Engine test position KM FIDE GP 2015

[zullil]

Man, it is not possible for both sides to have advantage at the same time.

If white has advantage above, how can black also have?

My attempt to quote you might have been confusing.

You claimed that White has a huge advantage (even before b4). SF says Black has a tiny advantage after b4 h6.

(I claim the position is a draw with correct play. )

[Lyudmil Tsvetkov]

Man, it is not possible for both sides to have advantage at the same time.

If white has advantage above, how can black also have?

My attempt to quote you might have been confusing.

You claimed that White has a huge advantage (even before b4). SF says Black has a tiny advantage after b4 h6.

(I claim the position is a draw with correct play. )

Not at all, I suppose after Bh4, the best move for black is Bb7, on which white plays f3.

[d]r2q1rk1/pbpn1pp1/1p1ppn1p/8/1PPP3B/P1Q1PP2/4N1PP/R3KB1R b KQ - 0 3

Let SF run longer on this position, white has huge advantage here.

Yakovenko was playing black, meaning he starts with some 15cps disadvantage, then he made the mistake Ba6, another 10-15cps, so white already has 25-30cps edge in the early mg, not very far from a serious advantage. Nakamura did not make any mistakes so far.

You know what the problem is, and not matter how much I would like, I can do nothing for SF?
The problem is, that this position favours white because of phalanx/connecting pawns, that SF does not understand, and also c6/f6 binds, that SF also does not understand. This position favours white in that only way.

But, try to implement that in SF - does not work at 15 sec. I guess such terms would suit SF excellently at TCEC time control, but no one tests at such TC.

So, no matter how I would like, I can do nothing for SF.

[Eelco de Groot]

At depth 34 there is still a slight plus for Black. But I immediately agree with Lyudmil, would never exchange the bishop on f6 unless there was no option. So I don't really understand why, have not checked if it's forced. It isn't as if Stockfish does not know about bishop pairs, there is a pretty large bonus for it in the material imbalance table

Always an excuse to make it even larger is welcome because that looks like better positional knowledge Who is first to try?

[D]r2q1rk1/p1pn1ppp/bp1ppn2/6B1/1PPP4/P1Q1P3/4NPPP/R3KB1R b KQ -

Engine: SfVerification_007E MOD MP (512 MB)
by Tord Romstad, Marco Costalba and Joona Kiiski

34 53:28 +0.24 10...h6 11.Bxf6{?!} Qxf6{!} 12.Ng3 Bb7
13.f3 a5 14.Bd3 Qg5 15.Kf2 f5
16.Rhe1 h5 17.h4 Qe7 18.Bc2 g6
19.Ba4 c6 20.Rh1 e5 21.Ne2 Rfe8
22.Qd2 Qf7 23.Bb3 a4 (8.034.372.588) 2504

34 53:28 +0.01 10...c5 11.dxc5 dxc5 12.Ng3 Qc7
13.Be2 h6 14.Bxf6 Nxf6 15.O-O Bb7
16.Rfd1 Rad8 17.b5 Ne4 18.Nxe4 Bxe4
19.Rxd8 Rxd8 20.a4 f6 21.a5 Qb7
22.f3 Bg6 23.a6 Qe7 (8.034.372.588) 2504

34 53:28 0.00 10...Rc8 11.Ng3 Bb7 12.f3 h6 13.Bxf6 Qxf6
14.Be2 c5 15.Rc1 cxd4 16.Qxd4 Rfd8
17.O-O Qe7 18.Rfd1 Nf8 19.Bd3 d5
20.cxd5 Rxd5 21.Qb2 Rcd8 22.Be2 Qg5
23.f4 Rxd1+ (8.034.372.588) 2504

[Lyudmil Tsvetkov]

At depth 34 there is still a slight plus for Black. But I immediately agree with Lyudmil, would never exchange the bishop on f6 unless there was no option. So I don't really understand why, have not checked if it's forced. It isn't as if Stockfish does not know about bishop pairs, there is a pretty large bonus for it in the material imbalance table

Always an excuse to make it even larger is welcome because that looks like better positional knowledge Who is first to try?

