A historical game

[Lyudmil Tsvetkov]

Here is the second game of the match Kasparov-Fritz from 2003.

[pgn][Event "Man-Machine World Chess Championship"]
[Site "New York Athletic Club, Manhattan"]
[Date "2003.11.13"]
[EventDate "?"]
[Round "2"]
[Result "1-0"]
[White "X3D Fritz (Computer)"]
[Black "Garry Kasparov"]
[ECO "C65"]
[WhiteElo "?"]
[BlackElo "?"]
[PlyCount "77"]

1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. d3 d6 5. c3 g6 6. O-O Bg7
7. Nbd2 O-O 8. Re1 Re8 9. d4 Bd7 10. d5 Ne7 11. Bxd7 Nxd7
12. a4 h6 13. a5 a6 14. b4 f5 15. c4 Nf6 16. Bb2 Qd7 17. Rb1
g5 18. exf5 Qxf5 19. Nf1 Qh7 20. N3d2 Nf5 21. Ne4 Nxe4
22. Rxe4 h5 23. Qd3 Rf8 24. Rbe1 Rf7 25. R1e2 g4 26. Qb3 Raf8
27. c5 Qg6 28. cxd6 cxd6 29. b5 axb5 30. Qxb5 Bh6 31. Qb6 Kh7
32. Qb4 Rg7 33. Rxe5 dxe5 34. Qxf8 Nd4 35. Bxd4 exd4 36. Re8
Rg8 37. Qe7+ Rg7 38. Qd8 Rg8 39. Qd7+ 1-0


[/pgn]

I must say that all of Kasparov's matches against the machines are extremely interesting. Kasparov's style is enterprising, and at the same time he frequently employs anti-computer strategies. In comparison to Kasparov-Deep Blue from 1996 and 1997, as well as Kasparov-Fritz, Kramnik-Fritz is boring, Adams-Hydra is boring, Nakamura-Stockfish is boring.

So that to anyone interested in computer-human games I recommend to study very well all of Kasparov's games in those matches.

[d]r2qr1k1/1ppnn1b1/p2p2pp/P2Ppp2/1PP1P3/5N2/3N1PPP/R1BQR1K1 b - - 0 15

This is a critical position in the game. Kasparov played 15...Nf6, later commited a couple of inaccuracies, a final mistake, and lost.

Still, here black is considerably better, and Kasparov could have won by adopting the appropriate strategy on time.

How many engines see the winning move for black?

PS. So that I do not get accused of bias, I must say that I am not 100% certain the position is won for black. All I know is black is better in terms of eval at the root by maybe some 30cps. However, my assessment might be imperfect, and those 30cps might actually be just 10cps, meaning a draw, or 50cps, already meaning a win for black in a rich middlegame. There is no doubt at that black has an observable advantage above.

[yanquis1972]

i would be shocked to see any engine evaluate this as anything but adv. white; even an evaluation under one decipawn would surprise me quite a bit.

komodo is more conservative than i have seen some engines in the past but clearly favors white by a quarter of a pawn with more concern about neutralizing whites space advantage on the QS than generating counterplay on the kingside. this is sadly typical of, as far as i know, all engines in these KID-type positions. i don't expect much anythign to change with greater depth here.

Analysis by Komodo 8 64-bit:

15...Nf6 16.Bb2
= (0.19) Depth: 7 00:00:00 44kN
+/= (0.34 --) Depth: 10 00:00:00 291kN
15...c6 16.c5
= (0.27) Depth: 10 00:00:00 517kN
+/= (0.39 --) Depth: 11 00:00:00 581kN
15...Nf6 16.Bb2 Qd7 17.Qb3 g5 18.Qc2 c6 19.dxc6 Nxc6 20.exf5 g4 21.Nh4 Nxb4 22.Qb3 Nc6 23.Ne4 Nxe4
+/= (0.32) Depth: 11 00:00:00 693kN
+/= (0.39) Depth: 14 00:00:00 4075kN
15...c6 16.dxc6 bxc6 17.Bb2 Rb8 18.Qa4 Qc7 19.Rad1 Bf6 20.h3 Rf8 21.Qb3 Kh7 22.Bc3 Rg8 23.Qb2 Bg7 24.Qa3 Bf6 25.Qa1 Bg7 26.Kh1 Qa7 27.Re3 Nf6 28.c5 dxc5 29.Bxe5 Nxe4 30.Nxe4
= (0.22 ++) Depth: 14 00:00:01 4869kN
= (0.26) Depth: 26 00:03:50 2529MN, tb=2

