/2NBP2P/PPQ5/2KR3R w - - 0 1[/diag]
Glaurung 0.2.3:
10 00:07 1.084.505 147.672 +0,12 1.Tdg1 Sf6 2.Dg2 g6 3.Dg5 S6d7 4.Dh6 Sxe5 5.fxe5 Db6
11 00:12 1.744.337 145.361 +0,15 1.Tdg1 Sf6 2.Dg2 g6 3.Dg5 S6d7 4.Dh6 Lf6 5.Lxf6 Dxf6 6.Kb1
12 00:26 3.802.724 143.753 0,00 1.Tdg1 Ld6 2.Lxh7+ Sxh7 3.Dg6 Lxe5 4.fxe5 Dh4 5.Tg4 Df2 6.Dxh5 Dxe3+ 7.Kb1 Te7
12 00:39 5.617.161 144.203 +0,21 1.Thg1 Sf6 2.Dg2 g6 3.Dg5 S6d7 4.Dh6 Sxe5 5.fxe5 Db6 6.Tdf1 Kh8
13 01:10 10.094.418 143.533 +0,19 1.Thg1 Sf6 2.f5 exf5 3.Lxf5 Ta8 4.Db3 b6 5.e4 g6 6.exd5 Sxd5 7.Kb1
13 01:36 13.780.656 143.594 +0,33 1.Tdg1 Sf6 2.Dg2 g6 3.h4 S6d7 4.h5 Sxe5 5.fxe5 g5 6.h6 Lb4 7.Tf1 Lxc3 8.bxc3
14 03:43 31.781.195 142.237 -0,03 1.Tdg1 b5 2.Dg2 g6 3.f5 exf5 4.Lxf5 Ta8 5.Lg4 Sf6 6.e4 dxe4 7.Sxe4 Kh8
14 04:45 40.603.541 142.312 +0,14 1.Thg1 Sf6 2.f5 exf5 3.Lxf5 Ta8 4.Db3 b6 5.e4 g6 6.exd5 Sxd5 7.Kb1 Lg5
15 09:44 83.619.147 143.237 +0,34 1.Thg1 Ld6 2.Tg5 Lxe5 3.Txh5 Lf6 4.Lxh7+ Sxh7 5.Dxh7+ Kf7 6.Tg1 Th8 7.Dg6+ Ke7 8.Dg4 Txh5 9.Dxh5 De8

16 15:47 135.395.501 143.013 +0,28 1.Thg1 Ld6 2.Tg5 Lxe5 3.Txh5 Lf6 4.Lxh7+ Sxh7 5.Dxh7+ Kf7 6.Tg1 Th8 7.Dg6+ Ke7 8.Txh8 Dxh8 9.Dg4 c5 10.Te1
17 36:25 314.948.140 144.127 +0,37 1.Thg1 Ld6 2.Tg5 Lxe5 3.Txh5 Lf6 4.Lxh7+ Sxh7 5.Txh7 c5 6.Dg6 cxd4 7.Dh5 Kf8 8.exd4 Dc7 9.Kb1 Dxf4 10.Th8+ Ke7

The +0.37 at depth 17 are ok, but this eval does not show, that black is certainly in deep troubles.

4) remote pawn (again)
taken from a game Kramnik - Korchnoj, where Kramnik knew of course, that
the majority on the queenside is a vital advantage, maybe even decisive.
Glaurung does see 0.10 at depth 17, thats low, but we allready had this.
For positional play the concepts of majority and as a result the remote passer
are really of highest importance, so it certainly would be good to implement this more generally, to have a better (not perfect of course) eval for a whole bunch of positions.

5k2/pR3ppp/3rb3/8/8/P3KB2/1P4PP/8 b - - 0 33
[diag]5k2/pR3ppp/3rb3/8/8/P3KB2/1P4PP/8 b - - 0 33[/diag]
Glaurung 0.2.3:
17 01:17 20.021.360 260.599 +0,10 33. ... Td7 34.Txd7 Lxd7 35.Kf4 Ke7 36.Ke5 Le6 37.Le4 f6+ 38.Kd4 g6 39.Ld5 Kd6 40.Lxe6 Kxe6 41.Ke4 f5+ 42.Kd4 Kf6 43.b3

