Hi... I'm working on engine research

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

FireDragon761138
Posts: 18
Joined: Sun Dec 28, 2025 7:25 am
Full name: Aaron Munn

Re: Hi... I'm working on engine research

Post by FireDragon761138 »

towforce wrote: Tue Dec 30, 2025 12:26 pm
FireDragon761138 wrote: Tue Dec 30, 2025 1:03 am...there should be self-similarity in evaluation from ply to ply in chess search in terms of evaluation (which translates into evaluation stability), because that's the fundamental nature of good chess analysis (stability, positional judgement)...

Engines with HCE (Hand Coded Evals) used to change their evaluations suddenly when search revealed something that the eval had missed. We know that HCEs missed a lot, because when they were replaced with NNs, the ELO rating jumped by hundreds of points, despite the eval taking a lot longer to run.

We also know that NNs miss a lot. They tend to encode a large number of simple "surface" features rather than encoding the deep underlying patterns of chess. We already knew this - but it was then demonstrated to us in a very stark way: in 2023 Adam Pelrine, an amateur Go player, researched KataGo (link), found a fundamental weakness, and used it to beat that engine in a match - link. Basically, he demonstrated that, at that time, KataGo didn't understand the concept of a group of stones - which, from a human perspective, is a remarkable knowledge gap for an engine which had been thought to be better than any human!

Chess engine developers are heavily focused on incrementally improving the strength of engines right now. To my knowledge, nobody is looking into how to uncover the deep underlying patterns of the game.
Yes, that's what I'm interested in, uncovering the deeper structure of chess than just looking at surface features. My approach is based on certain metaphysical and epistemic assumptions that might be "outside the box". I believe the concept of iteration is a flawed epistemological method if you don't focus on good first principles. And that involves engaging with more philosophic questions.
Ryan Benitez
Posts: 726
Joined: Thu Mar 09, 2006 1:21 am
Location: Portland Oregon

Re: Hi... I'm working on engine research

Post by Ryan Benitez »

FireDragon761138 wrote: Wed Dec 31, 2025 3:30 pm The way Stockfish tries to bake in depth 12 analysis at depth 0 into the neural network isn't the only way to design a neural network with rich density of information. In our case, we are working with an Lc0 game dataset. I believe Monte Carlo Tree Search gives more trustworthy positional information than simply having the engine iterate on itself, without a bias towards seeing chess as a series of forcing moves. I want to maintain strategic flexibility at lower depths of search.
The best reason I can think of to use MCTS over alpha-beta is the ability to perform large asynchronous batch evaluations. If you think you're getting better positional information with MCTS, it's almost certainly coming from either the NN itself or imagination. I don't doubt that a chess engine with some element of MCTS will eventually surpass Stockfish, I even consider it likely at this point, given hardware trends. That does not in any way imply that Stockfish is fundamentally flawed.
FireDragon761138
Posts: 18
Joined: Sun Dec 28, 2025 7:25 am
Full name: Aaron Munn

Re: Hi... I'm working on engine research

Post by FireDragon761138 »

Here's a sample annotation I did of a famous Kasparov vs. Deep Blue game, at 10 meganodes per move. Stockfish 17.1 Remote refers to the remote proxy I used, not the actual engine. It's less optimistic about Deep Blue's performance than Stockfish. Analysis is similar to Obsidian, another engine trained on Lc0 data. Analysis loses resolution or accuracy, but doesn't fundamentally change in character at different nodes/move. It's valuing piece coordination and flexibility more than Stockfish.

[Event "IBM Man-Machine"]
[Site "New York, NY USA"]
[Date "1997.05.04"]
[Round "2"]
[White "Deep Blue (Computer)"]
[Black "Garry Kasparov"]
[Result "1-0"]
[ECO "C93"]
[WhiteElo "?"]
[BlackElo "?"]
[EventDate "?"]
[PlyCount "89"]
[Opening "Spanish Opening: Closed, Smyslov Defence"]
[WhiteAccuracy "87.45"]
[BlackAccuracy "73.39"]
[TotalAccuracy "80.54"]

