bad eval for passers ?

[Daniel Anulliero]

hi all
Here is a game of my engine vs Beowulf :
[pgn]
[Event "TEST ISA 1_2_63"]
[Site "DANY-PC"]
[Date "2015.02.23"]
[Round "38"]
[White "Beowulf-24a-64-ja"]
[Black "ISA_1_2_63"]
[Result "1-0"]
[BlackElo "1899"]
[ECO "D46"]
[Opening "Semi-Slav"]
[Time "00:21:43"]
[Variation "Chigorin, 7.Qc2 Main Line, 9...b5"]
[WhiteElo "2242"]
[TimeControl "0+3"]
[Termination "normal"]
[PlyCount "169"]
[WhiteType "program"]
[BlackType "program"]

1. d4 d5 2. c4 c6 3. Nf3 Nf6 4. Nc3 e6 5. e3 Nbd7 6. Qc2 Bd6 7. Bd3 dxc4 8.
Bxc4 O-O 9. O-O {(O-O e5 Bd3 exd4 exd4 b6 Be3 Bb7) +0.36/8 6} b5 {(b7b5
c4b3 c8a6 e3e4 b5b4 c3e2 a6b5 c1g5) -0.15/8 3} 10. Bd3 {(Bd3 Bb7 Bd2 e5 Ne4
Qe7 Nxd6 Qxd6 b3) +0.54/9 7} Bb7 {(c8b7 c1d2 a8c8 a2a3 e6e5 a1d1 a7a5)
+0.01/7 3} 11. Bd2 {(Bd2 Qc7 Rac1 Rac8 Ne4 Nxe4 Bxe4 Nf6) +0.56/8 7} Rc8
{(a8c8 a2a3 d7b6 e3e4 b6c4 d2g5 d6e7) -0.05/7 3} 12. Ne4 {(Ne4 Nxe4 Bxe4 f5
Bd3 Qf6 Qb3 h6) +0.39/8 9} Nxe4 {(f6e4 d3e4 d7f6 a1c1 f6e4 c2e4 d6e7 f1d1
d8d5 e4b1) +0.49/10 3} 13. Bxe4 {(Bxe4 Nf6 Bd3 c5 Bxb5 Bxf3 gxf3 cxd4)
+0.20/8 2} Nf6 {(d7f6 f1d1 f6e4 c2e4 d8c7 f3e5 c6c5 e4f4 f8d8) +0.48/9 3}
14. Rac1 {(Rac1 Nxe4 Qxe4 Be7 Qd3 Qd5 b3 f6) +0.04/8 8} Nxe4 {(f6e4 c2e4
d6e7 f3e5 d8d5 e4d5 e6d5 f1e1) +0.35/8 3} 15. Qxe4 {(Qxe4 Qe7 Qd3 a6 Ba5 f6
b3 c5 Qe2 Bxf3 Qxf3 cxd4 exd4 Rxc1 Rxc1) -0.09/9 2} Re8 {(f8e8 e4b1 c6c5
d4c5 c8c5 c1c5 d6c5 e3e4) +0.43/8 3} 16. b3 {(b3 f5 Qd3 c5 Rc2 Bxf3 gxf3
cxd4 Rxc8 Qxc8 Qxb5) -0.06/8 11} Ba3 {(d6a3 c1d1 a3d6 d1c1) +0.47/8 3} 17.
Rc2 {(Rc2 Qe7 Qf4 c5 dxc5 Rxc5 Rxc5 Qxc5 Rd1) -0.22/9 10} Bd6 {(a3d6 f3e5
d8c7 f1d1 e8d8 a2a4 c7b8 c2c3) +0.37/8 3} 18. h3 {(h3 f6 Qd3 c5 Qxb5 Bxf3
gxf3 cxd4) +0.11/8 9} Re7 {(e8e7 f3g5 g7g6 f1d1 a7a5 e4h4 f7f6) +0.60/7 3}
19. Ng5 {(Ng5 f5 Qd3 h6 Nf3 c5 Qxb5 Bxf3 gxf3 cxd4 Rxc8 Qxc8 exd4) +0.29/9
8} f5 {(f7f5 e4d3 h7h6 g5f3 g8h7 e3e4 f5e4 d3e4) +0.35/8 3} 20. Qd3 {(Qd3
h6 Nf3 a6 g3 c5 dxc5 Bxc5 Qxd8+ Rxd8 Rxc5 Bxf3) +0.12/10 9} h6 {(h7h6 g5f3