5) King, development
From a Kramnik - Karpov game, white development is superior, black king
lost castling , in short terms: White is better.
There are some lines to calculate, too, making things kind of complicated,
but the eval is/should be clear. Glaury feels very good as black, certainly unlike Karpov did

r1bq1k1r/4bppp/p3pn2/n2P4/Bp3B2/2N2N2/PP3PPP/R2QR1K1 b - - 0 14
[diag]r1bq1k1r/4bppp/p3pn2/n2P4/Bp3B2/2N2N2/PP3PPP/R2QR1K1 b - - 0 14[/diag]

Glaurung 0.2.3:
9 00:03 503.411 167.803 -0,04 1. ... exd5 2.Se2 Kg8 3.Tc1 Sc4 4.Dd4 Lf5 5.Sg5 Tc8
14 04:01 36.549.061 151.665 -0,05 1. ... exd5 2.Se2 Lc5 3.Sed4 Kg8 4.Lg5 Db6 5.Lxf6 Dxf6 6.Te8+ Lf8 7.Dd3 Sc4 8.b3 Sb6
15 08:28 76.757.916 151.149 +0,08 1. ... exd5 2.Se2 Lc5 3.Sg3 Kg8 4.Dc2 Db6 5.Lg5 Ld7 6.Lxd7 Sxd7 7.Tad1 Db7 8.Se5 Sxe5 9.Dxc5

I tried to avoid the word "mobility" this time, but in the cases where "active pieces"
or "better development" are used, mobility may not be completely off topic,
but I have to leave this to the expert(s).

I dont know, if people allready noticed, that with Glaurung there is another open source engine
at a level of about Crafty, better in some respects, and for sure worse in some others, but it is certainly amazing. (but u cannot believe a fan, of course )

Peter

[Mihail Croitor]

its problem of all engines. 2 simple examples:
[diag]k6r/P7/8/1K6/4R3/8/8/6B1 w - - 0 0 bm ;=[/diag]
about this position Wildcat 4 said:
7 00:04 3.379 711 +7,17 e4c4
7 00:04 20.538 4.323 +7,17 e4c4
8 00:04 20.539 4.324 +7,17 e4c4
8 00:04 37.918 7.982 +7,18 :) g1c5
8 00:04 58.595 12.335 +7,18 g1c5 h8c8
8 00:05 94.759 18.951 +7,18 g1c5 h8c8
9 00:05 130.751 26.150 +7,18 g1c5 h8c8 e4e6 c8d8 e6e7 d8c8 b5b4

Ufim 600:
2 00:00 47 0 +8,00 Re4e7 Rh8c8
3 00:00 331 0 +8,08 Re4e7 Rh8h5+ Kb5c6 Rh5h6+ Kc6d5
4 00:00 2.137 0 +8,15 Re4e7 Rh8h5+ Kb5a6 Rh5h6+ Bg1b6 Rh6f6
5 00:00 7.187 0 +7,99 Re4e7 Rh8h5+ Kb5c4 Rh5c5+ Kc4d3 Rc5c3+ Kd3e4 Rc3a3
5 00:00 11.858 0 +8,28 Bg1e3 Rh8f8 Re4e7 Rf8f5+ Kb5c4 Rf5a5
6 00:00 20.299 0 +8,10 Bg1e3 Rh8c8 Re4e7 Rc8c5+ Kb5b6 Rc5c6+ Kb6a5 Rc6c8
7 00:00 51.630 0 +8,16 Bg1e3 Rh8c8 Re4e5 Rc8g8 Re5e7 Rg8g5+
8 00:00 146.562 0 +8,15 Bg1e3 Rh8g8 Be3d4 Rg8g5+ Bd4c5 Rg5g8 Re4e7 Rg8c8 Bc5b6
9 00:01 331.899 0 +8,23 Bg1e3 Rh8g8 Be3b6 Rg8g5+ Bb6c5 Rg5g8 Re4e7 Rg8c8 Re7h7 Rc8d8
10 00:02 730.440 374.000 +8,23 Bg1e3 Rh8g8 Be3d4 Rg8g5+ Bd4c5 Rg5g8 Bc5b6 Rg8g5+
11 00:03 1.499.383 400.000 +8,29 Bg1e3 Rh8h5+ Be3c5 Rh5h8 Bc5d4 Rh8g8 Bd4b6 Rg8f8 Kb5a6 Rf8g8

