SF loss in Game 25 of TCEC 21 Sufi: French Tarrasch

361 views
Skip to first unread message

Bryan

unread,
Jul 26, 2021, 10:15:44 AMJul 26
to FishCooking
SF as black lost to LC0 in Game 25 of the TCEC 21 Sufi, and then played to a draw in the reverse in Game 26. I used the latest net (nn-26abeed38351.nnue) to do a deep reverse search (d=50, threads=15) from moves 23 to 13 on Game 25, to see what this might reveal about where SF played suboptimal moves.

Move 13

At move 13 SF played a7-a6. According to the reverse analysis this lost about 14 centipawns, and the better move here is h7-h6 (which SF played at move 18) and delaying a7-a6 until move 16. In the reverse game LC0 played h7-h6 at move 13, and played a7-a6 at move 14.

Move 13: black to move: hash from subsequent moves: multipv=1: threads=15:
FEN: r1b1kb1r/p2n1ppp/1q2p3/1p1pP3/1P6/P2B1N2/5PPP/R1BQ1RK1 b kq b3 0 13
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
50/81    2:43:17     77,958,832k    7,957k    -0.79    h7-h6  Rf1-e1  Bf8-e7  Bc1-e3  Qb6-b7  h2-h4  a7-a6  h4-h5  O-O  Bd3-c2  f7-f5  e5xf6/ep  Nd7xf6  Nf3-e5  Be7-d6  Be3-d4  a6-a5  Qd1-d3  a5xb4  a3xb4  Ra8xa1  Re1xa1  Qb7-c7  Ne5-g4  Nf6-e4  Ra1-a7  Bc8-b7  f2-f3  Rf8-f4  f3xe4  Rf4xg4  e4-e5  Bd6xb4  Qd3-h7+  Kg8-f8  Qh7-h8+  Kf8-f7  Ra7-a1  Qc7-c8  Bc2-g6+  Rg4xg6  Qh8xc8  Rg6xg2+  Kg1xg2  Bb7xc8  Ra1-a7+  Kf7-e8  Ra7xg7  Bb4-f8  Rg7-a7  Bc8-d7  Ra7-a8+  Ke8-f7  Kg2-f3  Bd7-e8  Ra8-b8  b5-b4  Kf3-e2  Be8-a4  Ke2-d2  Ba4-b3  Rb8-b7+  Kf7-g8  Rb7-b6  Kg8-f7  Rb6-c6  Bb3-c4  Rc6-c7+  Kf7-e8  Bd4-e3  b4-b3  Rc7-c8+  Ke8-f7  Be3-c5  Bf8-e7  Bc5xe7  Kf7xe7

Move 14: black to move: hash from subsequent moves: multipv=1: threads=15:
FEN: r1b1kb1r/3n1ppp/pq2p3/1p1pP3/1P6/P2BBN2/5PPP/R2Q1RK1 b kq - 1 14
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
50/64    1:08:42     33,564,918k    8,142k    -0.93    Qb6-d8  h2-h4  Bc8-b7  Qd1-e2  h7-h6  h4-h5  Ra8-c8  Be3-d4  Rc8-c4  Qe2-e3  Qd8-c7  Bd4-b2  Bf8-e7  Ra1-e1  O-O  Nf3-d4  Qc7-b6  Re1-d1  Rf8-a8  Rf1-e1  Nd7xe5  Qe3xe5  Be7-f6  Qe5-e3  Bf6xd4  Bb2xd4  Rc4xd4  f2-f4  Qb6-a7  Bd3-h7+  Kg8xh7  Qe3xd4  Qa7xd4+  Rd1xd4  Ra8-d8  Rd4-d2  d5-d4  Re1-d1  g7-g6  Rd2xd4  Bb7-d5  Rd1-c1  g6xh5  f4-f5  Rd8-g8  f5xe6  f7xe6  Rc1-c7+  Kh7-g6  Rc7-a7  Kg6-f5  Rd4-d2  Kf5-e4  Ra7xa6  Ke4-e3  Rd2-b2  Ke3-d3  Kg1-h2  Rg8-g4  Ra6-b6  Kd3-c4  Rb2-d2  Kc4-c3  Rd2-f2

Move 15

At move 15 SF played Bc8-b7. According to the reverse analysis this lost about 19 centipawns, and the better move here is h7-h6 and delaying Bc8-b7 for 9 moves. In this game SF played h7-h6 at move 18. In the reverse game, LC0 played h7-h6 at move 13 and Bb8-b7 at move 15.

