Stockfish-Leela Game 66

[Michael Neish]

This is currently under way.

https://tcec-chess.com/

There's a sudden and massive jump in the evaluation of both engines within a couple of moves. The game was apparently heading for a draw before that, but there must have been something pretty significant beyond the search horizon of both engines that changed the whole outlook after it came into view.

Just in case anyone is interested in this, can someone figure out what was the crucial bit of understanding that both engines were lacking (assuming that this is what it is)?

Regards,

Mike.

[Michael Neish]

It would appear that Leela's 42 ... Rg3 was weak, a kind of blunder. Stockish didn't evaluate it because it didn't expect it. So I suppose that it points to tactical weakness in Leela, but pardon me I am not a strong Chess player and don't really know what I'm talking about.

Any comments would be interesting.

[Guenther]

This is currently under way.

https://tcec-chess.com/

There's a sudden and massive jump in the evaluation of both engines within a couple of moves. The game was apparently heading for a draw before that, but there must have been something pretty significant beyond the search horizon of both engines that changed the whole outlook after it came into view.

That's not true. Only Black evaluated it wrong and played the losing Rg3?? (with the faulty idea of Rxe3??)
SF knew about this and had the right Bd7 in the PV.

Just in case anyone is interested in this, can someone figure out what was the crucial bit of understanding that both engines were lacking (assuming that this is what it is)?

Regards,

Mike.

LC0 completely 'missed' the Q 'sac' 46. Qxe5 in its previous calculations (actually it's no sac, because W wins it immediately back with a dead won position).
That's it. Obviously a blind spot for LC0.

[peter]

Any comments would be interesting.

[pgn][Event "TCEC Season 21 - Superfinal"]
[Site "https://tcec-chess.com"]
[Date "2021.08.01"]
[Round "66.1"]
[White "Stockfish 14_202107131735"]
[Black "LCZero 0.28-dev+_69626"]
[Result "1-0"]
[ECO "E81"]
[WhiteElo "3639"]
[BlackElo "3623"]
[PlyCount "135"]
[EventDate "2021.??.??"]

1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. f3 O-O 6. Bg5 Nc6 7. Nge2 a6 8. Qd2
Bd7 {book} 9. h4 h5 10. O-O-O Na5 11. Nf4 Nc6 12. Nfe2 Na5 13. Nf4 Nc6 14. g4
hxg4 15. fxg4 Nh7 16. Nfd5 Re8 17. Bh6 e5 18. Bxg7 Kxg7 19. dxe5 Nxe5 20. g5 c6
21. Ne3 Qe7 22. Be2 Rab8 23. Rdf1 b5 24. Kb1 b4 25. Ncd1 a5 26. Rhg1 a4 27. h5
Bh3 28. Rf2 Kg8 29. Rh2 Bd7 30. Rhg2 Be6 31. Rh1 Kg7 32. Nf2 a3 33. b3 Nxg5 34.
hxg6 Rh8 35. Rxh8 Kxh8 36. gxf7 Ngxf7 37. Nd3 Nxd3 38. Bxd3 Rg8 39. Rh2+ Kg7
40. Qxb4 Qa7 41. e5 Kf8 42. Kc2 Rg3 43. exd6 Rxe3 44. Rh8+ Kg7 45. Qc3+ Re5 46.
Qxe5+ Nxe5 47. Rh7+ Kf6 48. Rxa7 c5 49. Be4 Nf7 50. Rxf7+ Bxf7 51. Bf5 Bg8 52.
d7 Ke7 53. Kc3 Bf7 54. b4 cxb4+ 55. Kxb4 Kd8 56. Kc5 Kc7 57. Be4 Kd8 58. Bd5
Bg6 59. Kd6 Be4 60. Be6 Bc6 61. Kxc6 Ke7 62. Bg8 Kf6 63. d8=Q+ Kg6 64. Be6 Kh5
65. Bf5 Kh6 66. Qf6+ Kh5 67. Qg6+ Kh4 68. Qg4# 1-0
[/pgn]

With automatically stored TCEC - GUI- comments from position after 42.Kc2 before the crucial blunder 42...Rg3?
BTW SF did evaluate that clearly won for white at once after it was played, the ponder- move in output of SF's, before 42. black move (after 42.Kc2), would have been 42...Bd7 :

[d]5kr1/q4n2/2ppb3/4P3/1QP5/pP1BN3/P1K4R/8 b - - 2 42

Code: Select all

42.Kc2 { d=50, sd=66, pd=dxe5, mt=63461, tl=2246745, s=121309838, n=7698322341, pv=Kc2 Bd7 Nd1 Bg4 Nf2 c5 Qc3 Nxe5 Be4 Qg7 Qa5 Nc6 Qxa3 Nd4+ Kb1 Bf5 Rh1 Qg2 Bxf5 Nxf5 Qb2 Qxh1+ Nxh1 Rg1+ Kc2 Rg2+ Kc3 Rxb2 Kxb2 Ke7 Nf2 d5 Kc3 Kd7 Ng4 Kc6 Kd3 dxc4+ Kxc4 Nd6+ Kc3 Nb5+ Kd3 Nd6 Ne5+ Kd5 Nc4 Ne8 Kc3 Nf6 Ne3+ Kc6 Kc4 Nd7 Kd3 Nf6 a3 Kb5 Nc4 Kc6 Nd2 Kd5 Kc3 Kc6 Kc4 Nd7, tb=14728182, h=15.5, ph=66.6, 
wv=0.37, R50=49, Rd=-11, Rr=-1000, mb=+1+0+0+0+0, } 