Glaurung 0.2.3:
7 00:00 16.996 339.920 +4,95 Bg1c5 Rh8c8 Re4a4 Rc8e8 Ra4a6 Re8e2 Kb5b6
8 00:00 37.751 311.991 +4,96 Bg1c5 Rh8d8 Re4e7 Rd8c8 Bc5d4 Rc8c5+ Kb5a6 Rc5c6+ Bd4b6 Rc6c2
9 00:00 48.691 345.326 +4,95 Bg1c5 Rh8d8 Re4e7 Rd8c8 Re7e6 Rc8d8 Re6a6 Rd8d2 Kb5b6
10 00:00 82.262 341.336 +4,96 Bg1c5 Rh8d8 Re4g4 Rd8c8 Rg4d4 Rc8e8 Rd4d7 Re8c8 Kb5b4 Rc8e8
11 00:00 128.434 376.639 +4,98 Bg1c5 Rh8d8 Re4g4 Rd8c8 Rg4d4 Rc8h8 Rd4f4 Rh8c8 Kb5c4 Rc8d8 Rf4f7
12 00:00 229.170 408.502 +4,96 Bg1c5 Rh8d8 Bc5b6 Rd8h8 Kb5a6 Rh8d8 Re4e7 Rd8h8 Re7d7 Rh8e8 Bb6c5 Re8e2
13 00:02 884.830 433.103 +5,04 Bg1c5 Rh8d8 Bc5b6 Rd8h8 Kb5a6 Rh8d8 Re4e7 Rd8h8 Re7d7 Rh8c8 Bb6c7 Rc8e8 Bc7d6
14 00:02 977.530 427.990 +5,00 Bg1c5 Rh8d8 Bc5b6 Rd8h8 Kb5a6 Rh8d8 Re4f4 Rd8c8 Rf4f7 Rc8e8 Rf7g7 Re8c8 Bb6c7 Rc8e8
15 00:03 1.526.242 426.801 +5,03 Bg1c5 Rh8d8 Bc5b6 Rd8h8 Kb5a6 Rh8d8 Re4f4 Rd8c8 Rf4f7 Rc8e8 Rf7g7 Re8c8 Bb6c7 Rc8e8 Bc7d6

Fruit 2.0:
8/21 00:00 26.028 0 +6,21 Bg1c5 Rh8c8 Re4e5 Rc8g8 Re5e7 Rg8d8 Re7f7 Rd8e8
9/21 00:00 34.619 0 +6,04 Bg1c5 Rh8c8 Re4e5 Rc8g8 Re5e7 Rg8c8 Re7e2 Rc8g8 Bc5d4
10/22 00:00 154.078 0 +6,03 Bg1c5 Rh8c8 Re4e5 Rc8g8 Re5e7 Rg8c8 Bc5d4 Rc8c5+ Kb5b6 Rc5e5 Re7f7
11/24 00:00 250.693 0 +6,03 Bg1c5 Rh8c8 Re4e5 Ka8b7 Re5e7+ Kb7a8 Re7e2 Rc8g8 Re2e6 Rg8c8 Bc5d4 Rc8f8
12/24 00:00 332.061 0 +6,06 Bg1c5 Rh8c8 Re4e5 Rc8d8 Re5e3 Rd8c8 Re3d3 Rc8xc5+ Kb5xc5 Ka8xa7 Rd3d7+ Ka7b8 Kc5c4 Kb8c8
13/28 00:01 715.064 0 +6,16 Bg1c5 Rh8c8 Re4e5 Rc8d8 Re5e3 Rd8g8 Re3d3 Rg8c8 Rd3d5 Rc8xc5+ Kb5xc5 Ka8xa7 Rd5d7+ Ka7b8 Kc5c4
14/28 00:01 1.125.038 879.121 +6,13 Bg1c5 Rh8c8 Re4a4 Ka8b7 Ra4h4 Rc8d8 Rh4h7+ Kb7a8 Rh7e7 Rd8c8 Re7e5 Rc8d8 Re5e3 Rd8g8 Re3e7
15/32 00:02 1.556.480 886.816 +6,15 Bg1c5 Rh8c8 Re4a4 Ka8b7 Ra4h4 Rc8d8 Rh4h7+ Kb7a8 Rh7e7 Rd8h8 Bc5d4 Rh8c8 Re7f7 Rc8c5+ Kb5b6 Rc5c6+ Kb6a5 Rc6c8
16/32 00:02 2.096.319 873.689 +6,15 Bg1c5 Rh8c8 Re4a4 Ka8b7 Ra4h4 Rc8d8 Rh4h7+ Kb7a8 Rh7e7 Rd8h8 Re7f7 Rh8c8 Bc5e3 Rc8d8 Kb5c4 Rd8d4+ Kc4c5 Rd4d7
17/34 00:03 2.736.710 898.802 +6,14 Bg1c5 Rh8c8 Re4a4 Ka8b7 Ra4h4 Rc8d8 Rh4h7+ Kb7a8 Rh7f7 Rd8c8 Rf7f4 Rc8d8 Rf4g4 Rd8e8 Rg4g7 Re8c8 Bc5d4 Rc8c5+ Kb5b4
18/34 00:04 3.499.946 906.732 +6,14 Bg1c5 Rh8c8 Re4a4 Ka8b7 Ra4h4 Rc8d8 Rh4h7+ Kb7a8 Rh7f7 Rd8e8 Rf7d7 Re8c8 Rd7d4 Rc8g8 Kb5c4 Rg8e8 Rd4d7 Re8e4+ Kc4b5 Re4b4+ Kb5c6