[D]r2q1rk1/p1pn1ppp/bp1ppn2/6B1/1PPP4/P1Q1P3/4NPPP/R3KB1R b KQ -

Engine: SfVerification_007E MOD MP (512 MB)
by Tord Romstad, Marco Costalba and Joona Kiiski

34 53:28 +0.24 10...h6 11.Bxf6{?!} Qxf6{!} 12.Ng3 Bb7
13.f3 a5 14.Bd3 Qg5 15.Kf2 f5
16.Rhe1 h5 17.h4 Qe7 18.Bc2 g6
19.Ba4 c6 20.Rh1 e5 21.Ne2 Rfe8
22.Qd2 Qf7 23.Bb3 a4 (8.034.372.588) 2504

34 53:28 +0.01 10...c5 11.dxc5 dxc5 12.Ng3 Qc7
13.Be2 h6 14.Bxf6 Nxf6 15.O-O Bb7
16.Rfd1 Rad8 17.b5 Ne4 18.Nxe4 Bxe4
19.Rxd8 Rxd8 20.a4 f6 21.a5 Qb7
22.f3 Bg6 23.a6 Qe7 (8.034.372.588) 2504

34 53:28 0.00 10...Rc8 11.Ng3 Bb7 12.f3 h6 13.Bxf6 Qxf6
14.Be2 c5 15.Rc1 cxd4 16.Qxd4 Rfd8
17.O-O Qe7 18.Rfd1 Nf8 19.Bd3 d5
20.cxd5 Rxd5 21.Qb2 Rcd8 22.Be2 Qg5
23.f4 Rxd1+ (8.034.372.588) 2504

This is almost a known theoretical position excluding b4.
White plays Bh4, f3 and then tucks away its bishop on f2, with advantage.

Kasparov used to play a lot similar lines.

But players like Kasparov are rare: neither Nakamura, even less Yakovenko are Kasparov.

Absolutely no sense to try on the framework - nothing works on the framework.

Take piece values - me and Joerg pushed patches giving more bonus to bishops, struggles at STC, does not pass, maybe at long it would have.
Krando pushes a patch on piece values, larger queen value, just the opposite of what should be done, passes green at STC, of course then fails at LTC. At STC simply it is easy to play with the queen.

Many other examples like that - only thing you need to succeed on the framework is luck.

SF 6 was released 4 months ago exactly, if I am not wrong.
Currently, it is at plus 30 elo. If it needs to add another 30 elo or so and be released not much after 8 months pass, as was the case with previous version, SF has another 4 months to add 30 elo. Bearing in mind that 2 of those are summer time...

Anyway, engine development is painful.

[zullil]

But, try to implement that in SF - does not work at 15 sec. I guess such terms would suit SF excellently at TCEC time control, but no one tests at such TC.

So, no matter how I would like, I can do nothing for SF.

Yes, I share your concerns about the current testing system for Stockfish: playing against itself at extremely fast time controls. But that system has resulted in perhaps the strongest engine ever, and how at long time controls do you get enough games for statistically significant assessment?

The fact remains that one can always fork Stockfish and modify it as one sees fit. Perhaps one can create a version that is stronger positionally at standard time controls than the current Stockfish. The catch is---how do you prove that it is, in any reasonable amount of time?

FWIW, I like b4, and I don't see the point of Bxf6 either. But I didn't study things either.

[zullil]

Krando pushes a patch on piece values, larger queen value, just the opposite of what should be done, passes green at STC, of course then fails at LTC.

You know "Krando" is your friend "Lanzo", who apparently has many, many aliases.

[Lyudmil Tsvetkov]

For example, after h6 Bh4 Bb7 f3 SF follows a line like Re8 Bf2 a5 b5 (SF wavers if to play b5, when this is the only obvious good move), e5, and now, SF wants to capture on e5 or play something else, which could still give white some advantage, but d5 instead gives white big advantage:

[d]r2qr1k1/1bpn1pp1/1p1p1n1p/pP1Pp3/2P5/P1Q1PP2/4NBPP/R3KB1R b KQ - 0 6

tremendous bind on c6 here, which even is not a king bind

white is much better here, but my SF still does not understand it; it needs another 5-10 moves/10-20 plies in order to see this, with major help from me.

If it does not see that the above position favours white, how will it see that the initial position favours white, when the above line is probably one of the best main lines for both sides?

And if SF does not see this, how will weaker engines see it?

