Lc0 lost with white against Boot 6.3.1 --Bug?

[MikeGL]

That's a click bait title, because it is just probably a bug on the part of Lc0.

[pgn] [Event "TCEC Season 14 D2"] [Site "?"] [Date "?"] [Round "18"] [White "LCZero v19.1-RC2-11248"] [Black "Booot 6.3.1"] [Result "0-1"] [Termination "unterminated"] [PlyCount "241"] [WhiteType "Engine"] [BlackType "Engine"] 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. f3 O-O 6. Be3 c6 7. Qd2 c5 8. dxc5 dxc5 9. Bxc5 Qxd2+ 10. Kxd2 Nc6 11. Nd5 Rd8 12. Kc2 Nxd5 13. cxd5 Nd4+ 14. Bxd4 Bxd4 15. Rd1 Be3 16. Ne2 Kg7 17. Nd4 Bd7 18. Bc4 Rac8 19. Kd3 Bf4 20. g3 Bd6 21. Rc1 Bc5 22. e5 Bb6 23. Rc3 Ba5 24. Rc2 Ba4 25. b3 Bd7 26. a4 a6 27. Ke4 Rc5 28. Rhc1 h5 29. f4 Bg4 30. Ra1 Bh3 31. Rf2 Bc3 32. Rc1 Bb4 33. Rd1 Bg4 34. Rdf1 Bh3 35. Rd1 Bg4 36. Ra1 Bc3 37. Rc1 Bb4 38. Ra1 Bc3 39. Rc1 Bb4 40. Rcf1 Bh3 41. Ra1 Bc3 42. Raa2 Bg4 43. Rg2 Bb4 44. Rgf2 Be1 45. Rf1 Bc3 46. Rff2 Be1 47. Rf1 Bc3 48. Rh1 Bb4 49. h4 Rdc8 50. Raa1 Bc3 51. Ra3 Bb4 52. Raa1 Kf8 53. Rhg1 Bh3 54. Rad1 Rd8 55. Rc1 Rdc8 56. Rcd1 Rd8 57. Rb1 Bd7 58. Rgc1 Bh3 59. Kd3 Bg4 60. Ke3 Kg7 61. Nc2 Ba5 62. Nd4 Bb4 63. Ra1 Bh3 64. Ra2 Bg4 65. Raa1 Bh3 66. Ra2 Bg4 67. Rac2 Kf8 68. Kf2 Ba5 69. Ke3 Bb4 70. Ra1 Kg7 71. Raa2 Bh3 72. Kf3 Bg4+ 73. Ke3 Bh3 74. Kf3 Bg4+ 75. Kf2 Rcc8 76. Ke3 Rc5 {3-fold repetition} 77. Rab2 Be1 78. Rg2 Bh3 79. Rge2 Bb4 80. Rbc2 Bg4 81. Rf2 Kf8 82. Rg2 Kg7 83. Rgf2 Be1 84. Rh2 Bb4 85. Rhg2 Bh3 86. Rgf2 Be1 87. Rh2 Bg4 88. Rhg2 Bb4 89. Rh2 Kg8 90. Ra2 b5 91. axb5 axb5 92. Be2 Rc3+ 93. Bd3 Rxd5 94. Rhc2 Rxc2 95. Rxc2 Bc5 96. Rxc5 Rxc5 97. Bxb5 Rc1 98. Bd3 Re1+ 99. Kf2 Ra1 100. Ke3 Bh3 101. Kd2 Kf8 102. b4 Ra2+ 103. Bc2 Ke8 104. b5 Kd7 105. Kc3 Ra1 106. Kb4 Rg1 107. Bb3 e6 108. Ne2 Re1 109. Nd4 Re3 110. b6 Rxg3 111. Ba4+ Kc8 112. Bc6 Rg4 113. Kc5 Rxh4 114. Ba8 Kb8 115. Bh1 Bf1 116. Nc6+ Ka8 117. Bf3 Rxf4 118. Bh1 Rf2 119. Nd8+ Bg2 120. Bxg2+ Rxg2 121. Kc6 0-1 [/pgn]