1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Ba4 Nf6 5.O-O Be7 6.Re1 b5 7.Bb3 d6 8.c3 O-O 9.h3
h6 10.d4 Re8 11.Nbd2 Bf8 12.Nf1 Bd7 13.Ng3 Na5 14.Bc2 c5 15.b3 Nc6 16.d5 Ne7
17.Be3 (17.a4 {+0.60/26/1.5" Stockfish 17.1 REMOTE} Ng6 18.Be3 Nh7 19.Qd2 Be7
20.Ra2 Nh4 21.Nxh4 Bxh4 22.Nf1 Rf8 23.Qe2 Bg5 24.Rea1 bxa4 25.bxa4 Bf4 26.Rb1
Qg5 27.Kh1 Qh4 28.Rab2 Ng5 29.f3 Bxe3 30.Nxe3 Bxh3 31.gxh3 Qxh3+ 32.Kg1 f5
33.exf5 Nxf3+ 34.Kf2) (17.Be3 {+0.60/24/1.5" Stockfish 17.1 REMOTE} Ng6 18.a4
Qc7 19.Qd2 Be7 20.Ra2 bxa4 21.bxa4 Nh7 22.Qd1 a5 23.Nf5 Rab8 24.c4 Bf6 25.Bd2
Rb7 26.Ra3 Ra8 27.g3 Ne7 28.Bc3 Nxf5 29.exf5 Qc8) Ng6 (17... Nh7 {+0.57/24/1.5"
Stockfish 17.1 REMOTE} 18.a4 Ng6 19.Ra2 Nf4 20.Qd2 Ng5 21.Nh2 g6 22.Ngf1 Nh7
23.Bxf4 exf4 24.Qxf4 Qe7 25.axb5 axb5 26.Rea1 Rxa2 27.Rxa2 Bg7 28.c4 Ng5 29.Ra7
Be5 30.Qd2 bxc4 31.bxc4 Bd4) (17... Ng6 {+0.61/24/1.5" Stockfish 17.1 REMOTE}
18.a4 Nh7 19.Qd2 Be7 20.Ra2 Nh4 21.Nxh4 Bxh4 22.Nf1 Rf8 23.Qe2 Qc7 24.Rea1 bxa4
25.bxa4 Rfb8 26.Bd3 a5 27.Nd2 Bg5 28.Bxg5 Nxg5 29.Qe3 Ra7 30.Re1 Nh7) 18.Qd2
(18.a4 {+0.60/22/1.7" Stockfish 17.1 REMOTE} Nh7 19.Ra2 Nf4 20.Qd2 Ng5 21.Nh2 g6
22.axb5 axb5 23.Rea1 Rxa2 24.Rxa2 h5 25.h4 Nh7 26.Bxf4 exf4 27.Qxf4 Qb8 28.Nf3
Qb6 29.Nxh5 gxh5 30.e5) (18.Qd2 {+0.55/27/1.4" Stockfish 17.1 REMOTE} Nh7 19.a4
Be7 20.Ra2 Nh4 21.Nxh4 Bxh4 22.Nf1 Rf8 23.Qe2 Qf6 24.axb5 axb5 25.Rea1 Rxa2
26.Rxa2 Qg6 27.Kh2 f5 28.exf5 Bxf5 29.Bxf5 Qxf5 30.Qxb5 Nf6 31.Qc4 Bg5 32.Bxg5
Qxg5 33.Ne3 Qf4+ 34.Kg1 Qxc4 35.Nxc4 Nxd5 36.Nxd6 Nxc3 37.Rc2 Rd8 38.Nb7 Rd1+
39.Kh2 Nb5 40.Rxc5 Nd4) Nh7 (18... Nh7 {+0.55/25/1.5" Stockfish 17.1 REMOTE}
19.a4 Be7 20.Ra2 Nh4 21.Nxh4 Bxh4 22.Nf1 Rf8 23.Qe2 Qf6 24.axb5 axb5 25.Rea1
Rxa2 26.Rxa2 Rb8 27.Ra7 Qd8 28.Bd3 Nf6 29.Qa2 Qc8 30.Nd2 Ne8 31.Qa6 Qxa6 32.Rxa6