g8h7 e3e4 c6c5 e4f5 b7f3 g2f3) +0.22/8 3} 21. Nf3 {(Nf3 b4 Rfc1 c5 Qc4 Bxf3
gxf3 Rcc7 Qd3 cxd4) -0.12/10 10} b4 {(b5b4 f1d1 c6c5 d3e2 c5d4 c2c8 d8c8
f3d4) +0.45/8 3} 22. Rfc1 {(Rfc1 c5 Qc4 Bd5 Qb5 Bxf3 gxf3 Rb8 Qc4 cxd4
Qxd4) -0.17/9 8} c5 {(c6c5 d3b5 b7f3 g2f3 c8b8 b5c4 c5d4 c4d4) +0.39/8 3}
23. Qb5 {(Qb5 Bxf3 gxf3 Rb8 Qc4 cxd4 Qxd4 e5 Qd5+ Kh8) -0.19/9 10} Bxf3
{(b7f3 g2f3 c8b8 b5c4 c5d4 c4d4 d8b6 c2c4 b6d4 c4d4) +0.55/9 3} 24. gxf3
{(gxf3 Rb8 Qa6 Qd7 Qc4 cxd4 Qxd4 Qd8 Kh1 a5) -0.16/10 10} Rb8 {(c8b8 b5a6
d8b6 a6b6 a7b6 g1g2 g8h7 f3f4 b8d8 g2f3) +0.38/10 3} 25. Qa6 {(Qa6 Kh7 dxc5
Bxc5 Rxc5 Qxd2 R1c2 Qd1+ Kg2 Rb6 Qf1) -0.06/10 10} Qb6 {(d8b6 a6b6 a7b6
g1g2 g8h7 f3f4 e7d7 g2f3 d6e7 c2c4 b8d8) +0.54/11 3} 26. Qxb6 {(Qxb6 axb6
Rd1 Rd7 Bc1 Rbd8 Rcd2 Bc7 Bb2 Rc8 dxc5 Rxd2 Rxd2 bxc5) -0.09/11 10} axb6
{(a7b6 g1g2 g8f7 c1d1 e7d7 f3f4 b8d8 g2f3 d6e7 c2c4 f7f6) +0.58/11 3} 27.
Rd1 {(Rd1 Rd7 Bc1 Be7 Rcd2 Kf7 Kg2 Ra8 dxc5 Rxd2 Bxd2 bxc5) -0.22/10 8} Rd8
{(b8d8 g1f1 e7a7 f1e2 d6e7 d4c5 b6c5 e3e4 a7d7 e4e5) +0.67/10 3} 28. Bc1
{(Bc1 Red7 Rcd2 g5 dxc5 bxc5 e4 fxe4 fxe4 e5 Bb2 Kh7) -0.28/11 10} Kf7
{(g8f7 g1g2 e7a7 d1d3 a7a6 c1d2 d8a8 d4c5 b6c5) +0.76/9 3} 29. dxc5 {(dxc5
Bh2+ Kxh2 Rxd1 c6 Rc7 Bd2 Rxc6 Rxc6 Rxd2 Rxb6 Rxa2 Rxb4 Rxf2+ Kg3) -0.13/11
10} bxc5 {(b6c5 g1g2 e7d7 a2a3 d6e7 d1d7 d8d7 a3b4 c5b4 c1b2 d7d1) +0.83/11
3} 30. Kg2 {(Kg2 Red7 a4 Be7 Rxd7 Rxd7 a5 Rd6 Rd2 Ra6 Ra2 g5) -0.16/12 10}
Red7 {(e7d7 a2a3 d6e7 d1d7 d8d7 c1b2 b4a3 b2a3 d7d5 a3b2) +0.82/10 3} 31.
a4 {(a4 Be7 Rxd7 Rxd7 Rd2 Rc7 Rc2 Rc6 Bb2 g5 Bd4) -0.04/11 8} Be7 {(d6e7
d1d7 d8d7 e3e4 d7d3 c1e3 f5f4 e3c5 e7c5 c2c5 d3b3) +0.88/11 3} 32. Rxd7
{(Rxd7 Rxd7 Rd2 Rc7 Rc2 g5 Kf1 Bd6 a5 Ra7 Ra2 Ra6 Bd2 f4) -0.11/13 10} Rxd7
{(d8d7 e3e4 f7f6 a4a5 e6e5 c1e3 d7c7 e4f5 f6f5 a5a6 f5e6) +0.76/11 3} 33.
Rd2 {(Rd2 Rc7 Rc2 g5 e4 Kg6 Be3 f4 Bd2 e5 Rc1 Rc6) -0.12/13 10} Rxd2 {(d7d2
c1d2 c5c4 b3c4 b4b3 d2c3 e7a3 g2g3 g7g5 f3f4 b3b2 c3b2 a3b2 f4g5 h6g5)
+1.85/12 3} 34. Bxd2 {(Bxd2 Ke8 Kf1 Kd7 Ke2 Kc6 Bc1 g5 Bb2 Kd5 Kd3 Bd6 Kc2)