Shredder 9 evaluate this position at +8.11 (20/40)
But i see only draw.

[diag]6b1/8/8/8/1p3k2/3K4/1P6/8 w - - 0 0 bm Kd4;=[/diag]
a theoretical position. all engines finded best move, but evaluated it...

Wildcat 4:
12 00:00 119.225 794.833 -4,98 d3d4 b4b3 d4d3
12 00:00 126.123 788.268 -4,98 d3d4 b4b3 d4d3
13 00:00 225.961 941.504 -4,98 d3d4 b4b3 d4d3
13 00:00 236.340 945.360 -4,98 d3d4 b4b3 d4d3
14 00:00 415.762 944.913 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2 h7d3 d2c1 d4e3 c1d1 e3d4
14 00:00 524.857 1.029.131 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2 h7d3 d2c1 d4e3 c1d1 e3d4
15 00:00 637.372 1.028.019 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2
15 00:01 668.987 1.045.292 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2
16 00:01 833.777 1.068.944 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2
16 00:01 869.966 1.074.032 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2
17 00:01 1.067.777 1.078.562 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2
17 00:01 1.121.388 1.088.726 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2
18 00:01 1.411.884 1.111.719 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2
18 00:01 1.504.231 1.122.560 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2
19 00:01 1.794.433 1.128.574 -5,08 d3d4 b4b3 d4d3 f4e5 d3d2 e5d4 d2c1 g8h7 c1d2

Ufim 600:
2 00:00 19 0 -5,61 Kd3d4 Bg8e6
3 00:00 85 0 -5,79 Kd3d4 Bg8e6 Kd4c5
4 00:00 459 0 -5,55 Kd3d4 Bg8e6 Kd4c5 b4b3
5 00:00 1.420 0 -5,74 Kd3d4 Bg8e6 Kd4c5 b4b3 Kc5d6
6 00:00 5.163 0 -5,65 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3e3 Bg8d5
7 00:00 10.636 0 -5,85 Kd3d4 b4b3 Kd4c3 Kf4e4 Kc3b4 Bg8e6 Kb4c5
8 00:00 33.479 0 -5,75 Kd3d4 b4b3 Kd4c5 Kf4e5 Kc5b5 Ke5d5 Kb5b6 Bg8e6
9 00:00 64.926 0 -5,87 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3e2 Ke5e4 Ke2d2 Ke4d4
10 00:00 113.608 0 -5,92 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3e2 Bg8h7 Ke2d2 Ke5d4 Kd2e2 Bh7d3+ Ke2f3
11 00:00 280.431 0 -6,01 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Ke5e4 Kc3b4 Ke4d3 Kb4a4 Kd3c4 Ka4a3
12 00:02 1.337.782 0 -6,00 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Ke5e4 Kc3b4 Ke4d3 Kb4a4 Bg8d5