Bd8 33.Ra7 Nf6 34.g4 g5 35.f3 Bb6 36.Ra1 Ne8 37.Nf1) 19.a4 (19.a4 {+0.56/23/1.5"
Stockfish 17.1 REMOTE} Be7 20.Ra2 Nh4 21.Nxh4 Bxh4 22.Nf1 Rf8 23.axb5 axb5
24.Rea1 Rxa2 25.Rxa2 Bg5 26.Ra7 Bxe3 27.Nxe3 Nf6 28.Qd1 Qb6 29.Qa1 Rb8 30.Ra6
Qc7 31.b4 cxb4 32.cxb4 Rc8 33.Qa3 g6 34.Qa1 Nh5 35.Ra7) Nh4 (19... Be7
{+0.49/22/1.7" Stockfish 17.1 REMOTE} 20.Bd3 Qc7 21.Reb1 Nh4 22.Nxh4 Bxh4 23.Qe2
Nf6 24.Nh5 Nxh5 25.Qxh5 Bf6 26.Qe2 Qb7 27.b4 c4 28.Bc2 Bg5 29.Bxg5 hxg5 30.Ra3
Reb8 31.Ra2 a5 32.axb5 axb4 33.Rxa8 Qxa8 34.Rxb4 Bxb5) (19... Nh4 {+0.72/25/1.3"
Stockfish 17.1 REMOTE} 20.Nxh4 Qxh4 21.b4 Rec8 22.Ra2 Qd8 23.Rea1 Nf6 24.c4 Rcb8
25.cxb5 axb5 26.bxc5 dxc5 27.axb5 Bxb5 28.Qc3 Rxa2 29.Rxa2 Qc7 30.Ra1 Ne8 31.Ba4
c4 32.Bd1 Bd7 33.Be2 Nd6 34.Qxe5 c3) 20.Nxh4 (20.Nxh4 {+0.75/22/1.5" Stockfish
17.1 REMOTE} Qxh4 21.b4 Rec8 22.Ra2 Qd8 23.Rea1 Nf6 24.c4 Rcb8 25.Qe2 bxc4
26.bxc5 dxc5 27.Qxc4 a5 28.Qe2 Qc7 29.Bd3 Ne8 30.Qc2 Rb4 31.Bd2 Nd6 32.Rc1) Qxh4
(20... Qxh4 {+0.73/23/1.4" Stockfish 17.1 REMOTE} 21.b4 Rec8 22.Qe2 c4 23.Ra3
Qd8 24.Rea1 g6 25.Bd1 h5 26.Qa2 Rab8 27.axb5 axb5 28.Ra6 Be7 29.Qa5 Qe8 30.Ra7
Bg5 31.Bb6 Bd8 32.Bxd8 Rxd8 33.Qc7 h4 34.Nf1 Nf6 35.Bc2 Qe7) 21.Qe2 (21.Qe2
{+0.72/22/1.5" Stockfish 17.1 REMOTE} Rec8 22.Reb1 g6 23.Bd3 Rab8 24.Ra2 Qe7
25.Rba1 Qe8 26.Bd2 Nf6 27.Bc1 Qd8 28.Be3 Bg7 29.Nf1 Nh5 30.axb5 axb5) Qd8 (21...
Rec8 {+0.69/21/1.5" Stockfish 17.1 REMOTE} 22.Bd3 Rab8 23.Ra2 Qd8 24.Rf1 Be7
25.f4 exf4 26.Bxf4 c4 27.bxc4 bxc4 28.Bxc4 Bg5 29.e5 Bxf4 30.Rxf4 dxe5 31.Qxe5
Qg5 32.Qxg5 hxg5 33.Rd4 Nf6 34.Bxa6 Rxc3 35.Bd3) (21... Qd8 {+0.94/24/1.4"
Stockfish 17.1 REMOTE} 22.b4 Qc7 23.Rec1 cxb4 24.cxb4 Be7 25.Bd3 Qb7 26.axb5
axb5 27.Ra5 Rxa5 28.bxa5 Bg5 29.Bxg5 hxg5 30.Qd2 Ra8 31.Qb4 Qa6 32.Ra1 Rb8
33.Nf1 g6 34.g4 Kg7 35.Nd2 Rh8 36.Bf1 Rc8 37.Kg2 Rc2 38.Bd3 Rc8) 22.b4 (22.Bd3