+0.11/13 3} c4 {(c5c4 b3c4 b4b3 d2c1 e7f6 g2g3 g7g6 c4c5 b3b2 c1b2 f6b2)
+2.06/11 3} 35. bxc4 {(bxc4 b3 Bc1 Ke8 a5 Kd7 e4 Bf6 a6 Bd4 Bb2 e5 Bxd4
exd4) +0.20/14 6} b3 {(b4b3 d2c1 e7f6 f3f4 b3b2 c1b2 f6b2 g2f3 g7g6 c4c5
f7f6) +2.03/11 3} 36. Bc1 {(Bc1 Ke8 Kf1 Kd7 Ke2 Bf6 Kd3 b2 Bxb2 Bxb2 a5 Kc6
a6 Kb6 Kc2 Be5) 0.00/15 10} Bf6 {(e7f6 c4c5 b3b2 c1b2 f6b2 c5c6 b2e5 f3f4
e5d6 g2f3 f7f6) +1.92/11 3} 37. a5 {(a5 Be7 Kf1 Ke8 Ke2 Kd7 Bb2 g5 Kd3 Kc6
Kc3 Bd8 a6 Kb6) +1.98/14 8} Be5 {(f6e5 f3f4 b3b2 c1b2 e5b2 a5a6 b2d4 e3d4
f7e7 g2f3 e7d6) -0.52/11 3} 38. f4 {(f4 Bb8 Kf1 Ke7 Ke2 Kd6 Bb2 g6 Kd3 Kc6
Kc3 Bc7 a6) +1.90/13 6} Bd6 {(e5d6 a5a6 d6b8 c1b2 g7g6 g2f3 f7e7 f3e2 g6g5
f4g5 h6g5) -0.94/11 3} 39. Bb2 {(Bb2 g6 Kf3 Ke7 Ke2 Kd7 Kd3 Kc6 Kc3 Bc7 Kb4
Bxa5+ Kxa5 Kc5 Bg7 Kxc4 Bxh6) +1.78/14 10} g6 {(g7g6 a5a6 d6c5 g2f3 f7e7
f3e2 e7d7 e2d3 h6h5 f2f3 d7c8 e3e4) -1.12/12 3} 40. Kf3 {(Kf3 Ke7 Ke2 Kd7
Kd3 Kc6 Bd4 Kb7 Kc3 Bc7 Bg7 Bxa5+ Kxb3 Be1 Bxh6 Bxf2) +2.06/15 10} Ke8
{(f7e8 f3e2 e8d7) -0.85/11 3} 41. Ke2 {(Ke2 Kd7 Kd3 Bb4 a6 Kc7 Bd4 Kb8 a7+
Ka8 f3 Ba5 e4 Bc7 Kc3) +1.95/14 10} Kd7 {(e8d7 e2d3 d7c6 d3c3 e6e5 f4e5
d6e5) -0.79/11 3} 42. Kd3 {(Kd3 Bb4 a6 Kc7 Bd4 Kb8 c5 Ka7 Kc4 Be1 f3 b2
Bxb2 Kxa6) +1.95/13 10} Kc6 {(d7c6 d3c3 d6c7 c3b4 c7d8 f2f3 d8e7) -0.78/11
3} 43. Kc3 {(Kc3 Bc7 a6 Kb6 Kxb3 Kxa6 Bg7 h5 Be5 Ba5 Bd4 Be1 f3 Bf2)
+1.96/14 10} Bc7 {(d6c7 c3b4 c7d8 f2f3 d8e7) -0.79/11 3} 44. Kb4 {(Kb4 h5
Bc3 Bxa5+ Kxa5 Kd6 Bb2 Kc6 Be5 Kc5 Ka4 Kxc4 Bb2 Kd3 Kxb3 Ke2) 0.00/14 12}
Bd8 {(c7d8 f2f3 d8e7) -0.78/11 3} 45. Bg7 {(Bg7 Bh4 f3 Bf2 Kxb3 Bxe3 Bxh6
Bd2 Ka4 Kc5 Bg5 Kxc4 a6) +2.35/13 9} Bh4 {(d8h4 f2f3 h4f2 e3e4 f2e3 b4b3
e3f4) -0.07/11 3} 46. f3 {(f3 Bf2 e4 Be3 Kxb3 Bxf4 Bd4 Kb7 Bb6 Kc6 Kc3 fxe4
fxe4 e5) +2.13/14 9} Bf2 {(h4f2 e3e4 f2e3 b4b3 e3f4 g7b2 h6h5 b3a4 h5h4
a5a6 c6b6 e4f5 e6f5) -0.15/12 3} 47. e4 {(e4 Be3 Kxb3 Bxf4 Kc3 h5 Kd4 Kb7
Kd3 Ka6 Bc3 Bc7 h4) +2.20/14 10} Be3 {(f2e3 b4b3 e3f4 g7d4 e6e5 d4f2 f4d2
e4f5 g6f5 b3a4 f5f4) -0.17/11 3} 48. Kxb3 {(Kxb3 Bxf4 Bd4 Kb7 Bb6 Kc6 Kc3