Glaurung 0.2.3:
16 00:00 81.029 424.235 -3,71 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Bg8d5 Kc3d3 Bd5e6 Kd3c3 Ke5e4 Kc3b4 Ke4d4 Kb4b5 Be6d5 Kb5b4 Bd5c4
17 00:00 116.067 428.291 -3,72 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Bg8d5 Kc3d3 Bd5e6 Kd3c3 Ke5e4 Kc3b4 Ke4d4 Kb4b5 Be6d5 Kb5a4 Kd4c4 Ka4a3
18 00:00 170.350 447.112 -3,75 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Bg8d5 Kc3d3 Bd5e6 Kd3c3 Ke5e4 Kc3b4 Ke4d4 Kb4b5 Be6d5 Kb5a4 Kd4c4 Ka4a3 Kc4d4
19 00:00 264.504 455.256 -3,75 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Bg8d5 Kc3d3 Bd5e6 Kd3c3 Ke5e4 Kc3b4 Ke4d4 Kb4b5 Be6d5 Kb5a4 Kd4c4 Ka4a3 Bd5e4 Ka3a4
20 00:01 343.898 445.463 -3,75 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Bg8d5 Kc3d3 Bd5e6 Kd3c3 Ke5e4 Kc3b4 Ke4d4 Kb4b5 Be6d5 Kb5a4 Kd4c5 Ka4a3 Kc5c4 Ka3a4 Kc4d4
21 00:01 526.120 452.771 -3,75 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Bg8d5 Kc3d3 Bd5e6 Kd3c3 Ke5e4 Kc3b4 Ke4d4 Kb4b5 Be6d5 Kb5a4 Kd4c5 Ka4a3 Kc5c4 Ka3a4 Bd5e4 Ka4a3
22 00:02 900.254 468.150 -3,74 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Bg8d5 Kc3d3 Bd5e6 Kd3c3 Ke5e4 Kc3b4 Ke4d4 Kb4b5 Be6d5 Kb5a4 Kd4c5 Ka4a3 Kc5c4 Ka3a4 Bd5e6 Ka4a3 Kc4d4
23 00:03 1.359.421 463.175 -3,75 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Bg8d5 Kc3d3 Bd5e6 Kd3c3 Ke5e4 Kc3d2 Ke4d4 Kd2e2 Be6c4+ Ke2d2 Bc4d3 Kd2c1 Bd3e2 Kc1d2 Be2f1 Kd2c1 Bf1d3 Kc1d2 Bd3e4
24 00:04 1.878.432 461.985 -3,75 Kd3d4 b4b3 Kd4d3 Kf4e5 Kd3c3 Bg8d5 Kc3d3 Bd5e6 Kd3c3 Ke5e4 Kc3d2 Ke4d4 Kd2e2 Be6c4+ Ke2d2 Bc4d3 Kd2c1 Bd3e2 Kc1d2 Be2c4 Kd2d1 Bc4d5 Kd1e2 Bd5e4 Ke2d2

Fruit 2.0:
15/19 00:01 724.256 708.583 -3,99 Kd3d4 b4b3 Kd4d3 Kf4f3 Kd3d2 Kf3e4 Kd2e2 Bg8c4+ Ke2d2 Ke4d4 Kd2d1 Kd4d3 Kd1c1 Bc4d5 Kc1d1 Kd3d4
16/21 00:01 959.682 717.720 -4,20 Kd3d4 b4b3 Kd4d3 Kf4f3 Kd3d2 Kf3f2 Kd2d1 Kf2e3 Kd1e1 Bg8e6 Ke1d1 Be6g4+ Kd1c1 Bg4f3 Kc1b1 Bf3e4+ Kb1c1 Ke3d4
17/21 00:01 1.328.917 818.710 -4,31 Kd3d4 b4b3 Kd4d3 Kf4f3 Kd3d2 Kf3f2 Kd2d1 Kf2e3 Kd1e1 Bg8e6 Ke1d1 Be6g4+ Kd1c1 Bg4f3 Kc1b1 Ke3d2 Kb1a1 Bf3e4
18/22 00:02 1.691.677 847.510 -4,31 Kd3d4 b4b3 Kd4d3 Kf4f3 Kd3d2 Kf3f2 Kd2d1 Kf2e3 Kd1e1 Bg8e6 Ke1d1 Be6g4+ Kd1c1 Bg4f3 Kc1b1 Ke3d2 Kb1a1 Bf3d5 Ka1b1 Bd5e4+ Kb1a1
19/23 00:02 2.104.096 943.865 -4,43 Kd3d4 b4b3 Kd4d3 Kf4f3 Kd3d2 Kf3f2 Kd2d1 Kf2e3 Kd1e1 Bg8e6 Ke1d1 Be6g4+ Kd1c1 Bg4f3 Kc1b1 Ke3d2 Kb1a1 Bf3d5 Ka1b1 Bd5e4+ Kb1a1 Kd2e3
20/25 00:02 2.656.149 964.048 -4,43 Kd3d4 b4b3 Kd4d3 Kf4f3 Kd3d2 Kf3f2 Kd2d1 Kf2e3 Kd1e1 Bg8e6 Ke1d1 Be6g4+ Kd1c1 Bg4f3 Kc1b1 Ke3d2 Kb1a1 Kd2e2 Ka1b1 Bf3e4+ Kb1a1 Ke2e3
21/27 00:03 3.452.608 1.041.459 -4,43 Kd3d4 b4b3 Kd4d3 Kf4f3 Kd3d2 Kf3f2 Kd2d1 Kf2e3 Kd1e1 Bg8e6 Ke1d1 Ke3d3 Kd1c1 Be6d5 Kc1d1 Kd3e3 Kd1c1 Ke3e2 Kc1b1 Bd5e4+ Kb1a1 Ke2e3
22/31 00:04 4.121.759 1.060.109 -4,43 Kd3d4 b4b3 Kd4d3 Kf4f3 Kd3d2 Kf3f2 Kd2d1 Kf2e3 Kd1e1 Bg8e6 Ke1d1 Ke3d3 Kd1c1 Be6d5 Kc1d1 Bd5f3+ Kd1c1 Kd3e2 Kc1b1 Bf3e4+ Kb1a1 Ke2d3 Ka1b1 Kd3d2+ Kb1a1 Kd2e3
23/32 00:04 4.865.814 1.118.028 -4,43 Kd3d4 b4b3 Kd4d3 Kf4f3 Kd3d2 Kf3f2 Kd2d1 Kf2e3 Kd1e1 Bg8e6 Ke1d1 Ke3d3 Kd1c1 Be6d5 Kc1d1 Bd5f3+ Kd1c1 Kd3e3 Kc1b1 Ke3d2 Kb1a1 Bf3g2 Ka1b1 Bg2e4+ Kb1a1 Kd2e3