42...Rg3 { d=36, sd=62, pd=Kc2, mt=1587, tl=3704166, s=183631, n=142951445, pv=Rg3 exd6 Rxe3 d7+ c5 Qd2 Rh3 d8=N Nxd8 Rf2+ Nf7 Bf1 Qb6 Bxh3 Bxh3 Qg5 Qd6 Qh5 Be6 Qh8+ Ke7 Qf6+ Kf8 Re2 Bd7 Qxd6+ Nxd6 Re5 Nb7 Kc3 Kf7 b4 Kf6 Rh5 cxb4+ Kxb4 Ke7 Kxa3 Kd6 Kb4 Nd8 Rh6+ Kc7 a4 Nc6+ Kc5 Na5 Ra6 Nb7+ Kd5 Nd8, tb=1881, h=0.0, ph=78.7, 
wv=0.08, R50=48, Rd=9, Rr=-1000, mb=+1+0+0+0+0, } 

43.exd6 { d=38, sd=63, pd=Bd7, mt=117691, tl=2139054, s=136344560, n=16046391336, pv=exd6 Bd7 Nf1 Rxd3 Kxd3 Qg1 Ke2 Qg4+ Ke3 Qf5 Qe1 Qe5+ Kf2 Qxd6 Kg1 Bf5 Qe3 Qe7 Qf3 Be4 Rh8+ Kg7 Qc3+ Qe5 Qxe5+ Nxe5 Ra8 Nd3 Nd2 Bf5 Rxa3 Kf6 Ra8 Ke5 Rd8 Ke6 a4 Ke7 Rh8 Kd6 Rh5 Bc8 Rg5 Be6 Rg7 Bc8, tb=24850187, h=25.6, ph=64.7, 
wv=5.76, R50=50, Rd=-11, Rr=-1000, mb=+2+0+0+0+0, } 

43...Rxe3 { d=1, sd=1, pd=exd6, mt=2, tl=3714164, s=69929829500, n=139859659, pv=Rxe3 d7+ c5 Qd2 Rh3 d8=N Nxd8 Rf2+ Nf7 Bf1 Qb6 Bxh3 Bxh3 Qg5 Qd6 Qh5 Be6 Qh8+ Ke7 Qf6+ Kf8 Re2 Bd7 Qxd6+ Nxd6 Re5 Nb7 Kc3 Kf7 b4 Kf6 Rh5 cxb4+ Kxb4 Ke7 Kxa3 Kd6 Kb4 Nd8 Rh6+ Kc7 a4 Nc6+ Kc5 Na5 Ra6 Nb7+ Kd5 Nd8, tb=0, h=0.0, ph=79.4, 
wv=0.08, R50=50, Rd=-11, Rr=-1000, mb=+2-1+0+0+0, }

44.Rh8+ { d=32, sd=38, pd=Bd7, mt=69063, tl=2079991, s=190016760, n=13123127551, pv=Rh8+ Kg7 Qc3+ Re5 Qxe5+ Nxe5 Rh7+ Kf6 Rxa7 Nf3 c5 Bg4 Re7 Be6 Bc4 Bf5+ Kc3 Ng5 Kb4 Bg4 Kxa3 Nh3 Rc7 Ng5 Kb4 Nh3 Rf7+ Ke5 Kc3 Ng5 Re7+ Kf4 a4 Bc8 a5, tb=13788352, h=9.1, ph=62.8, 
wv=13.07, R50=49, Rd=-11, Rr=-1000, mb=+2-1+0+0+0, }

 44...Kg7 { d=19, sd=45, pd=d7+, mt=2237906, tl=1486258, s=163533, n=366022854, pv=Kg7 Qc3+ Re5 Qxe5+ Nxe5 Rh7+ Kf6 Rxa7 c5 Rxa3 Nc6 Be4 Nb4+ Kd2 Ke5 Ke3 Kxd6 Ra8 Bd7 a3 Nc6 Bxc6 Bxc6 Ra6 Kc7 Rxc6+ Kd8 Ke4 Ke7, tb=415894, h=0.0, ph=77.1, 
 wv=33.38, R50=49, Rd=-11, Rr=-1000, mb=+2-1+0+0+0, }  

So LC0 took one move more to "see" the game lost for Black as well as SF did one move before already, in terms of evals.

Because the evals (wv=) always come at the ends of the very long output- lines, I tried to show them better with extra tabulators and double tabs between the plies regards