Where was LCZero blunder?

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

Jouni
Posts: 3792
Joined: Wed Mar 08, 2006 8:15 pm
Full name: Jouni Uski

Where was LCZero blunder?

Post by Jouni »

[pgn][Event "TCEC Cup 16 Quarterfinal"]
[Site "https://tcec-chess.com"]
[Date "2026.01.13"]
[Round "1.15"]
[White "Reckless 0.9.0-dev-d6e7e21f"]
[Black "LCZero 0.33-dev-7499292-BT4-6"]
[Result "1-0"]
[ECO "C45"]
[WhiteElo "3684"]
[BlackElo "3722"]
[PlyCount "235"]
[EventDate "2026.??.??"]

1. e4 e5 2. Nf3 Nc6 3. d4 exd4 4. Nxd4 Bc5 5. Nxc6 Qf6 6. Qf3 Qxc6 7. Nc3 Ne7
8. Bg5 a6 9. O-O-O h6 10. Bxe7 Bxe7 11. Nd5 Bd8 12. h4 d6 13. g4 Be6 14. Kb1
Qa4 15. b3 Qa5 16. Rd3 Qe1+ 17. Kb2 Qa5 18. g5 Qc5 19. b4 Qb5 20. Kb1 Bxd5 21.
exd5 Qxb4+ 22. Rb3 Qe1+ 23. Kb2 Qe5+ 24. Ka3 O-O 25. Re3 Qd4 26. c3 Qc5+ 27.
Kb2 c6 28. gxh6 g6 29. Bd3 cxd5 30. Bxg6 d4 31. Qg4 Kh8 32. Bc2 Bf6 33. Qe4 d3
34. Qxd3 Qb5+ 35. Qxb5 axb5 36. a3 Be5 37. Rf1 Rfc8 38. f4 Bf6 39. Rd1 Rc6 40.
h7 Rac8 41. Rg1 Bg7 42. Rgg3 d5 43. h5 Re6 44. Rxe6 fxe6 45. Re3 Rc6 46. Kb3
Bh6 47. Rg3 Rc8 48. Rg6 Bxf4 49. Rxe6 Bd2 50. Rd6 Rxc3+ 51. Kb2 Rc8 52. Rxd5
Be3 53. Rxb5 Bc5 54. a4 b6 55. Ka2 Ra8 56. Rb1 Bd4 57. Rd1 Bf2 58. Kb3 Rc8 59.
Bg6 Bc5 60. Kc4 Bf2+ 61. Kb5 Bc5 62. Ka6 Rb8 63. Ka7 Rf8 64. Kb7 Be3 65. Rd6
Bf2 66. Re6 Bd4 67. Kc7 Bc5 68. Kc6 Bf2 69. Kd5 Bg1 70. Ke4 Bh2 71. Bf5 Ra8 72.
Kf3 Bg1 73. Kf4 Bc5 74. Kg5 Rd8 75. Rc6 Rf8 76. Be6 Be3+ 77. Kg6 Rd8 78. h6 Bf4
79. Rxb6 Bc7 80. Rb7 Rd6 81. Kf5 Rc6 82. Bd7 Rc5+ 83. Kg6 Rd5 84. Be8 Rd4 85.
Bb5 Ba5 86. Kg5 Bd2+ 87. Kf5 Bxh6 88. a5 Be3 89. Ke5 Rd2 90. Ke6 Bd4 91. Rf7
Bc5 92. Rf5 Rd6+ 93. Kf7 Ba7 94. Rf3 Rd2 95. Ke6 Rf2 96. Rh3 Rf8 97. Rd3 Bg1
98. a6 Bf2 99. Rd7 Re8+ 100. Kf7 Ra8 101. Kg6 Rc8 102. Bd3 Bc5 103. Bf5 Ra8
104. Be6 Re8 105. Bd5 Rb8 106. Rc7 Bf2 107. Be6 Re8 108. Kf7 Ra8 109. Bc8 Bd4
110. Rc4 Bg7 111. Kg6 Rb8 112. a7 Rb6+ 113. Kg5 Bh6+ 114. Kh4 Rg6 115. Bf5 Ra6
116. Rc8+ Bf8 117. Rxf8+ Kg7 118. h8=Q# 1-0

[/pgn]

I was following this live. Draw everybody said. But it wasn't :) . Was this OCB ending really white win?
Jouni
peter
Posts: 3526
Joined: Sat Feb 16, 2008 7:38 am
Full name: Peter Martan

Re: Where was LCZero blunder?