[Event "TCEC Season 14 D2"]
[Site "?"]
[Date "?"]
[Round "18"]
[White "LCZero v19.1-RC2-11248"]
[Black "Booot 6.3.1"]
[Result "0-1"]
[Termination "unterminated"]
[PlyCount "241"]
[WhiteType "Engine"]
[BlackType "Engine"]

1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. f3 O-O 6. Be3 c6 7. Qd2 c5 8.
dxc5 dxc5 9. Bxc5 Qxd2+ 10. Kxd2 Nc6 11. Nd5 Rd8 12. Kc2 Nxd5 13. cxd5 Nd4+
14. Bxd4 Bxd4 15. Rd1 Be3 16. Ne2 Kg7 17. Nd4 Bd7 18. Bc4 Rac8 19. Kd3 Bf4
20. g3 Bd6 21. Rc1 Bc5 22. e5 Bb6 23. Rc3 Ba5 24. Rc2 Ba4 25. b3 Bd7 26. a4
a6 27. Ke4 Rc5 28. Rhc1 h5 29. f4 Bg4 30. Ra1 Bh3 31. Rf2 Bc3 32. Rc1 Bb4
33. Rd1 Bg4 34. Rdf1 Bh3 35. Rd1 Bg4 36. Ra1 Bc3 37. Rc1 Bb4 38. Ra1 Bc3
39. Rc1 Bb4 40. Rcf1 Bh3 41. Ra1 Bc3 42. Raa2 Bg4 43. Rg2 Bb4 44. Rgf2 Be1
45. Rf1 Bc3 46. Rff2 Be1 47. Rf1 Bc3 48. Rh1 Bb4 49. h4 Rdc8 50. Raa1 Bc3
51. Ra3 Bb4 52. Raa1 Kf8 53. Rhg1 Bh3 54. Rad1 Rd8 55. Rc1 Rdc8 56. Rcd1
Rd8 57. Rb1 Bd7 58. Rgc1 Bh3 59. Kd3 Bg4 60. Ke3 Kg7 61. Nc2 Ba5 62. Nd4
Bb4 63. Ra1 Bh3 64. Ra2 Bg4 65. Raa1 Bh3 66. Ra2 Bg4 67. Rac2 Kf8 68. Kf2
Ba5 69. Ke3 Bb4 70. Ra1 Kg7 71. Raa2 Bh3 72. Kf3 Bg4+ 73. Ke3 Bh3 74. Kf3
Bg4+ 75. Kf2 Rcc8 76. Ke3 Rc5 {3-fold repetition} 77. Rab2 Be1 78. Rg2 Bh3
79. Rge2 Bb4 80. Rbc2 Bg4 81. Rf2 Kf8 82. Rg2 Kg7 83. Rgf2 Be1 84. Rh2 Bb4
85. Rhg2 Bh3 86. Rgf2 Be1 87. Rh2 Bg4 88. Rhg2 Bb4 89. Rh2 Kg8 90. Ra2 b5
91. axb5 axb5 92. Be2 Rc3+ 93. Bd3 Rxd5 94. Rhc2 Rxc2 95. Rxc2 Bc5 96. Rxc5
Rxc5 97. Bxb5 Rc1 98. Bd3 Re1+ 99. Kf2 Ra1 100. Ke3 Bh3 101. Kd2 Kf8 102.
b4 Ra2+ 103. Bc2 Ke8 104. b5 Kd7 105. Kc3 Ra1 106. Kb4 Rg1 107. Bb3 e6 108.
Ne2 Re1 109. Nd4 Re3 110. b6 Rxg3 111. Ba4+ Kc8 112. Bc6 Rg4 113. Kc5 Rxh4
114. Ba8 Kb8 115. Bh1 Bf1 116. Nc6+ Ka8 117. Bf3 Rxf4 118. Bh1 Rf2 119.
Nd8+ Bg2 120. Bxg2+ Rxg2 121. Kc6 0-1

There are many alternatives to 90.Ra2 which would result to a draw, and 90.Ra2? move is triggered by some weird bug.