g5 Kd3 h5 Bd4 fxe4+) +2.23/13 10} Bxf4 {(e3f4 g7d4 e6e5 d4f2 f5e4 f3e4 h6h5
b3a4 c6d6 a5a6 f4g5) -0.33/11 3} 49. Kb4 {(Kb4 Bd2+ Bc3 Be3 Bf6 Bd2+ Ka4
Be3 Kb3 Bf2 Bg7 h5 Bf6 fxe4 fxe4 Ba7 h4 Be3) +2.18/13 10} e5 {(e6e5 c4c5
f5e4 f3e4 h6h5 b4c4 h5h4 a5a6 g6g5 a6a7) -0.03/10 3} 50. c5 {(c5 fxe4 fxe4
g5 Bxh6 g4 hxg4 Bxh6 Kc4 Bg5 a6 Be7 Kc3) +2.23/14 11} fxe4 {(f5e4 f3e4 f4d2
b4c4 d2a5 g7h6 a5d8 h6d2 d8e7 d2e3 g6g5 e3f2 e7f6) -0.85/12 3} 51. fxe4
{(fxe4 g5 Kc4 h5 a6 g4 hxg4 hxg4 Kd3 g3 Ke2 Kc7 a7 g2 Kf2 Kb7) +2.26/15 11}
Bd2+ {(f4d2 b4c4 d2a5 g7h6 a5e1 h6e3 e1h4 e3c1 h4f2 c1a3 f2e3 a3b4 e3g5)
-0.76/12 3} 52. Ka4 {(Ka4 Be3 Bxe5 Bxc5 Bg7 Be3 e5 Kd5 a6 Bc5 Kb5 Ba7 Ka4
Bc5 &rep.) +2.80/16 10} Be3 {(d2e3 g7f8 e3c5 f8h6 c5f2 a4b4 f2e1 b4a4 c6c5
h6e3 c5c6 a5a6 e1c3 h3h4) -0.72/12 3} 53. Bxe5 {(Bxe5 Bxc5 Bg7 Be3 Kb4 h5
Kc4 Bg5 Bd4 Be7 a6 h4 a7 Kb7 Kd3 g5 Ke3) +2.75/16 10} Bxc5 {(e3c5 e5b8 c6b7
a4b5 c5f2 b8e5 h6h5 a5a6 b7a7 e5c3 h5h4 b5a5 g6g5) -0.99/12 3} 54. Bg7
{(Bg7 Be3 e5 Bd4 Kb4 h5 Kc4 Bc5 Bf6 Bf2 e6 Be1 a6 Bh4) +2.88/15 8} h5
{(h6h5 g7e5 c5e7 e5d4 e7c5 d4e5) -0.26/12 3} 55. Bf6 {(Bf6 Ba7 h4 Bf2 e5
Bd4 Kb4 Kd5 Kb5 Be3 a6 Bf2 Kb4 g5 hxg5 Bd4 e6 Bc5+ Kc3 Kxe6 Be5 Ke7)
+2.94/16 10} Be3 {(c5e3 a4b4 c6d7 b4c4 d7e6 f6d4 e3d2 a5a6 d2f4 d4c3 h5h4)
-0.23/11 3} 56. h4 {(h4 Bd2 a6 Kb6 e5 Bh6 a7 Kxa7 Be7 Ka8 e6 Bg7 Bg5 Bf8
Bh6 Bxh6 e7 Bg7 e8=r+ Kb7 Re7+ Kc6 Rxg7) +3.02/16 10} Bf2 {(e3f2 a4b4 f2e1
b4a4 e1f2) -0.13/12 3} 57. e5 {(e5 Bd4 Kb4 Bb2 Kc4 Kc7 Kc5 Kb7 Bg7 Bc3 Kb5
Be1 a6+ Kb8 Bf6 Bb4 a7+ Ka8) +3.35/16 9} Kd5 {(c6d5 a4b5 f2e3 a5a6 e3a7
f6d8 a7d4 d8b6 d4b6 b5b6 d5e5 b6c6) -0.34/12 3} 58. Kb5 {(Kb5 Bd4 a6 Ba7
Be7 g5 hxg5 Bg1 Bf6 Ke6 g6 Bf2 Bh4 Be3 g7 Bg1 a7 Bxa7) +4.17/16 10} Be3
{(f2e3 f6g5 e3f2 a5a6 f2d4 e5e6 d5e6 b5c6 e6f5 c6b7 d4f6 g5f6 f5f6)
-0.69/12 3} 59. a6 {(a6 Ba7 Be7 g5 hxg5 h4 g6 h3 g7 h2 g8=q+ Kxe5 Bf6+ Kf5
Bd4 h1=q Bxa7 Qb1+ Ka5 Qe4) 0.00/15 10} Bf2 {(e3f2 f6d8 f2a7 d8b6 a7b8 a6a7
b8a7 b6a7 d5e5 b5c4 e5f4 c4d4 g6g5) -2.03/13 3} 60. e6 {(e6 Kxe6 Bd8 Ba7
