Strange game by Leela at CCC3

[Eduard]

I have not seen such a game yet!

[Event "CCC3: Rapid Redux Stage 3 (30|5)"]
[Site "?"]
[Date "2019.01.06"]
[Round "1"]
[White "Lc0"]
[Black "Houdini"]
[Result "1-0"]
[TimeControl "1800+5"]
[ECO "B07"]
[GameDuration "01:24:55"]
[GameEndTime "2019-01-06T21:01:17.832 PST"]
[GameStartTime "2019-01-06T19:36:22.051 PST"]
[Opening "Pirc defense"]
[PlyCount "319"]

1. e4 d6 2. d4 Nf6 3. Nc3 g6 4. Be3 Bg7 5. Qd2 O-O 6. O-O-O c6 7. h4 b5 8. f3
h5 9. Nce2 Nbd7 10. g4 hxg4 11. Ng3 e5 12. h5 exd4 13. Bh6 Ne8 14. hxg6 fxg6
15. f4 Qf6 16. N1e2 Nb6 17. Nxd4 Qxd4 18. Qxd4 Bxd4 19. Rxd4 Rf7 20. f5 Nc4 21.
Bxc4 bxc4 22. fxg6 Rf3 23. Rf1 Be6 24. Rdd1 c3 25. bxc3 Bc4 26. Rxf3 gxf3 27.
Rd4 Bxa2 28. Rd3 Nf6 29. Rxf3 Ng4 30. Bf4 Re8 31. Bxd6 Kg7 32. Rf4 Nh6 33. e5
Be6 34. Rf6 Bg4 35. Bc5 Ng8 36. Rxc6 Rc8 37. Nh5+ Kh6 38. Rxc8 Bxc8 39. Nf4 Kg5
40. Be3 Kf5 41. e6 Ke4 42. Bd2 Ke5 43. c4 Kf6 44. Bb4 Bb7 45. Bf8 Bc6 46. g7
Ke5 47. Ne2 Be8 48. e7 a6 49. Kb2 Nf6 50. Kb3 Ke6 51. Kb4 Bd7 52. Ka5 Ke5 53.
Kxa6 Be6 54. c5 Bd5 55. Kb6 Nd7+ 56. Kc7 Nf6 57. Nc3 Bf7 58. c6 Kd4 59. Na4 Be8
60. Nb2 Ke3 61. Nd3 Kd2 62. Kb6 Bf7 63. Nf2 Nd5+ 64. Kc5 Nc7 65. Ng4 Ne8 66. c4
Nc7 67. Kb6 Ne8 68. c5 Be6 69. Nf2 Ke2 70. Ne4 Ke3 71. Ng3 Nf6 72. Nf1+ Ke2 73.
Nh2 Nd5+ 74. Kb7 Nf6 75. Nf1 Bd5 76. Ng3+ Kf3 77. Nf5 Kf2 78. Kb6 Be6 79. Nd6
Bd7 80. Nb5 Be6 81. Na3 Nd5+ 82. Kb7 Nf6 83. Nc2 Bd5 84. Kb6 Bxc6 85. Na3 Be8
86. Nc4 Nd5+ 87. Kb7 Nf6 88. c6 Bf7 89. Ne5 Bh5 90. c7 Bf3+ 91. Kb8 Bg4 92. Ka7
Kg3 93. Kb6 Be6 94. Nd3 Kf3 95. Nc5 Bd7 96. Nb3 Kf4 97. Nc5 Kf3 98. g8=R Bc8
99. Rg6 Nd5+ 100. Ka7 Nxc7 101. Kb8 Bd7 102. Kxc7 Be8 103. Rg1 Kf2 104. Rg2+
Kxg2 105. Ne4 Bf7 106. Kb6 Kh1 107. Nd6 Bg6 108. Kc5 Kg2 109. Kd4 Kg3 110. Ke3
Kg4 111. e8=R Bxe8 112. Nxe8 Kg3 113. Ba3 Kg2 114. Bb2 Kg3 115. Bc1 Kg2 116.
Kd4 Kf3 117. Ba3 Kf4 118. Bb4 Kf3 119. Ba5 Kf4 120. Bb6 Kf5 121. Kc3 Ke6 122.
Ba5 Ke7 123. Nc7 Kd6 124. Nb5+ Kc6 125. Na3 Kb7 126. Bb4 Kc6 127. Kd4 Kb7 128.
Be1 Ka6 129. Kc5 Kb7 130. Nb1 Kb8 131. Bd2 Kb7 132. Bc3 Kc7 133. Kd5 Kb6 134.
Bd4+ Kc7 135. Nc3 Kc8 136. Kc6 Kb8 137. Nd5 Ka8 138. Bb2 Kb8 139. Nc7 Kc8 140.
Bd4 Kb8 141. Bb6 Kc8 142. Ba7 Kd8 143. Nd5 Ke8 144. Kd6 Kf7 145. Ne7 Kg7 146.
Be3 Kf6 147. Bf4 Kf7 148. Bg5 Ke8 149. Ke6 Kd8 150. Bf4 Ke8 151. Bc7 Kf8 152.
Nf5 Ke8 153. Ng7+ Kf8 154. Kf6 Kg8 155. Kg6 Kh8 156. Nf5 Kg8 157. Bd6 Kh8 158.
Be7 Kg8 159. Nh6+ Kh8 160. Bf6# 1-0

Position after move 15:

[d]r1b1nrk1/p2n2b1/2pp1qpB/1p6/3pPPp1/6N1/PPPQ4/2KR1BNR w - - 1 16

Leela plays like a world champion, or even better!

Position after move 50:

[d]4bB2/4P1P1/p3kn2/8/2P5/1K6/2P1N3/8 w - - 4 51

Leela plays fantastic! Great endgame! But now:

Position after move 104 white:

[d]4bB2/2K1P3/8/2N5/8/8/5kR1/8 b - - 4 104

Leela got 2 rooks to sacrifice both. No joke!!

Position after move 112:

[d]4NB2/8/8/8/8/4K1k1/8/8 w - - 1 113

Endgame KBNvK!

Position after move 160:

[d]7k/8/5BKN/8/8/8/8/8 b - - 96 160

Leela needed 48 moves for this CheckMate! Puh...

[AdminX]

I am not surprised by her endgames, this is why I still prefer 11258 overall. 32395 and 32425 may indeed be stronger, but those trolling endgames leave allot to be desired or as the young people say 'SMH', translation: Shaking My Head.

[M ANSARI]

I think this is quite normal and if no solution is found, maybe a hybrid engine makes a lot of sense. This is an exchange I read today that might make sense of what is going on

In the chat of TCEC you mentioned, “In this version of Komodo MCTS, once it sees it has a score of 7 pawns ahead (or 7 pawns behind), it uses Syzygy and regular search to finish off the game.” Does that comply with the rules that an engine has to be “unique”?

Mark Lefler: I will leave rule interpretation up to the people running TCEC, but the main reason we do this is that once a game is almost certainly won, alpha beta is just better at coming up with a faster end to the game. In alpha-beta search, win scores are scaled based on the number of moves until mate. In MCTS, a win is a win, and this gets converted to a 100% value, so shortest wins are not always found, so it can take a long time to get there. We thought it was not so interesting postponing the result. We are considering making this “mop up” feature an option people can turn on or off. There is not much of a reason to do the switch when we are totally losing, we may take that out. MCTS may well be better in such cases.