a similar-length search by a recent stockfish build, though i imagine louis will be able to post much more in depth analysis

New game
r2qr1k1/1ppnn1b1/p2p2pp/P2Ppp2/1PP1P3/5N2/3N1PPP/R1BQR1K1 b - - 0 1

Analysis by Stockfish 171114 64 BMI2 TB:

15...Nf6 16.Bb2 Qd7 17.exf5 Nxf5 18.Ra3 Rf8 19.Rb3 Rae8 20.Ne4 Nxe4 21.Rxe4 Ne7 22.Re1 Nf5 23.Qd3 Qf7 24.b5 axb5 25.Rxb5 Ra8 26.Bc3 Ra7 27.Re2 Bf6 28.Rb1 Rfa8 29.Ree1 Rf8 30.h3 Qg7 31.Qe4 h5 32.c5 Qf7
+/= (0.50) Depth: 7/7 00:00:00 8kN
+/= (0.38) Depth: 35/48 00:04:05 2781MN

[Lyudmil Tsvetkov]

i would be shocked to see any engine evaluate this as anything but adv. white; even an evaluation under one decipawn would surprise me quite a bit.

komodo is more conservative than i have seen some engines in the past but clearly favors white by a quarter of a pawn with more concern about neutralizing whites space advantage on the QS than generating counterplay on the kingside. this is sadly typical of, as far as i know, all engines in these KID-type positions. i don't expect much anythign to change with greater depth here.

Analysis by Komodo 8 64-bit:

15...Nf6 16.Bb2
= (0.19) Depth: 7 00:00:00 44kN
+/= (0.34 --) Depth: 10 00:00:00 291kN
15...c6 16.c5
= (0.27) Depth: 10 00:00:00 517kN
+/= (0.39 --) Depth: 11 00:00:00 581kN
15...Nf6 16.Bb2 Qd7 17.Qb3 g5 18.Qc2 c6 19.dxc6 Nxc6 20.exf5 g4 21.Nh4 Nxb4 22.Qb3 Nc6 23.Ne4 Nxe4
+/= (0.32) Depth: 11 00:00:00 693kN
+/= (0.39) Depth: 14 00:00:00 4075kN
15...c6 16.dxc6 bxc6 17.Bb2 Rb8 18.Qa4 Qc7 19.Rad1 Bf6 20.h3 Rf8 21.Qb3 Kh7 22.Bc3 Rg8 23.Qb2 Bg7 24.Qa3 Bf6 25.Qa1 Bg7 26.Kh1 Qa7 27.Re3 Nf6 28.c5 dxc5 29.Bxe5 Nxe4 30.Nxe4
= (0.22 ++) Depth: 14 00:00:01 4869kN
= (0.26) Depth: 26 00:03:50 2529MN, tb=2

a similar-length search by a recent stockfish build, though i imagine louis will be able to post much more in depth analysis

New game
r2qr1k1/1ppnn1b1/p2p2pp/P2Ppp2/1PP1P3/5N2/3N1PPP/R1BQR1K1 b - - 0 1

Analysis by Stockfish 171114 64 BMI2 TB:

15...Nf6 16.Bb2 Qd7 17.exf5 Nxf5 18.Ra3 Rf8 19.Rb3 Rae8 20.Ne4 Nxe4 21.Rxe4 Ne7 22.Re1 Nf5 23.Qd3 Qf7 24.b5 axb5 25.Rxb5 Ra8 26.Bc3 Ra7 27.Re2 Bf6 28.Rb1 Rfa8 29.Ree1 Rf8 30.h3 Qg7 31.Qe4 h5 32.c5 Qf7
+/= (0.50) Depth: 7/7 00:00:00 8kN
+/= (0.38) Depth: 35/48 00:04:05 2781MN

Both engines seem to be about the level of Kasparov: choosing the same line.