Notice that the move 75.Ke3= above would already result to a 3-fold repetition with draw. But of course Lc0 (white) avoided that drawing move since it cannot accept a mere draw considering it is already up a pawn plus some space advantage to booot. To be fair, Booot is no pushover since it is currently Top 4 on current standings for this tourn after the above win.

But this draw avoidance is similar bug which also haunted Rybka at ICC blitz which was punished by Nakamura. The bug is with draw detection and avoidance, even if GM Naka had a fortress or equal position, Rybka 3 at that time would push a pawn or would force a pawn capture move just to reset the 50-move-draw limit of the engine. That's where GM Naka punished Rybka after Naka won a pawn or two.

ps.
This is one of those games on how super GM Nakamura exploited this Rybka bug.
There are many more similar wins by Naka against Rybka (Top 1 engine at that time). http://www.chessgames.com/perl/chessgame?gid=1497429

Hope Lc0 programmers would fix this Lc0 bug too.

[Jouni]

Lc0 just lost in CCC3 to Houdini after +3 evaluation with white. Endgame blunder of course .

[Jouni]

Game here:

[pgn] [Event "CCC 3: Rapid Redux (30|5)"] [Site "Chess.com"] [Date "2018.12.12"] [Round "?"] [White "Lc0 0.19.0"] [Black "Houdini 6.03"] [Result "0-1"] [BlackElo "3400"] [ECO "D38"] [Opening "QGD"] [Time "06:04:55"] [Variation "Ragozin, Westphalia, 7.Qc2"] [WhiteElo "3200"] [TimeControl "1800+5"] [Termination "normal"] [PlyCount "370"] [WhiteType "program"] [BlackType "program"] 1. d4 Nf6 2. c4 e6 3. Nf3 d5 4. Nc3 Bb4 5. Bg5 Nbd7 6. cxd5 exd5 7. Qc2 c5 8. e3 c4 9. Be2 O-O 10. O-O Bxc3 11. bxc3 Re8 12. h3 h6 13. Bh4 b6 14. a4 Qc7 15. Nd2 Ne4 16. Nxe4 Rxe4 17. Bg3 Qc6 18. Qb2 Re8 19. Bf3 Nf6 20. Be5 Ne4 21. Bxe4 dxe4 22. Qb5 Qe6 23. Bg3 Rd8 24. a5 Bd7 25. Qe5 Qxe5 26. Bxe5 b5 27. a6 Bc6 28. Bc7 Rd7 29. Ba5 f5 30. Rfd1 Kf7 31. g3 Ke6 32. Kf1 Rc8 33. Bb4 g5 34. Ke2 Rcd8 35. h4 g4 36. Kd2 Rf7 37. Kc2 Kd5 38. Bc5 Rg7 39. Ra5 Rc7 40. Kb2 Rb8 41. Rb1 Rf7 42. Ra2 Ra8 43. Ra5 Re8 44. Rg1 Rb8 45. Ra2 Re8 46. Ra5 Re6 47. Rb1 Ref6 48. Rd1 Rc7 49. Re1 Re6 50. Rg1 h5 51. Bb4 Rf6 52. Ba3 Rd7 53. Bc5 Rh6 54. Kc2 Re6 55. Re1 Re8 56. Kb2 Re6 57. Kc2 Rf6 58. Rf1 Rg7 59. Kb2 Rc7 60. Re1 Rh6 61. Kc2 Rf6 62. Rb1 Be8 63. Re1 Rg6 64. Rd1 Rh7 65. Rb1 Re6 66. Re1 Rc7 67. Kb2 Bd7 68. Rea1 Re8 69. Rc1 Re6 70. Rca1 Rf6 71. Kc2 Re6 72. Rh1 Re8 73. Rg1 Rec8 74. Kb2 Rd8 75. Rd1 Rb8 76. Rh1 Bc6 77. Re1 Rbc8 78. Rb1 Rd7 79. Rf1 Rb8 80. Re1 Ra8 81. Rf1 Rb8 