{+0.90/21/1.5" Stockfish 17.1 REMOTE} Rb8 23.axb5 axb5 24.Ra7 Re7 25.Rea1 b4
26.cxb4 Rxb4 27.R7a3 Rb8 28.Bd2 g6 29.Nf1 Bc8 30.Ne3 Reb7 31.Qd1 h5 32.Nc4 Bd7
33.Qc2 Be8 34.Be3 Bd7 35.Na5 Rc7 36.Bd2) (22.b4 {+0.86/25/1.5" Stockfish 17.1
REMOTE} c4 23.Ra2 Be7 24.Rea1 Rb8 25.axb5 axb5 26.Ra6 Bc8 27.Ra8 Bd7 28.Rxb8
Qxb8 29.Bd1 Rc8 30.Ra7 Rc7 31.Ra5 Qc8 32.Qa2 Rb7 33.Bg4 Bxg4 34.hxg4 Bg5 35.Bxg5
Nxg5) Qc7 (22... Qc7 {+0.87/25/1.5" Stockfish 17.1 REMOTE} 23.Rec1 cxb4 24.cxb4
Qb7 25.axb5 axb5 26.Bd3 Be7 27.Ra5 Rxa5 28.bxa5 Bg5 29.Bxg5 hxg5 30.Qd2 Nf6
31.Qb4 Qa6 32.Ra1 g6 33.Qd2 g4 34.h4 Kg7 35.h5 Rh8 36.Rc1 Nxh5 37.Nxh5+ Rxh5
38.Rc7 g3 39.Rxd7 Qc8 40.Bxb5) 23.Rec1 (23.Rec1 {+0.89/23/1.6" Stockfish 17.1
REMOTE} Qb7 24.Rcb1 Rec8 25.c4 cxb4 26.Rxb4 a5 27.Rxb5 Bxb5 28.cxb5 g6 29.Rb1
Be7 30.Bxh6 Bg5 31.Bxg5 Nxg5 32.b6 Nh7 33.Bd3 Rc3 34.h4 Ra3 35.h5 Nf8 36.Qd2 Rb8
37.Nf5 gxf5 38.Qg5+ Kh7 39.Qf6 Rxd3 40.h6 fxe4) c4 (23... Qb7 {+0.89/23/1.6"
Stockfish 17.1 REMOTE} 24.Rcb1 cxb4 25.Rxb4 Rec8 26.c4 a5 27.Rxb5 Bxb5 28.cxb5
g6 29.Bd3 Be7 30.Nf1 Bg5 31.Nd2 Rc3 32.Nc4 Nf6 33.Bxg5 hxg5 34.Rb1 Qc7 35.b6 Qc5
36.Rb5) (23... c4 {+0.96/25/1.6" Stockfish 17.1 REMOTE} 24.Ra3 Rec8 25.Bd1 Qd8
26.Rca1 g6 27.Qa2 h5 28.axb5 Bxb5 29.Qd2 Rab8 30.Bc2 Be7 31.Nf1 Rb7 32.Nh2 Bg5
33.Nf3 Bxe3 34.Qxe3 Kg7 35.Nd2 h4 36.Rf1 Qb6 37.Qxb6 Rxb6 38.f4 Nf6 39.fxe5 dxe5
40.Nf3 Rh8 41.Nxe5) 24.Ra3 (24.Ra3 {+0.94/26/1.5" Stockfish 17.1 REMOTE} Rec8
25.Rca1 Qd8 26.Bd1 g6 27.Qa2 Rab8 28.axb5 Bxb5 29.Qd2 h5 30.Nf1 Be7 31.Bc2 Rb7
32.Ra5 Rcb8 33.Nh2 Bg5 34.Bxg5 Nxg5 35.h4 Nh7 36.g3 Qf8 37.Qe2 Qh6 38.Nf1 Rf8
39.Ne3 f5 40.exf5 g5 41.hxg5 Qxg5 42.Rxa6 Bxa6 43.Rxa6 h4 44.Rxd6 Nf6 45.Nf1
hxg3 46.Nxg3 Rg7 47.Qxe5 Ng4) Rec8 (24... Qb7 {+0.98/21/1.6" Stockfish 17.1