[Tord Romstad]

Hi Peter!

Hello, some more positions for Glaurung:

Great!

Thanks a lot for the new positions. Unfortunately I don't have much time for computer chess at the moment, but I save all your positions for future use. Some comments about your new positions below:

1) passer, active pieces, equal?!
3r4/p2r1k2/P1RPq1p1/5p1p/3RpQ2/6PP/5P1K/8 w - h6 0 56
[diag]3r4/p2r1k2/P1RPq1p1/5p1p/3RpQ2/6PP/5P1K/8 w - h6 0 56[/diag]

White pawn d6 is strong (ok, he may get lost, we dont know exactly),
and white initiative should be there, no matter how big the advantage is.
Glaury sees an equal position (and may find something after long thinking,
but here it is about evaluation).

Giving a big bonus for isolated, blocked passed pawns is often dangerous. Without a search, it is difficult to decide whether such a pawn is a strength or a weakness. Nevertheless I agree that white obviously have a clear advantage in this position, and I am disappointed that Glaurung does not understand it better. I will have to take a look at the values of the different components of the evaluation function in order to find out why Glaurung misevaluates this position so badly.

2) Active pieces
Black is ahead in development, has strong bishops, rook on open file,
an advanced strong pawn c4, creating a field d3 for knight.
In addition there is a pawn majority on queenside, which gives chances for a passer there . A first glance should give more than just a little advantage, later analysis may show, that things are not so easy, but black is certainly clearly better.
r2r2k1/pb3pp1/1p2qb1p/4nN2/2p1P3/P7/1PQ1BPPP/R1B1R1K1 w - - 0 21
[diag]r2r2k1/pb3pp1/1p2qb1p/4nN2/2p1P3/P7/1PQ1BPPP/R1B1R1K1 w - - 0 21[/diag]

The lack of understanding of development and piece activity is indeed one of Glaurungs main weaknesses. I have not yet found any simple and effective ways to evaluate this (Gothmog does this better, but the code is too complex and expensive for my taste). That this is a known weakness does not mean that posisitions which illustrate the problem are without value, of course. On the contrary, it is very useful to have several nice positions with which I can test attempted improvements.

Pawn majorities are an interesting subject. Glaurung does have an evaluation term for candidate passed pawns (new in version 0.2.3, if I recall correctly). This bonus currently depends only on the phase of the game (middle game or endgame) and the rank of the pawn. In your position, black's candidate passed pawn on c4 gets a bonus of 0.2 pawns (if I understand my own code correctly). However, white also has a candidate passed pawn, on e4. This pawn is only one rank less advanced, and is awarded a bonus of 0.12.