Kc6 g5 Bb6 Bb8 Kb7 Bg3 a7 Bxh4 a8=q Kf5 Qf8+ Ke5) +7.09/14 7} Kxe6 {(d5e6
f6d8 f2a7 d8b6 a7b8 b5c6 g6g5 c6b7 b8a7 b6a7 g5h4 a7e3 e6d5 b7a8) -3.86/14
3} 61. Bd8 {(Bd8 Ba7 Kc6 Kf5 Bb6 Bb8 Kb7 Bg3 a7 Bxh4 a8=q Bf6 Qc8+ Ke5 Bc7+
Kd4 Qd7+ Kc4 Qe6+ Kb4) +7.23/15 10} Ba7 {(f2a7 d8b6 a7b8 b6c7 b8a7 c7b6)
-3.86/14 3} 62. Bb6 {(Bb6 Bxb6 Kxb6 g5 a7 Kf7 a8=q g4 Qc6 g3 Ka7 Kf8 Kb8)
+8.76/15 10} Bb8 {(a7b8 b5c6 e6f5 c6b7 b8g3 a6a7 g3h4 a7a8 h4f6 a8e8 h5h4
b6d8 f6g5 d8g5 f5g5) -5.34/14 3} 63. Kc6 {(Kc6 Kf7 Bf2 g5 Kb7 Be5 a7 Bf6
hxg5 Bxg5 a8=q Bf6 Kc7 h4 Qh1 Be5+ Kd7) +9.24/15 10} Kf5 {(e6f5 c6b7 b8g3
a6a7 g3h4 a7a8 h4f6 a8e8 h5h4 b6d8 f6d8 e8d8 f5g4) -5.40/13 3} 64. Kb7
{(Kb7 Bf4 a7 Kg4 a8=q Kxh4 Qc8 g5 Bf2+ Bg3 Bxg3+ Kxg3 Qb8+ Kf2 Qh2+ Kf3
Qxh5+ Kf4 Qf7+ Ke5 Qe8+ Kf4 Qh5 g4 Qf7+ Kg5 Qd5+ Kg6 Qd6+ Kg5 Qd5+ &rep.)
+9.41/14 10} Bg3 {(b8g3 a6a7 g3h4 a7a8 h4f6 a8c8) -5.68/12 3} 65. a7 {(a7
Bxh4 a8=q g5 Qc8+ Ke5 Bc7+ Kf6 Qe8 Bf2 Bd8+ Kf5 Qf7+ Ke4 Qxh5 Bg3 Bxg5 Kd4
Bh4 Be5 Bg5 Bf4 Bxf4 Kc3) +10.60/12 7} Bxh4 {(g3h4 a7a8 h4g3 b6d8 h5h4 a8c8
f5f4 c8e6 g6g5 e6f6 f4e4 f6g6 e4d5 d8g5) -6.52/11 3} 66. a8=Q {(a8=q g5
Qf8+ Kg4 Qc8+ Kf4 Qe6 Kg3 Qe2 Kh3 Bc7 Bf2 Qf3+ Kh4 Qxf2+ Kg4 Qg3+ Kf5 Kc6
h4 Qe5+ Kg4 Qd4+ Kf3) +10.83/11 9} Bg3 {(h4g3 a8c8 f5f4 c8e6 h5h4 e6g6 f4f3
b7c6 f3f4 b6c7 f4f3 g6f5) -7.19/10 3} 67. Qc8+ {(Qc8+ Kg5 Qd8+ Kg4 Qd7+ Kg5
Bd8+ Kf4 Bc7+ Kg5 Bxg3 Kf6 Bh4+ Ke5 Qc7+ Kd4 Qd8+ Ke4 Qe7+ Kd5 Qd7+ Ke5
&rep.) +12.77/10 10} Kf4 {(f5f4 c8e6 h5h4 e6g6 f4f3 g6f5 f3g2 b7c6 h4h3
c6d5) -7.33/8 3} 68. Qc4+ {(Qc4+ Ke5 Bc7+ Kf6 Bd8+ Ke5 Qc7+ Ke6 Qxg3 Kd7
Qxg6 Kxd8 Qxh5 Kd7 Qh4 Ke6 Qd4) 0.00/11 10} Kf3 {(f4f3 c4c6 f3f4 b6c7 f4g5
c7g3 g5f5 c6f3 f5e6 f3e4 e6f6 e4e5 f6f7 b7c6) -9.05/8 3} 69. Qc6+ {(Qc6+
Kg4 Qxg6+ Kf3 Qxh5+ Kg2 Qg4 Kh2 Bc7 Bxc7 Kxc7 Kh1 Kd6) +M411/10 6} Kf4
{(f3f4 c6f6 f4g4 f6g6 g4h4 g6f5 g3e1 f5f6 h4g3 f6e5 g3g4 e5e1 h5h4)
-10.58/10 3} 70. Qf6+ {(Qf6+ Ke4 Qxg6+ Ke5 Qg5+ Kd6 Qxg3+ Kd5 Qf3+ Ke6 Qxh5
Kd7 Qh4 Ke6 Qg5 Kd6 Qh4 Ke6 &rep.) 0.00/11 10} Kg4 {(f4g4 f6g6 g4h4 b6d8