[Lyudmil Tsvetkov]

we continue the above line a bit: e4 f4 (SF wants to play something else, when f4 is the only obvious good move) Nc5 Nd4 Ng4 Be2 Nf2 Kf2

[d]r2qr1k1/1bp2pp1/1p1p3p/pPnP4/2PNpP2/P1Q1P3/4BKPP/R6R b - - 0 10
another 10 plies have passed, and even here, SF still does not see black is on the verge of losing!

white plays h4 with big winning chances

That is a position where search does not help: no winning captures, no mates - you simply need good eval to guide you to the better moves.

[Eelco de Groot]

For example, after h6 Bh4 Bb7 f3 SF follows a line like Re8 Bf2 a5 b5 (SF wavers if to play b5, when this is the only obvious good move), e5, and now, SF wants to capture on e5 or play something else, which could still give white some advantage, but d5 instead gives white big advantage:

[d]r2qr1k1/1bpn1pp1/1p1p1n1p/pP1Pp3/2P5/P1Q1PP2/4NBPP/R3KB1R b KQ - 0 6

tremendous bind on c6 here, which even is not a king bind

white is much better here, but my SF still does not understand it; it needs another 5-10 moves/10-20 plies in order to see this, with major help from me.

If it does not see that the above position favours white, how will it see that the initial position favours white, when the above line is probably one of the best main lines for both sides?

And if SF does not see this, how will weaker engines see it?

Just to illustrate the main line from Stockfish after 10. b4 h6 11. Bh4 c5 is with 12. dxc5 but after about depth 26 with multiple lines, Stockfish begins to see something different

depth 25:

25 2:21 -0.13 12.dxc5 bxc5 13.f3 Bb7 14.Rd1 Qc7 15.bxc5 Qxc5 16.Qb4 Qxb4+ 17.axb4 Rfc8 18.Bxf6 gxf6 19.Rd4 Ne5 20.Nf4 Kf8 21.b5 Ke7 22.Be2 a6 23.O-O axb5 24.cxb5 Rc2 25.Rb1 (347.553.037) 2463
25 2:21 -0.33 12.f3 cxd4 13.Nxd4 Rc8 14.b5 Bb7 15.Qb2 Qc7 16.Rc1 d5 17.Bg3 Qc5 18.Qb4 dxc4 19.Qxc5 Nxc5 20.Rxc4 Rfe8 21.Rc2 Nd5 22.Kf2 e5 23.Nc6 Bxc6 24.bxc6 Ne7 (347.553.037) 2463
25 2:21 -0.36 12.Rc1 Qc7 13.f3 Rfe8 14.dxc5 bxc5 15.Bxf6 Nxf6 16.Ng3 Bb7 17.Kf2 cxb4 18.axb4 a5 19.b5 a4 20.Qa3 Nd7 21.Be2 Nc5 22.Rhd1 Red8 23.Kg1 Qb6 24.Kf2 (347.553.037) 2463
25 2:21 -0.49 12.b5 Bb7 13.dxc5 bxc5 14.f3 a6 15.a4 axb5 16.axb5 Qb6 17.Rxa8 Rxa8 18.Bxf6 Nxf6 19.Nc1 d5 20.Be2 dxc4 21.Bxc4 Bd5 22.Kf2 Ra4 23.Bd3 Rb4 24.Rd1 Qc7 25.e4 (347.553.037) 2463

now it changes:

26 3:52 -0.10 12.f3 {!} cxd4 13.Qxd4 Rc8 14.Nc3 d5 15.cxd5 Bxf1 16.Kxf1 Qc7 17.dxe6 Qxc3 18.exd7 Nxd7 19.Qxc3 Rxc3 20.e4 Rfc8 21.Bg3 Rb3 22.h4 Nf6 23.Bd6 Rc2 24.Be7 Nh5 25.Kg1 (570.085.102) 2451
26 3:52 -0.20 12.dxc5 bxc5 13.f3 Bb7 14.Rd1 Qc7 15.bxc5 Qxc5 16.Qb4 Qxb4+ 17.axb4 Rfc8 18.Bxf6 gxf6 19.Rd4 Ne5 20.Nf4 Kf8 21.b5 Ke7 22.Be2 a6 23.O-O axb5 24.cxb5 Rc2 25.Rb1 (570.085.102) 2451
26 3:52 -0.26 12.Rc1 Qc7 13.f3 Rfe8 14.Bg3 e5 15.dxc5 bxc5 16.e4 cxb4 17.axb4 Rac8 18.b5 Bb7 19.Bf2 Nb6 20.Ng3 d5 21.c5 dxe4 22.fxe4 Bxe4 23.Nxe4 Nxe4 24.Qe3 Nxf2 25.Kxf2 (570.085.102) 2451
26 3:52 -0.32 12.a4 Bb7 13.a5 e5 14.bxc5 bxc5 15.Bxf6 Nxf6 16.d5 Ba6 17.Ng3 Rb8 18.Be2 Rb4 19.O-O Qc7 20.Qc2 Rfb8 21.Ra2 Rb3 22.Nf5 R8b4 23.Rd1 Qd7 24.Ng3 Bb7 (570.085.102) 2451