Move 15: black to move: hash from subsequent moves: multipv=1: threads=15:
FEN: r1bqkb1r/3n1ppp/p3p3/1p1pP3/1P5P/P2BBN2/5PP1/R2Q1RK1 b kq h3 0 15
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
50/74    1:29:52     43,824,957k    8,126k    -0.80    h7-h6  Ra1-c1  Nd7-b6  Rc1-c6  Nb6-c4  Bd3xc4  d5xc4  Qd1xd8+  Ke8xd8  Rf1-d1+  Kd8-e8  Rc6-c7  Bf8-e7  Be3-b6  c4-c3  Rc7xc3  Bc8-b7  Rd1-c1  f7-f6  h4-h5  Rh8-f8  Nf3-d4  Bb7-d5  Nd4-c6  Rf8-f7  Nc6xe7  Rf7xe7  e5xf6  g7xf6  Rc3-g3  Ke8-f7  Bb6-e3  Ra8-g8  Rg3xg8  Kf7xg8  Rc1-c8+  Kg8-g7  Rc8-d8  Bd5-b7  f2-f4  Bb7-d5  Be3-c5  Re7-c7  Kg1-f2  Kg7-f7  g2-g4  a6-a5  Rd8-b8  a5xb4  a3xb4  Rc7-b7  Rb8-f8+  Kf7-g7  Rf8-d8  Bd5-c4  Kf2-g3  Rb7-c7  Bc5-d4  Bc4-e2  Rd8-d6  Kg7-f7  g4-g5  h6xg5  f4xg5  f6xg5  h5-h6  Be2-d3  Bd4-e5  Bd3-f5  Rd6-d8  Rc7-e7

Move 16

At move 16 SF played Bf8-e7. According to the reverse analysis this lost about 12 centipawns, and the better move here is Ra8-c8 and delaying Bf8-e7 for 3 moves. In this game SF played Ra8-c8 at move 20. In the reverse game LC0 played Ra8-c8 at move 16, and played Bf8-e7 at move 25.

Move 16: black to move: hash from subsequent moves: multipv=1: threads=15:
FEN: r2qkb1r/1b1n1ppp/p3p3/1p1pP2P/1P6/P2BBN2/5PP1/R2Q1RK1 b kq - 0 16
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
50/70    2:00:14     58,820,590k    8,153k    -0.99    Ra8-c8  Qd1-e2  h7-h6  Be3-d4  Rc8-c4  Bd4-b2  Bf8-e7  Nf3-d4  O-O  Ra1-d1  Qd8-b6  Qe2-e3  Rf8-a8  Rf1-e1  Kg8-h8  Qe3-f4  Ra8-f8  Qf4-g3  Rf8-g8  Bd3xc4  b5xc4  Re1-e3  Bb7-c6  Qg3-f4  Be7-g5  Qf4xf7  Bg5xe3  f2xe3  Nd7xe5  Qf7xe6  Rg8-e8  Qe6-d6  Qb6-d8  Qd6xd8  Re8xd8  Nd4xc6  Ne5xc6  Kg1-f2  Kh8-g8  Kf2-e2  Rd8-d6  Bb2-d4  g7-g6  Bd4-c5  Rd6-d7  g2-g4  g6xh5  g4xh5  Kg8-f7  Rd1-g1  d5-d4  Rg1-g6  Nc6-e5  Rg6xa6  d4-d3+  Ke2-d2

Move 17

At move 17 SF castled short. According to the reverse analysis this lost about 4 centipawns, and the slightly better move here is Ra8-c8 and delaying castling for 4 more moves. In this game SF played Ra8-c8 at move 19. In the reverse game, LC0 played Ra8-c8 at move 16, and never castled.