The true reason why black's queenside pawn majority is more important than white's kingside majority is probably that both kings are on the kingside. This means that it is more risky for white to advance his pawn majority in the middle game, and also that black can use his king to block a hypothetical white passed pawn on the e-file in the endgame more easily than white can block black's c pawn. Perhaps I should let my candidate passed pawn evaluation depend on the locations of the kings. This would make my evaluation slightly more expensive (because I can no longer store candidate passed pawn scores in the pawn hash table), but it may be worth the price.

3) Open file, active pieces in good coordination -> strong attack
In some cases Glaurung gives too high value to rook and queen on open file,
even when there is no connection with the rest of the pieces, so that the attack has to fail:
Here it is the other way round, white has an excellent position with active pieces, open g file and the connectivity between all the pieces is excellent. The last criterion of connectivity may not exist at all,
but it is certainly the reason, why sometimes two rooks on open file are bad, and sometimes a single rook can be crushing, if he gets support of additional pieces, just as here.
2rqrnk1/1p2b1pp/p1p1p3/3pB2n/3P1P2/2NBP2P/PPQ5/2KR3R w - - 0 1
[diag]2rqrnk1/1p2b1pp/p1p1p3/3pB2n/3P1P2/2NBP2P/PPQ5/2KR3R w - - 0 1[/diag]
Glaurung 0.2.3:
10 00:07 1.084.505 147.672 +0,12 1.Tdg1 Sf6 2.Dg2 g6 3.Dg5 S6d7 4.Dh6 Sxe5 5.fxe5 Db6
11 00:12 1.744.337 145.361 +0,15 1.Tdg1 Sf6 2.Dg2 g6 3.Dg5 S6d7 4.Dh6 Lf6 5.Lxf6 Dxf6 6.Kb1
12 00:26 3.802.724 143.753 0,00 1.Tdg1 Ld6 2.Lxh7+ Sxh7 3.Dg6 Lxe5 4.fxe5 Dh4 5.Tg4 Df2 6.Dxh5 Dxe3+ 7.Kb1 Te7
12 00:39 5.617.161 144.203 +0,21 1.Thg1 Sf6 2.Dg2 g6 3.Dg5 S6d7 4.Dh6 Sxe5 5.fxe5 Db6 6.Tdf1 Kh8
13 01:10 10.094.418 143.533 +0,19 1.Thg1 Sf6 2.f5 exf5 3.Lxf5 Ta8 4.Db3 b6 5.e4 g6 6.exd5 Sxd5 7.Kb1
13 01:36 13.780.656 143.594 +0,33 1.Tdg1 Sf6 2.Dg2 g6 3.h4 S6d7 4.h5 Sxe5 5.fxe5 g5 6.h6 Lb4 7.Tf1 Lxc3 8.bxc3
14 03:43 31.781.195 142.237 -0,03 1.Tdg1 b5 2.Dg2 g6 3.f5 exf5 4.Lxf5 Ta8 5.Lg4 Sf6 6.e4 dxe4 7.Sxe4 Kh8
14 04:45 40.603.541 142.312 +0,14 1.Thg1 Sf6 2.f5 exf5 3.Lxf5 Ta8 4.Db3 b6 5.e4 g6 6.exd5 Sxd5 7.Kb1 Lg5
15 09:44 83.619.147 143.237 +0,34 1.Thg1 Ld6 2.Tg5 Lxe5 3.Txh5 Lf6 4.Lxh7+ Sxh7 5.Dxh7+ Kf7 6.Tg1 Th8 7.Dg6+ Ke7 8.Dg4 Txh5 9.Dxh5 De8

16 15:47 135.395.501 143.013 +0,28 1.Thg1 Ld6 2.Tg5 Lxe5 3.Txh5 Lf6 4.Lxh7+ Sxh7 5.Dxh7+ Kf7 6.Tg1 Th8 7.Dg6+ Ke7 8.Txh8 Dxh8 9.Dg4 c5 10.Te1
17 36:25 314.948.140 144.127 +0,37 1.Thg1 Ld6 2.Tg5 Lxe5 3.Txh5 Lf6 4.Lxh7+ Sxh7 5.Txh7 c5 6.Dg6 cxd4 7.Dh5 Kf8 8.exd4 Dc7 9.Kb1 Dxf4 10.Th8+ Ke7

The +0.37 at depth 17 are ok, but this eval does not show, that black is certainly in deep troubles.