REMOTE} 25.Rca1 Be7 26.Bd1 Bh4 27.axb5 axb5 28.Ra7 Rxa7 29.Rxa7 Qc8 30.Qh5 Bxg3
31.fxg3 Nf6 32.Qh4 Qd8 33.Bf3 Bc8 34.Kh2 Bd7 35.Kh1 Rf8 36.g4 Bc8 37.Kg1 Re8
38.Qf2 Nh7 39.Bd1 Rf8 40.Bc2 Bd7) (24... Rec8 {+0.98/28/1.5" Stockfish 17.1
REMOTE} 25.Rca1 Qd8 26.Bd1 g6 27.Qa2 Rab8 28.axb5 Bxb5 29.Qd2 h5 30.Nf1 Rb7
31.Bc2 Be7 32.Ra5 Kg7 33.Nh2 Bg5 34.Bxg5 Nxg5 35.h4 Nh7 36.g3 Qd7 37.Nf1 Qh3
38.f3 f5 39.Qg2 Qxg2+ 40.Kxg2 f4) 25.Rca1 (25.Bd1 {+0.98/25/1.5" Stockfish 17.1
REMOTE} g6 26.Rca1 Qd8 27.Qa2 h5 28.axb5 Bxb5 29.Qe2 Be7 30.Nf1 Rab8 31.Nd2 Bg5
32.Nf3 Bxe3 33.Qxe3 Rb7 34.Qh6 Qf6 35.Nd2 Qg7 36.Qxg7+ Kxg7 37.h4) (25.Rca1
{+0.96/26/1.6" Stockfish 17.1 REMOTE} Qd8 26.Bd1 g6 27.Qa2 h5 28.axb5 Bxb5
29.Nf1 Be7 30.Qe2 Bg5 31.Ra5 Rab8 32.Nd2 Kg7 33.Bc2 Rb7 34.Nf3 Bxe3 35.Qxe3 Qe7
36.g3 Rh8 37.Kg2 Rf8 38.Bd1 f5 39.exf5 gxf5) Qd8 (25... Qd8 {+0.92/25/1.6"
Stockfish 17.1 REMOTE} 26.Bd1 g6 27.Qa2 Rab8 28.axb5 axb5 29.Ra7 h5 30.Qa5 h4
31.Nf1 f5 32.Nd2 Be8 33.exf5 gxf5 34.Qxd8 Rxd8 35.Bc2 e4 36.Bd4 Rd7 37.f3 Bh6
38.Nf1 Bg7 39.Bxg7 Rxg7 40.fxe4 Rxa7 41.Rxa7 fxe4 42.Bxe4) 26.f4 (26.Bd1
{+0.89/23/1.7" Stockfish 17.1 REMOTE} g6 27.Qa2 f5 28.exf5 gxf5 29.axb5 f4
30.Rxa6 Rxa6 31.bxa6 fxe3 32.fxe3 Qh4 33.Qf2 Ra8 34.Bc2 Nf6 35.Qf3 Bg7 36.Ne4
Nxe4 37.Bxe4 Qg5 38.Kh2 h5 39.Rf1 Rf8 40.Qe2 Rxf1 41.Qxf1 Qxe3 42.Bf5) (26.f4
{+0.51/23/1.6" Stockfish 17.1 REMOTE} exf4 27.Bxf4 g6 28.Kh2 Nf6 29.Be3 Bg7
30.Bd4 h5 31.Qf3 h4 32.Ne2 Qe7 33.axb5 Bxb5 34.Bf2 Nh5 35.Nd4 Rab8 36.Ba4 Qe5+
37.Kh1 Re8) Nf6 ?! (26... exf4 {+0.53/21/1.7" Stockfish 17.1 REMOTE} 27.Bxf4 g6
28.Kh2 Ng5 29.Nf1 Bg7 30.Kh1 Qe7 31.axb5 Bxb5 32.Qe3 Re8 33.Re1 f5 34.exf5 Qf6
35.Qg3 Rxe1 36.Qxe1 Re8 37.Ne3) (26... Nf6 {+1.05/23/1.7" Stockfish 17.1 REMOTE}
27.Qf2 Be7 28.a5 Nh7 29.Rf1 Qf8 30.fxe5 dxe5 31.Nf5 Bxf5 32.Qxf5 Re8 33.Raa1
