Monday, April 26, 2021

Beating max level Stockfish with the Jerome Gambit (Part 1)

 


I was immediately interested in a title on Reddit: 

Beating max level Stockfish with the Bongcloud and the Jerome gambit

Seriously?

I was curious how strong lvl 8 Stockfish is on Lichess compared to a real engine running on a decent computer so i played some games where i essentially copied moves from Stockfish13. Here's the games:

...Win with the Jerome gambit: https://lichess.org/wqUokQSZ/white (the game had to be set up from position since stockfish prefers 3.Ng5 [3...Nf6] over 3.Bc5 [3...Bc5]) 

Since lichess.org was using Stockfish 13+ NNUE (Efficiently Updatable Neural Network) for its analysis, it looked like the game was a Stockfish vs Stockfish affair, but it turned out to be quite interesting, if a bit long... 

I have adjusted the move numbers to reflect a game that started with 1.e4.

This game should be titled "Beware the Jerome Pawns!"


razornfs - lichess AI level 8

15 10 standard, lichess.org, 2021


1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.Bxf7+


4...Kxf7 5.Nxe5+ Nxe5 6.Qh5+ Kf8


It is interesting that, in annotating the game, the lichess engine marked its own move here as an "Inaccuracy", preferring 6...Ke6. Probably 6...Kf8, a completely sound and reliable move, was in the engine's "book".

7.Qxe5 d6 8.Qg3 Nf6 9.Nc3 Nh5 


Yeow. I checked in The Database and found three previous Jerome Gambit games with this move, two by Bill Wall and one by 2200 rated Chess-for-All, all losses for White.

10.Qf3+ Qf6 11.d3 

White agrees to the exchange of Queens and the breakup of his Kingside pawns. Previously

11.Qxf6+ Nxf6 12.d3 Be6 13.f3 a6 14. Ke2 d5 15.a4 Kf7 16.a5 Rac8 17.e5 Nd7 18.f4 g6 19.Kf3 h6 20.Ne2 Be7 21.Be3 c5 22.Rhe1 Nb8 23.c3 Nc6 24.h3 25.cxd4 cxd4 26.Nxd4 Nxd4+ 27.Bxd4 Rhd8 28.Ke3 Rc2 29.Bb6 Rdc8 30.g4 Bd5 31.Rf1 Rxb2 32.Kd4 Bg2 33.Rab1 Rxb1 34.Rxb1 Rc2 35.f5 Bh4 36.e6+ Ke7 37.fxg6 Kxe6 38.Ke3 Ke5 39.Bd4+ Kd5 White resiged,  Wall,B - Guest1780185, PlayChess.com, 2019; and

11.Qxh5 Qxf2+ 12.Kd1 Qxg2 White resigned, Chess-For-All-Makabe, GameKnot.com, 2017

11...Qxf3 12.gxf3 Be6 

Also 12...Bd7 13.Ke2 Bd4 14.Be3 Bf6 15.Nd5 Bd8 16.Rhg1 Kf7 17.f4 c6 18.Nc3 g6 19.f3 Be6 20.Rad1 Rf8 21.d4 d5 22.Kd3 Bc7 23.Ne2 Ke7 24.e5 Ba5 25.c3 Bb6 26.a4 Rae8 27.Ra1 Ba5 28.b3 Kd7 29.Rg2 Kc8 30.Kc2 Ng7 31.Kb2 Kb8 32.b4 Bd8 33.a5 Bh3 34.Rg3 Bf5 35.Nc1 Ne6 36.Kb3 Be7 37.Rg1 Bh3 38.Nd3 g5 39.fxg5 Rxf3 40.Rg3 Rxg3 41.hxg3 Rg8 42.a6 Bxg5 43.Rh1 Bf5 44.Bxg5 Bxd3 45.Bh4 Bxa6 46.Kc2 Kc8 47.Rh2 Kd7 48.Rf2 Ke8 49.Rf6 Rg6 50.Rf5 b6 51.Rh5 h6 52.Bf6 Kd7 53.Rh3 Bc4 54.g4 a5 55.bxa5 bxa5 56.Rh2 a4 57.Kb2 Nf4 58.Ka3 Bb3 59.Rh4 Nd3 White resigned, Wall,B - Ahmadi,S, Chess.com, 2010. 


Black has the two Bishops, and a Knight for two pawns. However, those are "Jerome pawns", and their future is brighter than it appears.

13.Rg1 Ke7 14.Ne2 Raf8 15.a4 a5 16.f4 g6 17.Be3 Rhg8 18.h4


On behalf of razornfs the computer adopts an anti-computer strategy: keeping the position closed.

18...Bb4+ 

The lichess computer's post-game analysis suggests 18...Bb6 instead. My own Stockfish 13 suggests that White castle Queenside, and Black do the same, by hand.

19.c3 Bc5 20.Kd2 Kf6 21.Rgb1 Bxe3+ 22.fxe3 Ra8 23.b4 b6 24.b5 Ke7 


25.c4 Nf6 26.Rg1 Nd7 27.Rg3 h5 28.Nc3 Nf6 29.Nd5+ Bxd5 30.cxd5 Nd7 

My Stockfish 13 assesses the position as equal. The position is still locked up. Black has possible Knight outposts at g4 and c5. White has that impressive pawn mass in the center. My guess is that most club players would prefer to play White here. 

[to be continued]

No comments:

Post a Comment