and at depth 32 (and a little earlier) White has at least gained equality here. The difference is only 14 centipawns but if you lose 14 centipawns each move, then it can add up quickly... In practice Stockfish just does not play lines like this or if it has to play b4 from book or whatever, misjudges it.

[D]r2q1rk1/p2n1pp1/bp1ppn1p/2p5/1PPP3B/P1Q1P3/4NPPP/R3KB1R w KQ -

Engine: SfVerification_007E MOD MP (512 MB)
by Tord Romstad, Marco Costalba and Joona Kiiski

32 41:38 -0.01 12.f3 b5 13.dxc5 dxc5 14.cxb5 Bxb5
15.Bxf6 Qxf6 16.Qxf6 gxf6 17.Nf4 Rfb8
18.bxc5 Bxf1 19.Rxf1 Nxc5 20.Kf2 Rb2+
21.Kg1 Rb3 22.Rfc1 Nd3 23.Nxd3 Rxd3
24.e4 Rb8 25.Rc7 (6.234.780.761) 2495

32 41:38 -0.15 12.dxc5 bxc5 13.f3 Bb7 14.Rd1 Qc7
15.bxc5 Qxc5 16.Qb4 Qxb4+ 17.axb4 Rfc8
18.Nd4 Ne5 19.Bxf6 gxf6 20.Be2 Nxc4
21.Kf2 a5 22.bxa5 Nxa5 23.Rc1 Bc6
24.Ra1 Bd7 25.Rhc1 (6.234.780.761) 2495

32 41:38 -0.15 12.Rc1 Qc7 13.f3 Rfc8 14.dxc5 bxc5
15.b5 Bb7 16.Qd2 a6 17.Nc3 d5
18.cxd5 exd5 19.bxa6 Bc6 20.Nb5 Qb6
21.Nd6 Rd8 22.Be2 c4 23.e4 Rxa6
24.Bf2 Qc7 25.Bg3 (6.234.780.761) 2495

32 41:38 -0.37 12.a4 cxb4 13.Qxb4 Rc8 14.Nc3 d5
15.Nb5 dxc4 16.Nxa7 Ra8 17.Nc6 Qc7
18.Ne7+ Kh8 19.Be2 Bb7 20.Bg3 Qd8
21.Bf3 Ne4 22.Bxe4 Bxe4 23.f3 Bd3
24.Kf2 Nf6 25.Rhc1 (6.234.780.761) 2495

[Ferdy]

Position after b4 h6 Bh4
[d]r2q1rk1/p1pn1pp1/bp1ppn1p/8/1PPP3B/P1Q1P3/4NPPP/R3KB1R b KQ - 0 11
SF development version:

Code: Select all

Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz
Physical cores: 4, Logical cores: 8

id name Stockfish 210515 64 POPCNT
Hash: 128, Threads: 2, MultiPV: 2, AllocTime: 3.0m

FEN: r2q1rk1/p1pn1pp1/bp1ppn1p/8/1PPP3B/P1Q1P3/4NPPP/R3KB1R b KQ - 0 11

  +0.28/26  0h:02m:53s:675ms   11... c5 12. dxc5 dxc5 13. Rd1 Qe7 14. b5 Bb7 15. f3 Rfd8 16. e4 e5 17. Ng3 Qe6 18. Nf5 Nf8 
  +0.16/26  0h:02m:53s:675ms   11... Bb7 12. f3 a5 13. b5 e5 14. dxe5 dxe5 15. Rd1 Qe7 16. Ng3 Qe6 17. Bxf6 Nxf6 18. c5 Bd5 