I forgot to mention that the winning move is, of course, 15...f4.

[yanquis1972]

interesting choice of best move, was i do not see anything wrong with nf6 myself. but computers (at least stockfish, the only engine ive tried yet) really leap at 15...f4, so perhaps this position is of even more interest....i can't help but wonder if this is another case of a human player underestimating the resources available to the attacked side.



[d]r2qr1k1/1ppnn1b1/p2p2pp/P2Pp3/1PP1Pp2/5N2/3N1PPP/R1BQR1K1 w - - 0 1

Analysis by Stockfish 171114 64 BMI2:

16.Qb3 g5 17.h3 Nc8 18.Ba3 Nf8 19.Nh2 Nd7 20.Qd1 Qe7
+/- (0.83) Depth: 7/7 00:00:00 1kN
+/- (0.83) Depth: 10/10 00:00:00 3kN
16.Bb2 g5 17.c5 g4 18.Nh4 h5 19.Nc4 Qb8 20.Rc1 Bf6 21.Nf5 Nxf5 22.exf5
+/- (0.83) Depth: 11/12 00:00:00 7kN
+/- (0.83) Depth: 13/13 00:00:00 17kN
16.Qb3 g5
+/- (0.83) Depth: 14/17 00:00:00 158kN
+/- (0.84 --) Depth: 31/43 00:00:22 246MN
16.Ba3 Nc8 17.Rc1 Na7 18.h3 h5 19.Bb2 g5 20.Rc3 Rc8 21.Nh2 Nf6 22.Rb3 c6 23.g4 hxg4 24.Nxg4 Qd7 25.Kg2 Nxg4 26.Qxg4 Qxg4+ 27.hxg4 Kf7 28.Rh3 Rh8 29.Rhh1 Kg6 30.f3 Bf6 31.Rc1 Rh4 32.Rxh4 gxh4
+/- (0.90 ++) Depth: 31/47 00:01:22 938MN
+/- (0.81) Depth: 31/47 00:05:57 4167MN
16.h3 Nc8 17.Bb2 Qe7 18.Rc1 h5 19.Qe2 Na7 20.Rc3 Rac8 21.Rec1 c6 22.R3c2 Bf6 23.Nb3 Rc7 24.Nfd2 Rec8 25.Bc3 Re8 26.Nf3 Qf7 27.Nbd2 Rcc8 28.Ne1 Rc7 29.f3 Nc8 30.Nd3 cxd5 31.exd5 Ne7 32.Ne4
+/- (0.87 ++) Depth: 32/47 00:07:13 5040MN
+/- (0.90) Depth: 32/47 00:13:32 9383MN




here in fact, stockfish even seems to suggest that white is ahead on the attack, & that it is black who will have to defend. (edit-- i see now it takes 7+mins on my machine for SF to switch to a defensive minded move, 16. h3. the final posted eval is the one resolved after completing depth 32)

a very curious position; if no one else does the honor, i certainly wouldn't mind playing out whites next few moves with an engine (either komodo or SF). i think an an unaided ~10 minute search, provided you are only using engines to blundercheck, could be interesting.

[Lyudmil Tsvetkov]

interesting choice of best move, was i do not see anything wrong with nf6 myself. but computers (at least stockfish, the only engine ive tried yet) really leap at 15...f4, so perhaps this position is of even more interest....i can't help but wonder if this is another case of a human player underestimating the resources available to the attacked side.