h4h3 g6h5 h3g2 d8b6 g3d6 h5d5 g2h3 d5d6 h3g4 d6d4 g4f5 b7c6) -12.64/10 3}
71. Qxg6+ {(Qxg6+ Kf4 Bc7+ Kf3 Bxg3 h4 Bxh4 Ke2 Qe4+ Kd2 Be1+ Kc1 Kc6)
0.00/10 10} Kf3 {(g4f3 g6h5 f3g2 h5g4 g2h2 b6c5 h2g2 c5d6 g2f2 g4g3 f2e2
b7c6 e2d2) -12.72/10 3} 72. Qxh5+ {(Qxh5+ Kg2 Be3 Kf1 Qf3+ Ke1 Qxg3+ Ke2
Qf2+ Kd3 Qf4 Kc2 Qc4+ Kb2 Kc6) 0.00/10 10} Kg2 {(f3g2 h5d5 g2h3 d5h1 h3g4
h1g2 g4h4 b6d8 h4g4 d8c7 g4f5 c7g3 f5e6 g2c6) -12.39/10 3} 73. Bc5 {(Bc5
Bf2 Qe2 Kh3 Qxf2 Kg4 Be7 Kh5 Qf5+ Kh6 Bf8#) 0.00/10 4} Bf4 {(g3f4 h5g4)
-13.32/12 3} 74. Qg4+ {(Qg4+ Bg3 Bd6 Kf1 Bxg3 Kg1 Bc7+ Kf1 Qf3+ Kg1 Bb6+
Kh2 Qe2+ Kg3 Bc7+ Kh3 Qf3+ Kh4 Qf4+ Kh5 Bd8 Kg6 Qf6+ Kh7 Kc6 Kg8) 0.00/3 0}
Bg3 {(f4g3 c5d6 g2f2 g4g3 f2e2 b7c6 e2d2 d6f4 d2e2 g3e3 e2f1 e3d3 f1g2 d3e3
g2f1) -13.98/12 3} 75. Bd6 {(Bd6 Kf1 Bxg3 Kg1 Bc7+ Kf1 Qf3+ Kg1 Bb6+ Kh2
Qe2+ Kg3 Bc7+ Kh3 Qf3+ Kh4 Qf4+ Kh5 Bd8 Kg6 Qf6+ Kh7 Kc6 Kg8) 0.00/3 0} Kf2
{(g2f2 g4g3) -13.82/12 3} 76. Qxg3+ {(Qxg3+ Kf1 Qe3 Kg2 Qe2+ Kh1 Qf3+ Kg1
Qg4+ Kf2 Bc5+ Ke1 Be3 Kf1 Qf3+ Ke1 Qf2+ Kd1 Qd2#) 0.00/3 0} Ke2 {(f2e2 b7c6
e2d2 g3f3 d2c2 d6b4 c2b2 f3c3 b2b1 c6d5 b1a2 c3a3) -14.19/10 3} 77. Bb4
{(Bb4 Kd1 Qd3+ Kc1 Ba3#) 0.00/3 0} Kf1 {(e2f1 g3f3 f1g1 b4c5 g1h2 c5d6 h2g1
f3e2 g1h1 e2f1) -M5/11 1} 78. Qf3+ {(Qf3+ Kg1 Ka8 Kh2 Bd6+ Kg1 Bc5+ Kh2 Qg4
Kh1 Qh3#) +M10/2 0} Kg1 {(f1g1 f3e2 g1h1 e2f1 h1h2 b4d6) -M3/8 0} 79. Be7
{(Be7 Kh2 Bd6+ Kg1 Qg4+ Kf2 Bc5+ Ke1 Be3 Kf1 Qf3+ Ke1 Qf2+ Kd1 Qd2#) 0.00/3
0} Kh2 {(g1h2 f3g4 h2h1 g4h3 h1g1 e7c5) -M3/8 0} 80. Bd6+ {(Bd6+ Kg1 Qg4+
Kf2 Bc5+ Ke1 Be3 Kf1 Qf3+ Ke1 Qf2+ Kd1 Qd2#) 0.00/3 0} Kg1 {(h2g1 f3e2 g1h1
e2f1) -M2/8 0} 81. Qg3+ {(Qg3+ Kh1 Qe3 Kg2 Qe2+ Kh1 Qf3+ Kg1 &rep.) 0.00/3
0} Kf1 {(g1f1 g3e3 f1g2 e3e2 g2h3 e2f3 h3h4 d6e7) -M4/8 0} 82. Qe3 {(Qe3
Kg2 Qe2+ Kh1 Qf3+ Kg1 Qg3+ Kh1 Qe3 Kg2 &rep.) 0.00/3 0} Kg2 {(f1g2 e3e2
g2g1 d6c5 g1h1 e2f1 h1h2 c5d6) -M4/9 0} 83. Qe2+ {(Qe2+ Kh1 Qf3+ Kg1 Qg3+
Kh1 Qe3 Kg2 &rep.) +M10/2 0} Kh3 {(g2h3 e2f3 h3h4 d6e7) -M2/4 0} 84. Qf3+
{(Qf3+ Kh4 Be7#) +M7/2 0} Kh4 {(h3h4 d6e7) -M1/2 0} 85. Be7# {(Be7#) +M1/2
0} 1-0
[/pgn]