Move 17: black to move: hash from subsequent moves: multipv=1: threads=15:
FEN: r2qk2r/1b1nbppp/p3p3/1p1pP2P/1P6/P2BBN2/4QPP1/R4RK1 b kq - 2 17
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
50/65    2:30:28     73,128,724k    8,100k    -1.11    Ra8-c8  Be3-d4  Rc8-c4  Qe2-e3  Qd8-c7  Rf1-e1  h7-h6  Nf3-d2  O-O  Nd2xc4  d5xc4  Bd3-c2  a6-a5  Bd4-c3  a5xb4  Bc3xb4  Be7-c5  Bb4xc5  Qc7xc5  Qe3xc5  Nd7xc5  Ra1-b1  Bb7-c6  Re1-d1  Rf8-a8  Rd1-d6  Bc6-e8  Rb1-d1  Kg8-f8  Rd6-d8  Ra8xa3  Rd8-c8  Nc5-d3  f2-f4  Kf8-e7  Bc2xd3  c4xd3  Kg1-f2  Ra3-a2+  Kf2-e3  Ra2xg2  Ke3xd3  Rg2-a2  Rc8-c7+  Ke7-d8  Rd1-c1  Ra2-a3+  Kd3-e4  Ra3-a4+  Ke4-e3  Ra4-a3+  Ke3-f2  Ra3-a4  Rc7-c8+  Kd8-e7

Move 18

At move 18 SF played h7-h6. According to the reverse analysis this lost about 16 centipawns, and the better move here is Rf8-e8, allowing the thorn pawn on h6. In the game SF played Rf8-e8 at move 21.

Move 18: black to move: hash from subsequent moves: multipv=1: threads=15:
FEN: r2q1rk1/1b1nbppp/p3p3/1p1pP2P/1P1B4/P2B1N2/4QPP1/R4RK1 b - - 4 18
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
50/57    1:18:49     40,757,247k    8,617k    -1.15    Rf8-e8  h5-h6  g7-g6  Qe2-e3  Ra8-c8  Rf1-b1  Bb7-c6  Rb1-b2  Be7-f8  Bd3-f1  Bf8-e7  Rb2-a2  Rc8-c7  Bf1-d3  Re8-f8  Ra1-b1  Rf8-e8  Ra2-e2  Be7-f8  Rb1-a1  Bc6-b7  Qe3-f4  Rc7-c8  Re2-e1  Nd7-b6  Nf3-h2  Bf8-e7  Nh2-g4  Rc8-c6  Ra1-d1  Be7-g5  Qf4-f3  Qd8-e7  Qf3-h3  Qe7-c7  Qh3-g3  Qc7-d8  Rd1-a1  Re8-f8  Bd4-c5  Rc6xc5  b4xc5  Nb6-a4  Ra1-c1  Qd8-c7  Qg3-f3  Na4xc5  Ng4-f6+  Kg8-h8  Rc1-c2  Rf8-d8  Qf3-g4  Bg5xf6  e5xf6  Qc7-d6  Bd3-f1  d5-d4

Move 19

At move 19 SF played Ra8-c8. According to the reverse analysis this lost about 23 centipawns, and the better move here is Qd8-b8.