[d]r2qr1k1/1ppnn1b1/p2p2pp/P2Pp3/1PP1Pp2/5N2/3N1PPP/R1BQR1K1 w - - 0 1

Analysis by Stockfish 171114 64 BMI2:

16.Qb3 g5 17.h3 Nc8 18.Ba3 Nf8 19.Nh2 Nd7 20.Qd1 Qe7
+/- (0.83) Depth: 7/7 00:00:00 1kN
+/- (0.83) Depth: 10/10 00:00:00 3kN
16.Bb2 g5 17.c5 g4 18.Nh4 h5 19.Nc4 Qb8 20.Rc1 Bf6 21.Nf5 Nxf5 22.exf5
+/- (0.83) Depth: 11/12 00:00:00 7kN
+/- (0.83) Depth: 13/13 00:00:00 17kN
16.Qb3 g5
+/- (0.83) Depth: 14/17 00:00:00 158kN
+/- (0.84 --) Depth: 31/43 00:00:22 246MN
16.Ba3 Nc8 17.Rc1 Na7 18.h3 h5 19.Bb2 g5 20.Rc3 Rc8 21.Nh2 Nf6 22.Rb3 c6 23.g4 hxg4 24.Nxg4 Qd7 25.Kg2 Nxg4 26.Qxg4 Qxg4+ 27.hxg4 Kf7 28.Rh3 Rh8 29.Rhh1 Kg6 30.f3 Bf6 31.Rc1 Rh4 32.Rxh4 gxh4
+/- (0.90 ++) Depth: 31/47 00:01:22 938MN
+/- (0.81) Depth: 31/47 00:05:57 4167MN
16.h3 Nc8 17.Bb2 Qe7 18.Rc1 h5 19.Qe2 Na7 20.Rc3 Rac8 21.Rec1 c6 22.R3c2 Bf6 23.Nb3 Rc7 24.Nfd2 Rec8 25.Bc3 Re8 26.Nf3 Qf7 27.Nbd2 Rcc8 28.Ne1 Rc7 29.f3 Nc8 30.Nd3 cxd5 31.exd5 Ne7 32.Ne4
+/- (0.87 ++) Depth: 32/47 00:07:13 5040MN
+/- (0.90) Depth: 32/47 00:13:32 9383MN




here in fact, stockfish even seems to suggest that white is ahead on the attack, & that it is black who will have to defend. (edit-- i see now it takes 7+mins on my machine for SF to switch to a defensive minded move, 16. h3. the final posted eval is the one resolved after completing depth 32)

a very curious position; if no one else does the honor, i certainly wouldn't mind playing out whites next few moves with an engine (either komodo or SF). i think an an unaided ~10 minute search, provided you are only using engines to blundercheck, could be interesting.

This is another case of a human player having a better grasp of the position.

SF does not see the right black moves, it misses at least a couple of vital moves, after which black gets better. The position is simply too deep for it.

Let's wait and see if someone else contributes. If Louis does not want, then I will be happy to play a game with you.

For the time being the clue is that Kh8 and Rg8 are important moves of the black plan.

[carldaman]

Helping the attacked side (White) is the lack of a light-squared Black Bishop (often critical in KID-type setups) and the available Kf1-e2 escape route for the King. Piece exchanges tend to ease White's defensive concerns in such positions.

Regards,
CL

interesting choice of best move, was i do not see anything wrong with nf6 myself. but computers (at least stockfish, the only engine ive tried yet) really leap at 15...f4, so perhaps this position is of even more interest....i can't help but wonder if this is another case of a human player underestimating the resources available to the attacked side.