Isa played not too bad in openning and middlegame but in late endgame it seems it has some problems to evaluate correctly the passers.
In my eval I have for the passers :

Code: Select all

1)bonus "Row dependant"
 int bonus_avance_pp[8] = {0,60,40,25,15,10,10,0}; for white passer 
reverse for black
2) +20 bonus if the pawn is !isolated

For the late endgame (only a Rook or a minor of each color):
3) bonus for his  own king near the promo square vs ennemy king near the promo square :int bonus_roi_sqpromo[8] = {0,100,60,20,20,5,5,0};
bonus = (2 * own king near) - (ennemy king near)
For the pawns endgames : 
4) Classical square rule  (if unstoppable --> passer = VAL_QUEEN + VAL_PAWN)

It seems It's not sufficent to beat Beowulf !
All critics and ideas are welcome
bests
Dany

[Ferdy]

hi all
Here is a game of my engine vs Beowulf :

Isa played not too bad in openning and middlegame but in late endgame it seems it has some problems to evaluate correctly the passers.
In my eval I have for the passers :

1)bonus "Row dependant"
int bonus_avance_pp[8] = {0,60,40,25,15,10,10,0}; for white passer
reverse for black

Perhaps you may try scoring with interpolation given mg and eg phase values, mg for mid-game and eg for end-game.

2) +20 bonus if the pawn is !isolated

You can increase accuracy by saying if this passer is protected by own pawn or even the rank distance between these 2 pawns is less than or equal to 1.
But there is also a little trick about scoring for non-isolated passer. If the passer is not isolated it is effective against opp rook, but it is not effective against opp bishop.

For the late endgame (only a Rook or a minor of each color):
3) bonus for his own king near the promo square vs ennemy king near the promo square :int bonus_roi_sqpromo[8] = {0,100,60,20,20,5,5,0};
bonus = (2 * own king near) - (ennemy king near)

One refinement is to assure protection of the front sq of the passer, because a safe front sq guarantees its potential progress.
You can add of course a penalty for blocked passer, a passer whose front sq is occupied.

For the pawns endgames :
4) Classical square rule (if unstoppable --> passer = VAL_QUEEN + VAL_PAWN)

Be careful with this rule and scoring, say you have 2 unstoppable but not too far advanced and your opp has only 1 unstoppable but a little bit more advanced, now compare the score on both sides because you might have given too much bonus for the side with 2 pawns but actually the side with 1 pawn promotes first.

[Evert]

Isa played not too bad in openning and middlegame but in late endgame it seems it has some problems to evaluate correctly the passers.
In my eval I have for the passers :

Code: Select all

1)bonus "Row dependant"
 int bonus_avance_pp[8] = {0,60,40,25,15,10,10,0}; for white passer 
reverse for black
2) +20 bonus if the pawn is !isolated

For the late endgame (only a Rook or a minor of each color):
3) bonus for his  own king near the promo square vs ennemy king near the promo square :int bonus_roi_sqpromo[8] = {0,100,60,20,20,5,5,0};
bonus = (2 * own king near) - (ennemy king near)
For the pawns endgames : 
4) Classical square rule  (if unstoppable --> passer = VAL_QUEEN + VAL_PAWN)

It seems It's not sufficent to beat Beowulf !

• If you haven't already, make sure that you have some sort of mg/eg scaling for passers. You don't want to push them too aggressively in the middle game.
• Don't make the bonus too large to begin with! This will break things badly. Start slowly and if it works, tune the thing up.
• Don't give a bonus for having the own king closer to the pawn, give a penalty for the other king being nearer. I know this looks like the same thing (and it is, technically), but it's a bit easier to tune this way because you don't need to renormalise the base bonus (I may not have explained what I meant here properly, so ask if it's unclear).
• Rule of the square applies only to pawn end-games. Not sure if you have it that way.
• The king should always support passers in the end-game, not just in the "late" end game.
• "if unstoppable --> passer = VAL_QUEEN + VAL_PAWN" Really? Consider what this implies: the passer is worth more than a queen even before factoring in advancement and king placement? If you do this the engine is going to defer promotion as long as possible (right into the 50-move barrier) because it sees promotion as a drop in score. The value of a passer should always be less than a queen, after factoring in all bonuses and penalties. I have found it best to let it approach the value of a minor piece rather than a queen: this is accurate if it can be stopped (because the other side will likely give up a minor for it) and if it can't be stopped it doesn't matter anyway, as long as you promote the thing.

[Daniel Anulliero]

hi all
Here is a game of my engine vs Beowulf :

Isa played not too bad in openning and middlegame but in late endgame it seems it has some problems to evaluate correctly the passers.
In my eval I have for the passers :

1)bonus "Row dependant"
int bonus_avance_pp[8] = {0,60,40,25,15,10,10,0}; for white passer
reverse for black

Perhaps you may try scoring with interpolation given mg and eg phase values, mg for mid-game and eg for end-game.

I have the scores for mg and eg but use the same table bonus_avance_pp[] may be I can decrease these values for the op/middle ?

2) +20 bonus if the pawn is !isolated

You can increase accuracy by saying if this passer is protected by own pawn or even the rank distance between these 2 pawns is less than or equal to 1.
But there is also a little trick about scoring for non-isolated passer. If the passer is not isolated it is effective against opp rook, but it is not effective against opp bishop.