Move 19: black to move: hash from subsequent moves: multipv=1: threads=15:
FEN: r2q1rk1/1b1nbpp1/p3p2p/1p1pP2P/1P1B4/P2BQN2/5PP1/R4RK1 b - - 1 19
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
50/69    55:19     27,656,935k    8,332k    -1.31    Qd8-b8  g2-g4  Rf8-c8  g4-g5  h6xg5  Nf3xg5  Rc8-c4  h5-h6  g7xh6  Ng5xe6  Rc4xd4  Ne6xd4  Qb8xe5  Qe3xh6  Qe5-g5+  Qh6xg5+  Be7xg5  Bd3-f5  Nd7-b6  Ra1-d1  Nb6-c4  Rd1-d3  Kg8-f8  Bf5-h3  Ra8-e8  Nd4-b3  Bg5-e7  Bh3-g2  Re8-d8  Nb3-c5  Be7xc5  b4xc5  Rd8-d7  Rf1-e1  Rd7-c7  Bg2xd5  Bb7xd5  Rd3xd5  Nc4xa3  Re1-a1  b5-b4  Kg1-g2  Rc7-c6  Ra1-c1  b4-b3  Rd5-d8+  Kf8-g7  Rd8-b8  Na3-b5  Rc1-b1  Rc6xc5  Rb1xb3  Nb5-c3  Rb8-b7  a6-a5  Rb7-a7  Kg7-g6  Rb3-b7  f7-f6  Rb7-c7  Rc5xc7

Move 20: black to move: hash from subsequent moves: multipv=1: threads=15:
FEN: 2rq1rk1/1b1nbpp1/p3p2p/1p1pP2P/1P1B4/P2BQ3/5PPN/R4RK1 b - - 3 20
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
50/80    54:05     27,256,575k    8,399k    -1.54    Rf8-e8  Nh2-g4  Rc8-c4  Ra1-d1  Qd8-c7  Ng4xh6+  g7xh6  Qe3xh6  Nd7-f8  Bd3-b1  Re8-c8  Rd1-d3  Be7-h4  Rf1-d1  Rc4-c1  Qh6-f4  Rc1xd1+  Rd3xd1  Bh4-d8  Qf4-g4+  Kg8-h8  h5-h6  Nf8-g6  Bb1xg6  f7xg6  Qg4xe6  Kh8-h7  Bd4-c5  Bb7-c6  Bc5-b6  Qc7-b7  Bb6xd8  Rc8xd8  Qe6-f6  Rd8-a8  Rd1-c1  Bc6-e8  Qf6-f8  Be8-c6  Qf8-d6  Ra8-c8  Qd6-f6  d5-d4  e5-e6  Rc8-g8  Qf6xd4  Bc6xg2  Qd4-f4  Bg2-c6  Qf4-f7+  Qb7xf7  e6xf7  Rg8-f8  Rc1xc6  Rf8xf7  Rc6xa6  Rf7-f5  Kg1-g2  Rf5-g5+  Kg2-f3  Rg5-f5+  Kf3-g3  Rf5-g5+  Kg3-h4  Rg5-f5  Ra6-a5  Kh7xh6  Kh4-g3  Kh6-h5  f2-f4  g6-g5  a3-a4  Rf5xf4  Ra5xb5  Rf4-c4  a4-a5  Rc4-c3+  Kg3-h2  Rc3-c2+  Kh2-g1

Move 22


At move 22 SF played Be7-f8. According to the reverse analysis this lost about 46 centipawns, and the better move here is Bb7-c8, and then, after Kg8-h8, putting the knight on f8 instead and sacrificing the rook on c4 to open the long diagonal for black's light-squared bishop.

Move 22: black to move: hash from subsequent moves: multipv=1: threads=15:
FEN: 3qr1k1/1b1nbpp1/p3p2p/1p1pP2P/1PrB2N1/P2BQ3/5PP1/3R1RK1 b - - 7 22
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
50/69    3:30:12    103,732,792k    8,225k    -1.58    Bb7-c8  Qe3-g3  Kg8-h8  Bd3-b1  Nd7-f8  Qg3-e3  Be7-g5  f2-f4  Bg5-e7  Qe3-d2  a6-a5  Ng4-e3  a5xb4  Ne3xc4  d5xc4  a3xb4  Qd8-c7  Bd4-c3  Bc8-b7  Qd2-f2  Kh8-g8  Bb1-c2  Re8-d8  Rd1xd8  Be7xd8  Rf1-a1  Qc7-c6  Ra1-d1  Bd8-b6  Bc3-d4  Bb6xd4  Rd1xd4  Qc6-c8  Rd4-d6  Qc8-a8  Kg1-h2  Bb7-d5  Qf2-d2  Qa8-a7  Rd6-d8  Qa7-e7  Rd8-b8  Qe7-h4+  Kh2-g1  Qh4xh5  Rb8xb5  Qh5-g4  Rb5-b8  g7-g5  f4xg5  c4-c3  Qd2-f2  h6xg5  b4-b5  Bd5-e4  Rb8-c8  Be4xc2  Qf2xc2  Qg4-d4+  Kg1-f1