[d]r2qr1k1/1ppnn1b1/p2p2pp/P2Pp3/1PP1Pp2/5N2/3N1PPP/R1BQR1K1 w - - 0 1

Analysis by Stockfish 171114 64 BMI2:

16.Qb3 g5 17.h3 Nc8 18.Ba3 Nf8 19.Nh2 Nd7 20.Qd1 Qe7
+/- (0.83) Depth: 7/7 00:00:00 1kN
+/- (0.83) Depth: 10/10 00:00:00 3kN
16.Bb2 g5 17.c5 g4 18.Nh4 h5 19.Nc4 Qb8 20.Rc1 Bf6 21.Nf5 Nxf5 22.exf5
+/- (0.83) Depth: 11/12 00:00:00 7kN
+/- (0.83) Depth: 13/13 00:00:00 17kN
16.Qb3 g5
+/- (0.83) Depth: 14/17 00:00:00 158kN
+/- (0.84 --) Depth: 31/43 00:00:22 246MN
16.Ba3 Nc8 17.Rc1 Na7 18.h3 h5 19.Bb2 g5 20.Rc3 Rc8 21.Nh2 Nf6 22.Rb3 c6 23.g4 hxg4 24.Nxg4 Qd7 25.Kg2 Nxg4 26.Qxg4 Qxg4+ 27.hxg4 Kf7 28.Rh3 Rh8 29.Rhh1 Kg6 30.f3 Bf6 31.Rc1 Rh4 32.Rxh4 gxh4
+/- (0.90 ++) Depth: 31/47 00:01:22 938MN
+/- (0.81) Depth: 31/47 00:05:57 4167MN
16.h3 Nc8 17.Bb2 Qe7 18.Rc1 h5 19.Qe2 Na7 20.Rc3 Rac8 21.Rec1 c6 22.R3c2 Bf6 23.Nb3 Rc7 24.Nfd2 Rec8 25.Bc3 Re8 26.Nf3 Qf7 27.Nbd2 Rcc8 28.Ne1 Rc7 29.f3 Nc8 30.Nd3 cxd5 31.exd5 Ne7 32.Ne4
+/- (0.87 ++) Depth: 32/47 00:07:13 5040MN
+/- (0.90) Depth: 32/47 00:13:32 9383MN




here in fact, stockfish even seems to suggest that white is ahead on the attack, & that it is black who will have to defend. (edit-- i see now it takes 7+mins on my machine for SF to switch to a defensive minded move, 16. h3. the final posted eval is the one resolved after completing depth 32)

a very curious position; if no one else does the honor, i certainly wouldn't mind playing out whites next few moves with an engine (either komodo or SF). i think an an unaided ~10 minute search, provided you are only using engines to blundercheck, could be interesting.

[peter]

For the time being the clue is that Kh8 and Rg8 are important moves of the black plan.

So you'd play Kh8 instantly after e.g. 15...f4 16.h3 too?

[Lyudmil Tsvetkov]

Helping the attacked side (White) is the lack of a light-squared Black Bishop (often critical in KID-type setups) and the available Kf1-e2 escape route for the King. Piece exchanges tend to ease White's defensive concerns in such positions.

Regards,
CL

interesting choice of best move, was i do not see anything wrong with nf6 myself. but computers (at least stockfish, the only engine ive tried yet) really leap at 15...f4, so perhaps this position is of even more interest....i can't help but wonder if this is another case of a human player underestimating the resources available to the attacked side.



[d]r2qr1k1/1ppnn1b1/p2p2pp/P2Pp3/1PP1Pp2/5N2/3N1PPP/R1BQR1K1 w - - 0 1

Analysis by Stockfish 171114 64 BMI2:

16.Qb3 g5 17.h3 Nc8 18.Ba3 Nf8 19.Nh2 Nd7 20.Qd1 Qe7
+/- (0.83) Depth: 7/7 00:00:00 1kN
+/- (0.83) Depth: 10/10 00:00:00 3kN
16.Bb2 g5 17.c5 g4 18.Nh4 h5 19.Nc4 Qb8 20.Rc1 Bf6 21.Nf5 Nxf5 22.exf5
+/- (0.83) Depth: 11/12 00:00:00 7kN
+/- (0.83) Depth: 13/13 00:00:00 17kN
16.Qb3 g5
+/- (0.83) Depth: 14/17 00:00:00 158kN
+/- (0.84 --) Depth: 31/43 00:00:22 246MN
16.Ba3 Nc8 17.Rc1 Na7 18.h3 h5 19.Bb2 g5 20.Rc3 Rc8 21.Nh2 Nf6 22.Rb3 c6 23.g4 hxg4 24.Nxg4 Qd7 25.Kg2 Nxg4 26.Qxg4 Qxg4+ 27.hxg4 Kf7 28.Rh3 Rh8 29.Rhh1 Kg6 30.f3 Bf6 31.Rc1 Rh4 32.Rxh4 gxh4
+/- (0.90 ++) Depth: 31/47 00:01:22 938MN
+/- (0.81) Depth: 31/47 00:05:57 4167MN
16.h3 Nc8 17.Bb2 Qe7 18.Rc1 h5 19.Qe2 Na7 20.Rc3 Rac8 21.Rec1 c6 22.R3c2 Bf6 23.Nb3 Rc7 24.Nfd2 Rec8 25.Bc3 Re8 26.Nf3 Qf7 27.Nbd2 Rcc8 28.Ne1 Rc7 29.f3 Nc8 30.Nd3 cxd5 31.exd5 Ne7 32.Ne4
+/- (0.87 ++) Depth: 32/47 00:07:13 5040MN
+/- (0.90) Depth: 32/47 00:13:32 9383MN




here in fact, stockfish even seems to suggest that white is ahead on the attack, & that it is black who will have to defend. (edit-- i see now it takes 7+mins on my machine for SF to switch to a defensive minded move, 16. h3. the final posted eval is the one resolved after completing depth 32)

a very curious position; if no one else does the honor, i certainly wouldn't mind playing out whites next few moves with an engine (either komodo or SF). i think an an unaided ~10 minute search, provided you are only using engines to blundercheck, could be interesting.

How are you doing, Carl?

I am afraid here the white king going to the center would be an impossible enterprise, as only d and e files are fully closed by blocked pawns, while the queen side is pretty flexible, meaning black can attack also through that side.

The lack of a light-squared bishop indeed looks to be extremely favourable to white, but only on the surface. In the particular position this is not felt at all, as the bishop of white is belittled behind the longer central blocked chain, while the bishop of black's alleged weakness of colour is made irrelevant precisely because of the same factor. As you rightly point out, exchanges would ease white's defence, but the blocked central chain makes even those very difficult.

In a word, the chain acts like a magical wand, transforming black's weaknesses to strengths and nullifying white's strong points.

[PK]

I think 15...f4 and 15...Nf6 are about equal. Black's setup is far from ideal of a KID attacking setup, whereas White has already made some progress on queenside. At first I thought that Nb3 is the key move (simultaneously preparing Nfd2 and c5), but then Black has just enough time to get g5-g4 at about the same time White gest c5xd6.

15...f4 16.Nb3 g5 16.h3 Ng6 17.Nfd2 h5 18.f3 Nf6 19.c5 and now either 19...Qd7 or 19...g4!?

Still, 15...f4 in this piece setup feels a bit premature. It takes the pressure off White's pawn chain, thus giving White free hand on the queenside. However the line that I have just given shows that normal moves are not enough to prove it. That leaves only

15...f4 16.c5!

Black really cannot take the pawn:

16... dxc5? 17. bxc5 Nxc5 18. Ba3 Nd7 19. Qb3

So the correct line might be something like:

15...f4 16.c5! Nf6 17.Nc4 g5 18.h3 Ng6 when White at has interesting options of 19. Qb3 and of delaying 19. Nh2

[Lyudmil Tsvetkov]

For the time being the clue is that Kh8 and Rg8 are important moves of the black plan.

So you'd play Kh8 instantly after e.g. 15...f4 16.h3 too?

Not instantly. First I would include the moves f4, g5, and Nc8.

Here is a tentative game played where me + SF have the better of SF left without any help.