82. Rd1 Rc7 83. Rg1 Ra8 84. Rf1 Rd7 85. Rd1 Re8 86. Rg1 Re6 87. Rb1 Re8 88. Ra2 Rh7 89. Ra5 Rf7 90. Ra2 Re6 91. Ra5 Be8 92. Kc2 Ref6 93. Ba3 Rg6 94. Bc5 Bc6 95. Ba3 Be8 96. Bc5 Bc6 97. Kb2 Re6 98. Ka2 Ref6 99. Kb2 Ke6 100. d5+ Kxd5 101. Bd4 Rd6 102. Ka3 Ke6 103. Kb2 Rc7 104. Ka3 Rcd7 105. Kb4 Rf7 106. Ka3 Be8 107. Kb2 Rc7 108. Kc2 Rd8 109. Re1 Rd5 110. Kb2 Rf7 111. Ka3 Bc6 112. Rd1 Rd8 113. Rg1 Rb8 114. Kb2 Be8 115. Ka3 Rd8 116. Rd1 Rh7 117. Kb2 Rd5 118. Rc1 Bc6 119. Ka3 Rd8 120. Ka2 Rc7 121. Kb2 Rd5 122. Ka3 Rh7 123. Kb2 Rdd7 124. Raa1 Rd6 125. Ra5 Rhd7 126. Ka3 Rd5 127. Kb2 Ke7 128. Ka3 Kd6 129. Kb2 Re7 130. Kc2 Be8 131. Rb1 Rc7 132. Rc1 Bc6 133. Re1 Rd7 134. Kb2 Rc7 135. Kc2 Kd7 136. Rc1 Kc8 137. Kb2 Be8 138. Kc2 Kb8 139. Rb1 Bc6 140. Ra2 Ka8 141. Ra5 Rh7 142. Rd1 Rd8 143. Raa1 Rh6 144. Bg7 Rhd6 145. Rxd6 Rxd6 146. Be5 Rd5 147. Bd4 Kb8 148. Kb2 Ka8 149. Ka3 Kb8 150. Kb4 Bd7 151. Ra5 Be8 152. Ra2 Ka8 153. Bh8 Kb8 154. Bf6 Bf7 155. Bd4 Be8 156. Ka5 b4+ 157. Kxb4 Rb5+ 158. Kxc4 Kc7 159. Ra4 Kc6 160. Bg7 Bf7+ 161. Kd4 Rd5+ 162. Kc4 Bg8 163. Bh6 Rd6+ 164. Kb4 Rxh6 165. Ra5 Be6 166. Rb5 Rh7 167. Rb8 Rc7 168. Rh8 Kd6 169. Rxh5 Rc6 170. Rh7 Rb6+ 171. Ka4 Rxa6+ 172. Kb4 Rb6+ 173. Ka5 Rb2 174. h5 Bd7 175. Ka6 Bc6 176. Kxa7 Kc5 177. c4 Ra2+ 178. Kb8 Ra8+ 179. Kc7 Ra7+ 180. Kc8 Rxh7 181. h6 Kxc4 182. f3 Kc5 183. fxe4 fxe4 184. Kd8 Kd6 185. Kc8 Rh8# 0-1 [/pgn]

[MikeGL]

Lc0 just lost in CCC3 to Houdini after +3 evaluation with white. Endgame blunder of course .

Not sure if you're serious.
But I reviewed the game. Seems it is the same bug. Notice the last pawn push and the draw-counter starts at move 50...h5 of black, and the draw would therefore happen at move 100. But see how both engines conspired to enable white (Lc0) to make a useless pawn push at move one hundred with 100.d5? which only gave up a pawn for nothing, note that that d5 move can also be done at move 36 of Lc0 with the same idea of openning up the d-file.

So,
1. Is it a coincidence that upon reaching the 50-move-counter, Lc0 gave up d5 for free? I think laws of probability won't agree.
2. Why wait for move 100 when the same idea (move d5) can be done at move 36?

Takes two to tango, Houdini also felt the 50-move-draw on its next move and therefore made the K move to make Lc0 push a pawn (d5) to reset the draw counter. Maybe the bug is in syzygy.h 50 move draw counter or other modules, but i only understand C and not C++

I think Lc0 has some codes which is not as refined and as robust as that of Stockfish.