I am actually a bit happier about Glaurung's evaluation of this position than your previous positions. Note that Glaurung does think that white has a small advantage, despite the fact that black is a pawn up. In other words, it evaluates white's positional advantage to be worth more than a pawn. I agree that white's real advantage is even bigger, but for a simple and stupid engine like Glaurung, allowing very big positional scores will backfire more often than not.

I hope that Glaurung will eventually be able to undersand such positions better, but a lot of knowledge must be added before this happens.

4) remote pawn (again)
taken from a game Kramnik - Korchnoj, where Kramnik knew of course, that
the majority on the queenside is a vital advantage, maybe even decisive.
Glaurung does see 0.10 at depth 17, thats low, but we allready had this.
For positional play the concepts of majority and as a result the remote passer
are really of highest importance, so it certainly would be good to implement this more generally, to have a better (not perfect of course) eval for a whole bunch of positions.

5k2/pR3ppp/3rb3/8/8/P3KB2/1P4PP/8 b - - 0 33
[diag]5k2/pR3ppp/3rb3/8/8/P3KB2/1P4PP/8 b - - 0 33[/diag]

My comments about candidate passed pawns above apply here, too. White's pawn majority is more important than black's because the kings are located on the kingside. I think considering this when evaluating candidate passed pawns is really worth trying.

5) King, development
From a Kramnik - Karpov game, white development is superior, black king
lost castling , in short terms: White is better.
There are some lines to calculate, too, making things kind of complicated,
but the eval is/should be clear. Glaury feels very good as black, certainly unlike Karpov did

r1bq1k1r/4bppp/p3pn2/n2P4/Bp3B2/2N2N2/PP3PPP/R2QR1K1 b - - 0 14
[diag]r1bq1k1r/4bppp/p3pn2/n2P4/Bp3B2/2N2N2/PP3PPP/R2QR1K1 b - - 0 14[/diag]

Glaurung 0.2.3:
9 00:03 503.411 167.803 -0,04 1. ... exd5 2.Se2 Kg8 3.Tc1 Sc4 4.Dd4 Lf5 5.Sg5 Tc8
14 04:01 36.549.061 151.665 -0,05 1. ... exd5 2.Se2 Lc5 3.Sed4 Kg8 4.Lg5 Db6 5.Lxf6 Dxf6 6.Te8+ Lf8 7.Dd3 Sc4 8.b3 Sb6
15 08:28 76.757.916 151.149 +0,08 1. ... exd5 2.Se2 Lc5 3.Sg3 Kg8 4.Dc2 Db6 5.Lg5 Ld7 6.Lxd7 Sxd7 7.Tad1 Db7 8.Se5 Sxe5 9.Dxc5

This is another position where I am not entirely unhappy. After 1... exd5 2. Ne2 black is a pawn up, but Glaurung thinks white has full compensation for the pawn. Glaurung understands and evaluates the positional factors you mention above. Once again you could argue that the positional scores should have been even bigger, but I think this would make the engine misevaluate too many other positions.

I tried to avoid the word "mobility" this time, but in the cases where "active pieces"
or "better development" are used, mobility may not be completely off topic,
but I have to leave this to the expert(s).

Yes, a decent mobility eval should help to solve some of these problems. As mentioned in my previous message in this thread, straightforward mobility as used in most other programs doesn't seem to work well in Glaurung, for some mysterious reason.

I dont know, if people allready noticed, that with Glaurung there is another open source engine at a level of about Crafty, better in some respects, and for sure worse in some others, but it is certainly amazing. (but u cannot believe a fan, of course )

Thanks for the compliments!

The statement that Glaurung is close to the level of Crafty is not true without certain qualifications. In particular, Crafty would obviously be far stronger on any computer with multiple CPUs. When playing at fast time controls on single-CPU 32-bit computers, however, the two programs do seem to be of comparable strength. I think this is very strange. Crafty calculates about three times as many nodes per second than Glaurung, and knows much more about virtually everything. That Crafty is not a couple of hundred Elo points stronger is one of the biggest mysteries of computer chess.

Tord

[Uri Blass]

Here is rating from AEGT

It seems that Glaurung performs well not only at blitz.

30 Glaurung 0.2.3 : 2534 36 37 229 41.0 % 2596
35 Crafty 19.15 : 2522 22 22 620 52.1 % 2507
38 Crafty 19.19 : 2518 46 46 139 43.2 % 2566


My guess is that part of the knowledge that Crafty has is simply counter productive and it is possible to remove part of Crafty's evaluation and do it stronger(at least against computers).

Uri