Rad8 34.Rad1 g6 35.Qf3 h5 36.Kh1 Qg7 37.Bc5 Bd6 38.Rf2 f6 39.Rdf1 Rd7 40.Qg3 Kh8
41.h4 Rf8) 27.fxe5 (27.Qf2 {+1.07/22/1.7" Stockfish 17.1 REMOTE} Be7 28.a5 Qf8
29.Rf1 Nh7 30.fxe5 dxe5 31.Nf5 Bxf5 32.Qxf5 Re8 33.Raa1 Rad8 34.Kh1 Rd6 35.Bc5
Rf6 36.Qd7 Rxf1+ 37.Rxf1 Nf6 38.Qa7 Bxc5 39.bxc5 Qe7 40.Qb6 Nh5) (27.fxe5
{+0.87/25/1.7" Stockfish 17.1 REMOTE} dxe5 28.a5 Nh7 29.Rf1 g6 30.Raa1 h5 31.Rf3
f6 32.Raf1 Kh8 33.Qf2 Qe7 34.Kh1 Qg7 35.Bc5 h4 36.Ne2 g5 37.Ng1 Bxc5 38.bxc5 g4
39.hxg4 Bxg4 40.c6 Bxf3 41.Nxf3) dxe5 (27... dxe5 {+0.86/23/1.7" Stockfish 17.1
REMOTE} 28.a5 Nh7 29.Rf1 g6 30.Raa1 Qh4 31.Qf3 Be8 32.Kh2 Rd8 33.Nh1 Ng5 34.Qd1
Nxe4 35.g3 Nxg3 36.Nxg3 Bd7 37.Qf3 Qxh3+ 38.Kg1 f5 39.Bb6 Re8 40.Rf2 Bd6 41.Rh2
e4 42.Rxh3 exf3 43.Rf1) 28.Qf1 ?! (28.a5 {+1.05/23/1.9" Stockfish 17.1 REMOTE}
Nh7 29.Rf1 g6 30.Raa1 Qh4 31.Qf3 Be8 32.Nh1 Kg7 33.Rf2 h5 34.Kh2 Be7 35.Raf1 Ng5
36.Qe2 Rab8 37.Ng3 Rd8 38.Bb6 Rd6 39.Bc7 Rb7 40.Bxd6 Bxd6) (28.Qf1
{+0.21/26/1.6" Stockfish 17.1 REMOTE} a5 29.axb5 axb4 30.Rxa8 Rxa8 31.Rxa8 Qxa8
32.Qxc4 Qa1+ 33.Kh2 Qe1 34.Bg1 Qxc3 35.Qxc3 bxc3 36.b6 Bc8 37.Ne2 Bb4 38.Nc1 Nd7
39.Ba4 Nc5 40.Na2 Nxa4 41.Nxb4 Kf8 42.Be3 Ke8 43.Kg3 f6 44.Kf3 Kd7 45.Ke2 Kd6
46.Kd3 Nc5+ 47.Kxc3 Nxe4+ 48.Kd3 f5) Ne8 ?! (28... a5 {+0.27/25/1.6" Stockfish
17.1 REMOTE} 29.axb5 axb4 30.Rxa8 Rxa8 31.cxb4 Bxb5 32.Nf5 Ne8 33.Bc5 Rxa1
34.Qxa1 g6 35.Ne3 Qg5 36.Nf1 Bd7 37.Be3 Qe7 38.Ba4 Nd6 39.Bc5 Bxh3) (28... Ne8
{+1.03/24/1.5" Stockfish 17.1 REMOTE} 29.Qf2 Nd6 30.R3a2 Qe8 31.Bb6 g6 32.axb5
axb5 33.Ba7 Be7 34.Nf1 Qf8 35.Ra6 Ne8 36.Be3 Rxa6 37.Rxa6 Bd6 38.Nd2 Kh7 39.Nf3
Kg7 40.Qh4 g5 41.Qg3 Qe7 42.Bd1 Kh8 43.Nh2 Nc7 44.Ra7 f5) 29.Qf2 (29.Qf2
{+0.99/22/1.4" Stockfish 17.1 REMOTE} Nd6 30.R3a2 Qe8 31.Bb6 Be7 32.Kh2 bxa4