[pgn][PlyCount "55"]
[Event "Blitz 2m+2s"]
[Site "Sofia"]
[Date "2014.11.20"]
[White "Stockfish 5 64 SSE4.2"]
[Black "L Tsvetkov + Stockfish, Lyudmil"]
[Result "*"]
[TimeControl "120+2"]
[SetUp "1"]
[FEN "r2qr1k1/1ppnn1b1/p2p2pp/P2Ppp2/1PP1P3/5N2/3N1PPP/R1BQR1K1 b - - 0 1"]
[Annotator "Tsvetkov,Lyudmil"]
[MLNrOfMoves "27"]
[MLFlags "000100"]

{192MB, Dell XPS 4Cores} 1... f4 {[%emt 0:00:00]} 2. Qb3 {[%emt 0:00:05]} 2...
g5 {[%emt 0:00:01]} 3. h3 {[%emt 0:00:03]} 3... Nc8 {[%emt 0:00:03]} 4. Bb2
{[%emt 0:00: 11]} 4... Kh8 {[%emt 0:00:02]} 5. Rec1 {[%emt 0:00:13]} 5... Rg8
{[%emt 0:00:03]} 6. Qd1 {[%emt 0:00:07]} 6... Qe8 {[%emt 0:00:00]} 7. Ne1
{[%emt 0:00:06]} 7... h5 {[%emt 0:00: 19]} 8. Nd3 {[%emt 0:00:12]} 8... Nf6
{[%emt 0:00:13]} 9. f3 {[%emt 0:00:04]} 9... Bh6 {[%emt 0:00:03]} 10. Qe1
{[%emt 0:00:07]} 10... Qg6 {[%emt 0:00:08]} 11. Kh1 {[%emt 0: 00:11]} 11... g4
{[%emt 0:00:04]} 12. Qf2 {[%emt 0:00:07]} 12... Ne7 {[%emt 0:00:03]} 13. Rg1
{[%emt 0:00:04]} 13... Rg7 {[%emt 0:00:07]} 14. c5 {[%emt 0:00:04]} 14... Rag8
{[%emt 0:00:03]} 15. Qf1 {[%emt 0:00:07]} 15... Nd7 {[%emt 0:00:12]} 16. Rc1
{[%emt 0:00:09]} 16... Qf6 {[%emt 0:00:04]} 17. Nf2 {[%emt 0:00:04]} 17... gxh3
{[%emt 0:00:04]} 18. Nxh3 {[%emt 0:00:05]} 18... Ng6 {[%emt 0:00:03]} 19. c6
{[%emt 0:00:04]} 19... bxc6 {[%emt 0:00: 03]} 20. Rxc6 {[%emt 0:00:09]} 20...
Nh4 {[%emt 0:00:03]} 21. Rxc7 {[%emt 0:00:03]} 21... Nxg2 {[%emt 0:00:04]} 22.
Rxg2 {[%emt 0:00:02]} 22... Rxg2 {[%emt 0:00:03]} 23. Qxg2 {[%emt 0:00:03]}
23... Rxg2 {[%emt 0:00:08]} 24. Kxg2 {[%emt 0:00:03]} 24... Qg7+
{[%emt 0: 00:08]} 25. Kf1 {[%emt 0:00:03]} 25... Qg3 {[%emt 0:00:03]} 26. Nf2
{[%emt 0:00:03]} 26... Nf6 {[%emt 0:00:13]} 27. Rc8+ {[%emt 0:00:11]} 27... Kg7
{[%emt 0:00:03]} 28. Rc7+ {[%emt 0:00:04]} 28... Kf8 {[%emt 0:00:03]} *
[/pgn]

Again, I played for black the first 5 or 6 moves, and then let SF alone, as it already started seeing black advantage. As you see, black has great chances.

[d]r1n1q1rk/1ppn2b1/p2p4/P2Pp1pp/1PP1Pp2/7P/1B1N1PP1/R1RQN1K1 w - - 0 8
I leave SF here, and then it plays perfectly on its own.

As a rule, engines recognize good positions and play acceptable chess: only thing you should do is execute the first couple of moves for them.