Evaluating Evaluation

Websites, such as LiChess and Chess.com, advertise the benefits of nearly instant computer analysis following a game. This analysis is provided by a resident version of Stockfish, usually the newest edition. While I make liberal use of this analysis when examining my suspicions concerning cheating (suspicions that are nearly always unfounded--see "Playing Online [Pros and Cons]"), my postgame analysis usually begins sans engine, then I turn on an engine to check my analysis. However, yesterday I played two 30 + 10 games, much longer than my usual live play online. I used the instant computer evaluation to count the inaccuracies (three in 46 moves!). Today, I played one game, the computer evaluation was less encouraging.

What follows is the game, as annotated by Stockfish on LiChess. I added some annotations of my own, and also checked Stockfish's with Komodo's. Stockfish's evaluations must be treated with some skepticism aimed at understanding the reasons. I left in the move-by-move computer score.

Internet Opponent (1831) -- Stripes,J (2112) [A45]
Rated Classical game lichess.org, 02.08.2020

1.d4 [0.02/010] Nf6 [A45 Indian Game 0.27/010] 2.c3 [0.00/02] g6 [0.09/05] 3.Nd2 [0.01/08] Bg7 [0.34/04]

3...c5 4.e3 cxd4 5.exd4 Bg7 6.Ngf3 0–0 7.Bc4 d5 8.Bd3 Bf5 9.Bxf5 gxf5 10.0–0 e6 Perhaps Black expected to mount an attack along the half open g-file. It did not happen. White won in 42 moves, Timoshenko,I (2266) -- Erendzhenov,S (2280), Nevinnomyssk 2010.

4.e3 [0.14/01] d6 [0.29/050] 5.Be2?! [0.29/04]

5.Ngf3 was best, according to Stockfish on LiChess 5...b6 6.Bb5+ c6 7.Bd3 0–0 8.0–0 c5 9.e4 Bb7;
5.f4 was a move I expected with a sort of reversed Dutch Stonewall.;
5.Bd3 is probably a better square for the bishop.

5...e5 [0.14/017] 6.h4 [0.05/06]

Black to move

6...h5?! [0.16/011]

6...Nc6 was best, according to Stockfish, allowing what I intended to prevent. One inaccuracy (debatable). 7.h5.
6...Nbd7 is Komodo's choice. 7.h5 Nxh5 8.Bxh5 gxh5 9.Qxh5 Qf6 Black is slightly better.

7.Nb3? [-0.81/017]

7.dxe5 was best (Stockfish) 7...dxe5 8.Ngf3 Nbd7 9.Qc2 0–0 10.Ng5 Qe7 11.0–0 a5 12.Rd1 b6 13.a4 Bb7

7...Be6?? [0.61/032]

7...e4 was a move I considered, and should have been played, according to Stockfish. It is not among Komodo's top three choices. My blunder allows White to open up the board and deprive me of castling. The advantage shifts from Black to White.

8.f3? [0.82/011]

8.dxe5 dxe5 9.Qxd8+ Kxd8 10.Nf3 Ne4 11.Nbd2 Nxd2 12.Ng5 Ne4 13.Nxe4 f6 14.b3 Nd7

8...e4?!

Second "inaccuracy", according to Stockfish. One blunder. Komodo likes my move. 0.31/028
8...Nbd7 9.e4 c6 10.Nh3 Bxh3 11.Rxh3 a5 12.Bg5 a4 13.Nd2 b5 14.f4 Qb6 15.fxe5

9.f4?? [-1.74/020]

9.Qc2 d5

9...0–0 [-0.13/052] 10.Nd2 [-1.58/018] d5 [-1.62/03] 11.Nh3 [-1.34/05] Ng4 [-1.43/013] 12.Nf1 [-2.89/018] 

Black to move

12...Nh6? [-0.57/01:16] 

12...Qxh4+ is a move that I should not have missed. 25% of my moves so far have been errors of one sort or another, although I disagree with the computer's assessment in one case. Komodo disagrees with Stockfish in the other case.

13.Ng5 [-0.96/013] Bg4 [-0.52/02] 14.g3?? [-2.44/013]

14.Ng3

14...c5?! [-1.74/030]

14...Bxe2 is part of a better sequence (Stockfish). The change from -2.44 to –1.74 is more severe than Komodo's evaluation: -2.01 to –1.68. It is still an inaccuracy. 15.Qxe2 Nd7 (15...c5 could be played now, according to Komodo) 16.Bd2 c5 17.0–0–0 Qb6 18.Kb1 Rac8 19.Nh2 cxd4 20.exd4 Nf5 21.Be1

15.Nh2?! [-2.44/026]

15.Bxg4 (Stockfish's suggestion) was my hope 15...Nxg4 with pressure on e3, but perhaps that pressure is less useful than my other idea in the position: opening the c-file.

15...Bxe2 [-2.59/019] 16.Qxe2 [-2.94/036] Nd7 [-2.69/03] 17.g4 [-2.56/08] Nf6 [-2.85/07] 18.gxh5 [-2.97/012]

18.f5 Komodo's line--Black is still winning. 18...hxg4 19.fxg6 fxg6

18...Nxh5 [-2.54/00] 19.Rg1 [-2.27/08]

Black to move

19...Qd7? [-0.91/018]

19...cxd4 was best (Stockfish). I wanted to prevent Ne6 if I played f6. I thought perhaps I could trap the knight. Also, connecting the rooks seemed like a good idea, as both might come to the c-file as matters developed further. 20.exd4 Qa5 21.Ng4 Nf5 22.Ne5 (an error, according to Komodo, preferring 22.Bd2) 22...Nxd4 23.Qd1 Nf5 24.Ngxf7 Qc5 25.Qg4 Rxf7 26.Qxg6 may seem favorable from Black's perspective to the chess engine, but I prefer to avoid such a position.

Black to move

Analysis Diagram

20.Qg2?? [-3.16/020]

20.Ng4 (Stockfish) 20...Nf5 was my intent. White's heavy pieces along the opened g-file do not offer the prospects for attack that my opponent imagined.

20...Nf5?! [-2.37/016]

20...cxd4 was best (Stockfish). The number of my moves judged errors now exceeds 25%. 21.exd4

21.Bd2? [-5.98/019]

21.dxc5 Rac8

21...cxd4 [-4.36/016] 22.cxd4 [-4.27/016] Rac8 [-4.97/05] 23.b3?! [-7.61/023]

23.Nxf7 was best (Stockfish). Nonetheless, the computer's line gives Black an easy game with a clear advantage. The text move also gave me an easy game with some tactical threats. 23...Qxf7 24.Qxg6 Qxg6 25.Rxg6 Rc2 26.Rg5 Nhg3 27.Rc1 Rxb2 28.Kd1 Kh8 29.Ng4 Rxa2

23...Rc2 [-5.63/02:26]

White to move

24.Kd1?! [-10.96/013]

24.Ng4 Stockfish sees the threats to the e-pawn with the second rank pin. I disagree with Stockfish that 24.Kd1 is an inaccuracy. It was a blunder. 24...Nxf4

24...Nxe3+[-11.98/012]

Wins the queen

25.Bxe3 [-10.27/07] Rxg2 [-10.84/02] 26.Rxg2 [-10.84/03] f6[-9.36/026]

I also win the knight (thanks to an earlier "mistake") 0–1