33.Qf3 a5 34.bxa5 Rcb8 35.Nh5 f5 36.exf5 e4 37.Qg4 Nxf5 38.Bc7 Qf7 39.Bxa4 Bxa4
40.Bxb8 Rxb8 41.Rxa4 Bd6+ 42.Kg1 Bc5+ 43.Kh1 Qxh5 44.Rxc4 Qxg4 45.hxg4 Ng3+
46.Kh2 Bd6 47.Rc6 Bf4) Nd6 (29... Nd6 {+0.90/23/1.5" Stockfish 17.1 REMOTE}
30.R3a2 Qe8 31.Bb6 Be7 32.Kh2 bxa4 33.Rxa4 Bxa4 34.Bxa4 Nb5 35.Nf5 Qf8 36.Bxb5
axb5 37.Rxa8 Rxa8 38.Bc5 Bxc5 39.bxc5 g6 40.Nd6 h5 41.Nb7 b4 42.c6 bxc3 43.d6
Qh6 44.d7 Qd2 45.Qf6 c2 46.d8=Q+ Rxd8 47.Nxd8 Qf4+ 48.Qxf4 exf4) 30.Bb6 (30.R3a2
{+0.99/25/1.5" Stockfish 17.1 REMOTE} Qe8 31.Bb6 Be7 32.Qf3 Bg5 33.Kh1 g6 34.Nf1
Kg7 35.Bc5 Nb7 36.Qf2 h5 37.Bb6 a5 38.bxa5 Nxa5 39.axb5 Nb7 40.Be3 Bxe3 41.Nxe3
Rxa2 42.Rxa2 Bxb5 43.Ra7 Rc7 44.Qg3 f6 45.Bd1 Kh7 46.Qh4 Qb8 47.Bxh5 Qxa7)
(30.Bb6 {+0.90/24/1.7" Stockfish 17.1 REMOTE} Qh4 31.Bc5 Qd8 32.Rf1 a5 33.axb5
axb4 34.Bxb4 f6 35.b6 Rxa3 36.Bxa3 Rb8 37.Rb1 Nb5 38.Bxf8 Qxf8 39.Rb4 Nxc3
40.Rxc4 Nb5 41.Nh5 Be8 42.Bd1 Qd6 43.Kh2 Qxb6 44.Qg3 Bxh5 45.Bxh5 Nd4 46.Qg4 Kh8
47.Rc3 Rd8 48.Bg6) Qe8 (30... Qh4 {+0.90/21/1.7" Stockfish 17.1 REMOTE} 31.Bc5
Rab8 32.Rf1 Qf4 33.Qe1 Qg5 34.Kh2 a5 35.Qf2 Qf4 36.Qxf4 exf4 37.Rxf4 bxa4
38.Bxa4 axb4 39.Bxb4 Bxa4 40.Rxa4 f6 41.Rf2) (30... Qe8 {+0.97/23/1.6" Stockfish
17.1 REMOTE} 31.R3a2 Be7 32.Qf3 Bg5 33.Kh1 g6 34.Nf1 Kg7 35.Bc5 Nb7 36.Qf2 h5
37.Bb6 Nd6 38.Nd2 a5 39.bxa5 bxa4 40.Bxa4 Bxd2 41.Qxd2 Bxa4 42.Rxa4 Nxe4 43.Qe1
Nf6 44.a6 Nxd5 45.Bf2 f6 46.Ra5 Qc6 47.Rc5 Qd7 48.Qe4 Rxc5 49.Bxc5) 31.R3a2
(31.Kh2 {+1.00/23/1.6" Stockfish 17.1 REMOTE} h5 32.axb5 h4 33.Ne2 axb5 34.Qxh4
Rxa3 35.Rxa3 Ra8 36.Rxa8 Qxa8 37.Ng3 Qa2 38.Bd1 Qd2 39.Bf3 Qxc3 40.Qd8 Qxb4
41.Qxd7 c3 42.Be3 c2 43.Bc1 Qc5 44.Qc6 Qxc6 45.dxc6 b4) (31.R3a2 {+0.97/24/1.6"
Stockfish 17.1 REMOTE} Be7 32.Qf3 Bg5 33.Kh1 g6 34.Nf1 Kg7 35.Bc5 Nb7 36.Qf2 Bf4
37.axb5 Nxc5 38.bxc5 axb5 39.c6 Bxc6 40.dxc6 Rxa2 41.Rxa2 Qxc6 42.g3 Bg5 43.Ra7