Post by peter »

Reckless' eval jumped online from 1.34 to 1.77 at move 78...Bf4, online- Stockfish went from 1.57 to 1.86 there.
So SF online was already near to +- one move earlier, at the move 77...Rd8 it went up from 1.19 before, rising to that already from 75...Rf8 (0.84).

My SF dev. (260101) installation sees White already winning much sooner after Backward. At 55:

Reckless 0.9.0-dev-d6e7e21f - LCZero 0.33-dev-7499292-BT4-6 1-0 1.15, TCEC Cup 16 Quarterfinal 2026
[d]2r4k/7P/1p6/1Rb4P/P7/8/K1B5/8 b - - 0 1

Analysis by Stockfish dev-20260101-28844fc6:

55...Tf8 56.Tb1 Lf2 57.Td1 Ta8 58.Kb3 Tf8 59.Td6 Lc5 60.Te6 Tc8 61.Lg6 Lg1 62.Kb4 Td8 63.Kb5 Tc8 64.Te2 Td8 65.Te1 Lf2 66.Te6 Ta8 67.Te2 Tf8 68.Ka6 Lc5 69.Kb7 Lf2 70.Kc6 Lg1 71.Te1 Lc5 72.Kb7 Lf2 73.Te6 Ld4 74.Te7 Lg1 75.Te4 Td8 76.Kc7 Ta8 77.Te2 Ld4 78.Kd7 Tf8 79.Te6 Kg7 80.Tc6 Le3 81.Tc2 Lf4 82.Tc4 Le3 83.Tc6
Tiefe: 65/115 00:01:10 2615MN, tb=3980945
+/- (1.54)

Lc0 played 55...Ra8 there, but doesn't make much eval- difference to home- SF. SF online was at 0.60 there.

After 45.Re3:

[d]2r4k/1p4bP/4p3/1p1p3P/5P2/P1P1R3/1KB5/8 b - - 0 1

Analysis by Stockfish dev-20260101-28844fc6:

45...Te8 46.Te1 d4 47.Ka2 dxc3 48.Tb1 Lh6 49.Tb4 b6 50.Txb5 Lxf4 51.Kb3 Le3 52.a4 e5 53.Tb4 Tc8 54.Tg4 Ld4 55.Tg6 Te8 56.Le4 Tc8 57.Kc2 Te8 58.Tc6 Td8 59.Kb3 Tb8 60.h6 Td8 61.Lc2 Te8 62.Kc4 Le3 63.Lg6 c2
Tiefe: 45/37 00:00:48 1477MN, tb=3259422
Die Stellung ist ausgeglichen.
= (0.09 ++)

And one move (45...Rc6) later:
[d]7k/1p4bP/2r1p3/1p1p3P/5P2/P1P1R3/1KB5/8 w - - 0 1
Analysis by Stockfish dev-20260101-28844fc6:

46.Kb3 Lh6 47.Tg3 Tc8 48.Tg6 Lxf4 49.Txe6 Lc1 50.Tb6 Ta8 51.Txb7 Txa3+ 52.Kb4 Ta8 53.Td7 Tb8 54.Txd5 Lf4 55.Kb3 Lh6 56.Ld3 Lg7 57.Kc2 Tc8 58.Lf5 Tb8 59.Kb3 Lf6 60.Td6 Lg7 61.Tc6 Lf8 62.Kc2 Lg7 63.Ld3 Lf8 64.Kb3 Lg7 65.Kb4 Lf8+ 66.Ka5 Td8 67.Lg6 Tb8 68.Lc2 Le7 69.Tg6 Lb4+
Tiefe: 49/99 00:00:51 1463MN, tb=3625498
+/- (1.07)

At this one point SF online went from 0.00 to 0.39, Reckless from 0.25 to 0.78.

So I'd say, yes, the endgame was won for White since quite some more time than the engines showed the surely winning evals online, e.g SF evaluates some sooner some higher at Backward even with some weaker hardware. Yet it's really difficult to say, from which point onwards it was won for sure, as well as it's even difficult to say, when the "endgame" started at all
:)
Peter.