Big bug found in Kiwi 0.6b

[Guenther Simon]

Well...this happened a few minutes ago live in RWBC.

New Kiwi was completely winning against the strong new Matacz with
Black, but did not see a stalemate, or did not consider stalemate
in the search at all, despite the PV already ends with the stalemate
in ply 3?
(I am not 100% sure though if it could win without tablebases?)

Guenther
P.S. Kiwi played with its bitbases of course

[Site "RWBC Open Class 6th Edition, Running since October 20th, Espresso PIV 2.67Ghz "]
[Date "10-22-2005"]
[White "Matacz_10 Book "]
[Black "Kiwi 0.6b "]
[Result "1/2-1/2"]
[TimeControl "40150"]

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 e6 6. Be2 a6 7. O-O
Be7 8. Qd3 b5 9. Bd2 Bb7 10. b4 Nc6 11. Nxc6 Bxc6 12. Bf4 O-O 13. Rad1
Rc8 14. Qe3 Qc7 15. Rd3 Nd7 16. Rb1 Ne5 17. Rd2 Kh8 18. a4 Be8 19. Bxe5
dxe5 20. Nd1 f5 21. axb5 axb5 22. Bd3 h6 23. Qe2 Rd8 24. c3 Bc6 25. f3
Qb6 26. Qf2 Qb7 27. exf5 exf5 28. Bc2 e4 29. fxe4 fxe4 30. Rxd8 Rxd8
31. Ra1 Qd7 32. Ne3 Qd2 33. Qxd2 Rxd2 34. Ra6 Bg5 35. Nf1 Rxc2 36. Rxc6
e3 37. Re6 Rc1 38. h4 Bf4 39. g3 Bxg3 40. Rxe3 Bxh4 41. Re5 Rxc3
42. Rxb5 Rc2 43. Rd5 Be7 44. b5 Bc5 45. Kh1 Kh7 46. Nd2 Be3 47. Nf1
Bf4 48. Rd4 g5 49. Rb4 h5 50. b6 h4 51. b7 Bb8 52. Ne3 Rh2 53. Kg1
h3 54. Rb5 Re2 55. Nf1 Kg6 56. Rb6 Kf7 57. Rb1 g4 58. Kh1 Kg6 59. Rb4
Kf5 60. Rb1 Kf4 61. Rb4 Re4 62. Rb1 g3 63. Nxg3 Kxg3 64. Rg1 Kh4
65. Rb1 Ra4 66. Kg1 Rg4 67. Kh1 h2 68. Rc1 Rb4 69. Kg2 Rb2 70. Kh1
Kg4 71. Re1 Bf4 72. b8=Q Rxb8 73. Kg2 Bg3 74. Rf1 Re8 75. Kh1 Kh3
76. Rd1 [D]Re1 77. Rxe1 Bxe1
1/2-1/2 {stalemate}

...Re1??
[diag]4r3/8/8/8/8/6bk/7p/3R3K b - - 0 76[/diag]
4r3/8/8/8/8/6bk/7p/3R3K b - - 0 76

Code: Select all

3393720 <first : move f1d1
3393720 >second: time 11733
otim 9466
3393720 >second: f1d1
3393730 <second: 3 624 0 19 Re2 Rb1 Re1+ Rxe1 Bxe1
3393730 <second: 4 624 0 223 Re1+ Rxe1 Bxe1
3393730 <second: 5 624 0 296 Re1+ Rxe1 Bxe1
3393730 <second: 6 624 0 614 Re1+ Rxe1 Bxe1
3393730 <second: 7 624 0 1008 Re1+ Rxe1 Bxe1
3393730 <second: 8 624 0 1481 Re1+ Rxe1 Bxe1
3393730 <second: 9 624 0 2551 Re1+ Rxe1 Bxe1
3393740 <second: 10 624 0 3889 Re1+ Rxe1 Bxe1
3393750 <second: 11 624 1 12212 Re1+ Rxe1 Bxe1
3393790 <second: 12 624 6 44334 Re1+ Rxe1 Bxe1
3393960 <second: 13 624 23 170282 Re1+ Rxe1 Bxe1
3394601 <second: 14 624 87 593591 Re1+ Rxe1 Bxe1
3396524 <second: 15 624 278 1969474 Re1+ Rxe1 Bxe1
3403203 <second: 15 624 946 6525643 Re1+ Rxe1 Bf2 Rd1 Kg3 Rc1 Kh3 Rb1 Bg3 Rb5
3403213 <second: 16 624 947 6535678 Re1+ Rxe1 Bxe1
3417203 <second: 16 624 947 6535678 Re1+ Rxe1 Bxe1
3417203 <second: move e8e1

[Alessandro Scotti]

Hi Guenther,
thanks for the position! I doubt Kiwi could win this without tablebases but I'll try to see why it doesn't try to avoid the stalemate.

P.S. BTW the half-point against Matacz is not bad at all!

[Alessandro Scotti]

Hi Guenther,
well it was not difficult to reproduce... it's a bug with bitbases!
I'll now try to look better into the problem...

[Guenther Simon]

Alessandro said:

Hi Guenther,
well it was not difficult to reproduce... it's a bug with bitbases! Sad
I'll now try to look better into the problem...

Oh, sorry to hear this. Shall I disable the bitbases or parts of them
in the next games?

Best regards,
Guenther

[Alessandro Scotti]

Oh, sorry to hear this. Shall I disable the bitbases or parts of them in the next games?

Hi Guenther,
I have found the bug in the probing code. It only happens with KBPK/black, so it shouldn't be too dangerous. I would say to keep using bitbases for now (...and maybe discover some other problem! ) until next Kiwi...
Thanks again for spotting this bug!

[Pradu]

[diag]4r3/8/8/8/8/6bk/7p/3R3K b - - 0 76[/diag]
4r3/8/8/8/8/6bk/7p/3R3K b - - 0 76

...Re1??

How would black start to win this position? I think it is drawn already if I'm not mistaken. Kiwi's egtb bug may not be critical. Returning a non-draw score is indeed strange nevertheless.

[Uri Blass]

Black simply sacrifice the pawn and wins KRB vs KR position

White has no choice but to capture the pawn because otherwise the pawn promote.

1...Bf4 2.Rf1 Kg4 3.Rb1 Rc8 4.Rf1 Be3 is a possible line and now best for white is 5.Kxh2 that is losing based on tablebases

alternative is 5.Re1 Kf3 6.Rf1+ Kg3 7.Re1 Kf2 8.Ra1 Rb8 9.Ra2+ Kf3 10.Rb2(Ra1 Rg8 Ra4 Rg1+ is even worse for white) Rc8 11.Rb1(Rc2 Rd8 Rc3 Rd1+ Kxh2 Rd5 is also winning) Rg8 and white cannot prevent Rg1+ with winning tablebases position.

[Pradu]

Ah, didn't know KRB vs KR is won.

[Uri Blass]

KRB vs KR is not generally won and the better side can win only in part of the cases but in this case black can sacrifice the pawn in a condition that force white to lose the game.

Uri

[Alessandro Scotti]

Returning a non-draw score is indeed strange nevertheless.

Hi Pradu,
in Kiwi I compute bitbases for white and then mirror the position for black. In this case I forgot to properly mirror so if you just invert colors in the above diagram the black king would not be stalemated by the white pawn!