Move 23: black to move: hash from subsequent moves: multipv=1: threads=15:
FEN: 3qrbk1/1b1n1pp1/p3p2p/1p1pP2P/1PrB2N1/P2B2Q1/5PP1/3R1RK1 b - - 9 23
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
50/82    07:06      3,877,233k    9,085k    -2.04    Qd8-g5  Bd3-e2  Kg8-h8  f2-f4  Qg5xh5  Ng4-f6  Qh5xe2  Nf6xe8  Bb7-c6  f4-f5  e6xf5  Rf1-e1  Qe2-g4  Qg3xg4  f5xg4  e5-e6  f7xe6  Re1xe6  Kh8-g8  Ne8-d6  Bf8xd6  Re6xd6  h6-h5  g2-g3  Kg8-f7  Kg1-f2  Nd7-f6  Kf2-e3  h5-h4  g3xh4  g4-g3  Rd1-g1  Kf7-e7  Bd4-c5  Bc6-d7  Rd6xd5+  Ke7-f7  Rd5-d4  Rc4-c3+  Rd4-d3  Rc3-c2  Rg1xg3  Bd7-f5  Rd3-d4  Rc2-c3+  Ke3-f2  Rc3-c2+  Kf2-g1  Rc2-c1+  Kg1-h2  Nf6-e4  Rg3-f3  Kf7-e6  Rd4-d8  Ne4xc5  b4xc5  Rc1xc5  Rd8-e8+  Ke6-f6  Re8-a8  Kf6-e5  Kh2-g3  Bf5-c8  Rf3-e3+  Ke5-d4  Re3-e7  g7-g6  Re7-e8  Bc8-f5  Ra8xa6  Rc5-c1  a3-a4  b5xa4  Ra6xa4+  Kd4-c5

Move 29

I also noticed that the latest SF still has difficulty finding the zwischenzug move 29.Rf1-e1. On one thread, even to d=57 and 13.8 billion nodes SF does not find it.

Move 29: white to move: hash cleared: multipv=1: threads=1
FEN: 4Nb1k/3n1pp1/p1b4p/1p1pPp2/1PrB4/P5Q1/4q1P1/3R1RK1 w - - 0 29
Stockfish_21072418_x64_bmi2:
NNUE evaluation using nn-26abeed38351.nnue enabled
[...]
 57/85 3:17:37 13,844,522k 1,168k +1.16 e5-e6 f5-f4 Qg3xf4 Qe2xe6 Qf4xf7 Qe6-g4 Qf7-f4 Qg4xf4 Rf1xf4 Kh8-g8 Rf4-f1 h6-h5 Ne8-c7 Bf8-e7 Rf1-f5 Nd7-f6 Nc7xa6 Kg8-f7 Na6-c5 Be7xc5 Bd4xc5 Kf7-g6 Rf5-e5 d5-d4 Re5-e6 Bc6-d5 Re6-b6 Rc4-c2 Rd1xd4 Rc2xg2+ Kg1-f1 Rg2-g5 Rb6xb5 Nf6-e4 a3-a4 Rg5-f5+ Kf1-g1 Ne4-c3 Rb5-b6+ Kg6-h7 Rd4-d2 Nc3xa4 Rb6-d6 Na4-c3 Rd6-d8 Bd5-e4 Rd8-e8 Be4-c6 Re8-e6 Bc6-a8 Rd2-d4 Rf5-g5+ Kg1-f2 Rg5-f5+ Kf2-e1 Ba8-d5 Re6-e7 Kh7-h6 Rd4-d2 g7-g5 Re7-e3 Nc3-b5 Rd2-f2 Rf5xf2 Ke1xf2 Kh6-g6 Re3-e5 Nb5-c3

Mehdi Amini Valashani son of Bahram

unread,
Jul 28, 2021, 8:14:52 AMJul 28
to FishCooking
On move 15 ... Bb7 seems like a mistake according to my brief analysis. 15 ... h6 , then castle then develop queen-side.