  +0.34/25  0h:02m:12s:738ms   11... c5 12. dxc5 bxc5 13. f3 Ne5 14. b5 Bb7 15. a4 a6 16. e4 Ned7 17. Bxf6 Nxf6 18. Ng3 axb5 
  +0.13/25  0h:02m:22s:263ms   11... e5 12. Bxf6 Qxf6 13. d5 Bb7 14. Ng3 a5 15. Be2 axb4 16. axb4 Rxa1+ 17. Qxa1 Ra8 18. Qc1 Qd8 

  +0.37/24  0h:01m:56s:321ms   11... c5 12. dxc5 dxc5 13. Rd1 Qe7 14. b5 Bb7 15. f3 Rfd8 16. e4 e5 17. Ng3 Qe6 18. Nf5 Nf8 
  +0.13/24  0h:02m:05s:992ms   11... e5 12. Bxf6 Qxf6 13. d5 Qg5 14. h4 Qg6 15. a4 Bb7 16. a5 bxa5 17. Rxa5 Rfb8 18. Ng3 c5 

  +0.25/23  0h:01m:45s:404ms   11... c5 12. dxc5 bxc5 13. f3 Bb7 14. Rd1 Qb6 15. bxc5 Qxc5 16. Qb4 Qxb4+ 17. axb4 Rfc8 18. Nc3 Ne8 
  +0.12/23  0h:01m:50s:551ms   11... e5 12. Bxf6 Qxf6 13. d5 Qg5 14. h4 Qg6 15. Ng3 f5 16. Be2 f4 17. Bd3 Qf6 18. Ne4 Qe7 

  +0.35/22  0h:00m:33s:982ms   11... c5 12. dxc5 dxc5 13. Rd1 Qc8 14. b5 Bb7 15. f3 a6 16. a4 axb5 17. axb5 Ra4 18. Qc2 Qa8 
  +0.00/22  0h:01m:16s:988ms   11... e5 12. Bxf6 Qxf6 13. d5 Rac8 14. Qc2 c6 15. dxc6 Rxc6 16. Qa4 Bb7 17. Nc3 Rc7 18. Rd1 Rfc8 

  +0.34/21  0h:00m:28s:280ms   11... Re8 12. Qc2 
  +0.34/21  0h:00m:18s:792ms   11... c5 12. dxc5 bxc5 13. f3 Re8 14. b5 Bb7 15. a4 d5 16. a5 dxc4 17. Ra3 Ne4 18. Bxd8 Nxc3 

Deuterium development version:

Code: Select all

Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz
Physical cores: 4, Logical cores: 8

id name Deuterium v2015.1.35.169
Hash: 128, Threads: 1, MultiPV: 3, AllocTime: 5.0m

FEN: r2q1rk1/p1pn1pp1/bp1ppn1p/8/1PPP3B/P1Q1P3/4NPPP/R3KB1R b KQ - 0 11

  +0.33/19  0h:04m:44s:473ms   11... c5 12. dxc5 bxc5 13. f3 Rc8 14. b5 Bb7 15. O-O-O d5 16. Bxf6 Nxf6 17. Ng3 Qb6 18. a4 Qc7 
  +0.19/19  0h:05m:22s:672ms   11... Rc8 12. b5 Bb7 13. Qb3 c5 14. bxc6 Bxc6 15. Nc3 Qe7 16. f3 e5 17. O-O-O d5 18. Kb2 dxc4 
  +0.05/19  0h:07m:41s:218ms   11... Bb7 12. Qb2 a5 13. b5 e5 14. d5 g5 15. Bg3 Nh5 16. Nc3 Nc5 17. Be2 Nxg3 18. hxg3 Qf6 

  +0.20/18  0h:03m:30s:891ms   11... Rc8 12. Rd1 c6 13. Qb3 d5 14. cxd5 exd5 15. Nc3 Bxf1 16. Kxf1 b5 17. Bxf6 Nxf6 18. Kg1 Re8 
  +0.12/18  0h:02m:34s:745ms   11... Re8 12. Qc2 c5 13. dxc5 bxc5 14. Nc3 Ne5 15. Ne4 g5 16. Bg3 Nxe4 17. Qxe4 f5 18. Qc2 Bb7 
  +0.12/18  0h:02m:22s:176ms   11... Qe7 12. f3 Rfe8 13. Qb3 e5 14. d5 Bb7 15. Nc3 e4 16. f4 a5 17. Ra2 axb4 18. axb4 Rxa2