Oh yes , true ! I count the bonus for each passer non isolated , even if the second pawn is far from the passer .. not very good ... I'll change that

For the late endgame (only a Rook or a minor of each color):
3) bonus for his own king near the promo square vs ennemy king near the promo square :int bonus_roi_sqpromo[8] = {0,100,60,20,20,5,5,0};
bonus = (2 * own king near) - (ennemy king near)

One refinement is to assure protection of the front sq of the passer, because a safe front sq guarantees its potential progress.
You can add of course a penalty for blocked passer, a passer whose front sq is occupied.

Good idea !

For the pawns endgames :
4) Classical square rule (if unstoppable --> passer = VAL_QUEEN + VAL_PAWN)

Be careful with this rule and scoring, say you have 2 unstoppable but not too far advanced and your opp has only 1 unstoppable but a little bit more advanced, now compare the score on both sides because you might have given too much bonus for the side with 2 pawns but actually the side with 1 pawn promotes first.

Yes I will change my bonus in square rule , Evert pointed that too ...

bests
dany

[Daniel Anulliero]

Isa played not too bad in openning and middlegame but in late endgame it seems it has some problems to evaluate correctly the passers.
In my eval I have for the passers :

Code: Select all

1)bonus "Row dependant"
 int bonus_avance_pp[8] = {0,60,40,25,15,10,10,0}; for white passer 
reverse for black
2) +20 bonus if the pawn is !isolated

For the late endgame (only a Rook or a minor of each color):
3) bonus for his  own king near the promo square vs ennemy king near the promo square :int bonus_roi_sqpromo[8] = {0,100,60,20,20,5,5,0};
bonus = (2 * own king near) - (ennemy king near)
For the pawns endgames : 
4) Classical square rule  (if unstoppable --> passer = VAL_QUEEN + VAL_PAWN)

It seems It's not sufficent to beat Beowulf !

• If you haven't already, make sure that you have some sort of mg/eg scaling for passers. You don't want to push them too aggressively in the middle game.
• Don't make the bonus too large to begin with! This will break things badly. Start slowly and if it works, tune the thing up.

yes I have a mg/eg eval but use the same table ,I will decrease these values for the op/mg , as I said in my last post

[*]Don't give a bonus for having the own king closer to the pawn, give a penalty for the other king being nearer. I know this looks like the same thing (and it is, technically), but it's a bit easier to tune this way because you don't need to renormalise the base bonus (I may not have explained what I meant here properly, so ask if it's unclear).

Mmm ... Can you explain that more clearly please , bonus or penalty for me seems the same no?

[*]Rule of the square applies only to pawn end-games. Not sure if you have it that way.

Yes I have

[*]The king should always support passers in the end-game, not just in the "late" end game.

This heuristic is material dependant yes? If the oponent have Q + 2R ? It's dangerous for our King to go after his passer no?

[*]"if unstoppable --> passer = VAL_QUEEN + VAL_PAWN" Really? Consider what this implies: the passer is worth more than a queen even before factoring in advancement and king placement? If you do this the engine is going to defer promotion as long as possible (right into the 50-move barrier) because it sees promotion as a drop in score. The value of a passer should always be less than a queen, after factoring in all bonuses and penalties. I have found it best to let it approach the value of a minor piece rather than a queen: this is accurate if it can be stopped (because the other side will likely give up a minor for it) and if it can't be stopped it doesn't matter anyway, as long as you promote the thing.
[/list]

ouch! yes ... I'll change that silly thing ,thanks !
bests
dany

[Evert]

[*]Don't give a bonus for having the own king closer to the pawn, give a penalty for the other king being nearer. I know this looks like the same thing (and it is, technically), but it's a bit easier to tune this way because you don't need to renormalise the base bonus (I may not have explained what I meant here properly, so ask if it's unclear).

Mmm ... Can you explain that more clearly please , bonus or penalty for me seems the same no?

It's easier to keep the score within bounds. Consider two scenarios:

1. You assign a base-line for the value of a passer, then you apply bonuses.
2. You assign a base-line for the value of a passer, then you apply penalties.

Question: what is the largest possible value (say M, I'll use this later) that the code can give to a passer (which you need to know is sane, ie, << the value of a queen)?

In scenario 1 this is a bit tricky to keep track of, because you need to add up the various bonuses. In scenario 2 this is easy: it is the base value you chose. Of course the value needs to still be positive, but this is far less likely to be a problem in practice.

It's also a question of what people call "orthogonality" of evaluation terms. If you implement a feature as a bonus and want to tune the bonus value, you don't necessarily want to tune the maximum value M at the same time, which means you need to adjust the base value at the same time as the bonus, but that may affect some other terms as well.
The book-keeping is just slightly easier if the feature is implemented as a penalty rather than a bonus, even though mathematically the two descriptions are equivalent.

This heuristic is material dependant yes? If the oponent have Q + 2R ? It's dangerous for our King to go after his passer no?

Yes and no.
Don't make it material dependent, make it game-phase dependent. Otherwise you introduce nasty evaluation discontinuities.