Laoshi Ma

unread,
Aug 13, 2021, 1:34:18 AMAug 13
to FishCooking
Some of these I'm not so sure about. Yes, 16. h5 Rc8 (or at least 17. ... Rc8 which may transpose;
in the game, 19. ... Rc8 came too late) would likely have helped, going for active counterplay
instead of preparing to castle into White's king-side attack, as far as I understand it. But this also
means that things aren't yet so bad for Black after 13. ... a6 and 15. ... Bb7, while after 13. ... h6
Black also has to contend with the possibility of 14. Be3. So it's not clear to me that 13. ... h6 is
actually better; more generally, when the eval difference is just a couple cp, one iteration more or
less could well change the move preference.

Instead of going deeper and deeper into the game, isn't there something at the beginning
as well? Couldn't SF on its first move have played 7. ... b4 8. 0-0 Nc6, avoiding the (too?) early
Qb6, which in the game led to some painful tempo losses? If this line holds up, then the way
to play the book exit would be 7. a3 (SF's choice), a move order which discourages 7. ... b4.

To the previous responder:
>On move 15 ... Bb7 seems like a mistake according to my brief analysis. 15 ... h6 ,
>then castle then develop queen-side.
Bryan _already_ posted that 15. ... h6 probably leads to 19. ... Ke8xd8, after which Black
lost castling rights? Give-or-take a transposition, I also get his line in that variation. You may
of course disagree with his PV for 15. ... h6, but then please tell us where and why.

Mehdi Amini Valashani son of Bahram

unread,
Aug 13, 2021, 11:44:47 AMAug 13
to FishCooking
It seems to me after 15...h6 , 16.Rc1 , 16...Be7 is better than 16...Nb6 .

Laoshi Ma

unread,
Aug 13, 2021, 3:08:38 PMAug 13
to FishCooking
> It seems to me after 15...h6 , 16.Rc1 , 16...Be7 is better than 16...Nb6 .
Hm; SF surely likes 15. h4 h6 16. Rc1 Be7 17. Bb1 -- for white. :-) And then there's
17. Rc6 as a more aggressive option; don't know how either move works out at high
depth. Then again, now I've seen a bit more, 15. ... Bb7 16. h5 Rc8 isn't that clear
yet either, so alternatives should certainly be explored.

All in all, I don't know of a reply to 13. b4 yet that one can feel *sure* is good enough
for a draw. What is it that makes that move so strong? Prepares Bb2? Claims the
c5 square for White? Prevents b4 by Black? A little bit of each? Only thing that's
clear to me so far is that this 6. ... b5 Tarrasch is a good deal worse for Black than
the main lines.

Laoshi Ma

unread,
Aug 15, 2021, 2:22:31 PMAug 15
to FishCooking
OK I'm calling it: this line 15. h4 h6 16. Rc1 Nb6 etc is a solid way to draw after 13. b4
(of course, I'm not claiming it's the only drawing line). I agree with Bryan's PV for it until
22. ... c3; after that the variations branch out a bit, but all of them moob hard by move
30 or so. It's not just the evals: the resulting more open positions with queens off, plus
two center pawns and a pair of minors traded, look (but what do I know) as something
where one can trust an SF moob to be the real thing.

The resulting eval drop for this 13. b4 line, played by both SuFi contestants, reorders
the whole variation tree, leading to the question if White's early play could be improved.
For example, in game 26, should SF have tried 9. Nf3 instead of 9. 0-0? Is this an
example of what cc players seem to say, "Don't castle prematurely if you want the full
point"?
Reply all
Reply to author
Forward
0 new messages