Rc7 44.Rxc7 Qxc7 45.h4 Be7 46.Kg2 Bc5 47.Qd2 Qa7 48.Qd5) Be7 (31... Be7
{+0.94/23/1.6" Stockfish 17.1 REMOTE} 32.Qf3 Bg5 33.Kh1 g6 34.Nf1 Kg7 35.Qf2 h5
36.Nd2 Bf4 37.Bc5 Nb7 38.Be3 Bxe3 39.Qxe3 Nd6 40.Qb6 Qe7 41.axb5 Rab8 42.Qe3
axb5 43.Ra6 Re8 44.Kh2 Rbd8) 32.Bc5 (32.Qf3 {+0.95/21/1.6" Stockfish 17.1
REMOTE} Bg5 33.Kh1 g6 34.Nf1 Kg7 35.Bc5 Nb7 36.axb5 axb5 37.Ba7 Rd8 38.Qf2 h5
39.Nd2 Qe7 40.Nf3 Bf4 41.Ra6 Bc8 42.Bb6 Rxa6) (32.Bc5 {+0.59/23/1.4" Stockfish
17.1 REMOTE} a5 33.Nh5 f6 34.Qf3 axb4 35.Bxb4 Rcb8 36.Kh2 Nc8 37.Bxe7 Qxe7
38.axb5 Rxa2 39.Rxa2 Nd6 40.Qf2 Bxb5 41.Ng3 g6 42.Ba4 Kg7 43.Bxb5 Rxb5 44.Ra7
Rb7 45.Ra6 h5 46.Qe3 h4 47.Nf1 f5 48.exf5 gxf5 49.Qc5 Rd7 50.Qc6 Rd8) Bf8 ?!
(32... a5 {+0.49/22/1.6" Stockfish 17.1 REMOTE} 33.Nh5) (32... Bf8
{+1.24/23/1.6" Stockfish 17.1 REMOTE} 33.axb5 axb5 34.Nh5 Rxa2 35.Rxa2 Qd8
36.Qg3 Qg5 37.Qxg5 hxg5 38.Ra7 Rd8 39.Ng3 g4 40.h4 Ne8 41.Kf1 Nf6 42.Be3 Be7
43.Ke1 Kh7 44.Bc5 Bf8 45.Bb6 Re8 46.Be3 Bd6 47.h5 Bb8 48.Rb7 Rd8 49.Bc5) 33.Nf5
(33.axb5 {+1.13/24/1.6" Stockfish 17.1 REMOTE} axb5 34.Ra6 Qd8 35.Bb6 Qe8 36.Ba7
Qd8 37.Bd1 g6 38.Bb6 Rxa6 39.Rxa6 Qg5 40.Bc2 Ne8 41.Ra7 Qe7 42.Nf1 Nf6 43.Nd2
Qd6 44.Nf3 Kg7 45.Be3 Re8 46.Bc5 Qb8 47.Ra6 Nh5 48.Rb6 Qc7) (33.Nf5
{+0.89/25/1.5" Stockfish 17.1 REMOTE} Bxf5 34.exf5 a5 35.f6 g6 36.Qg3 axb4
37.cxb4 Rd8 38.axb5 Rxa2 39.Rxa2 Nxb5 40.Bxf8 Qxf8 41.Bxg6 fxg6 42.Qxg6+ Kh8
43.Ra5 Rb8 44.f7 c3 45.Ra6 Nd4 46.Qh5 Kh7 47.Qxe5 Nf3+ 48.gxf3 Qg7+ 49.Qxg7+
Kxg7 50.Kf2 Kxf7 51.Ke3 Rxb4 52.Rc6 Rh4 53.Rxc3 Rxh3 54.Ke4) Bxf5 (33... Bxf5
{+1.00/24/1.6" Stockfish 17.1 REMOTE} 34.exf5 a5 35.f6 g6 36.Qg3 axb4 37.cxb4 c3
38.h4 Rxa4 39.Bxa4 bxa4 40.Re1 Qb5 41.Qxc3 Nb7 42.Qh3 Bxc5+ 43.bxc5 Rxc5 44.h5
g5 45.Qa3 Rxd5 46.Qxa4 Qb6+ 47.Rf2 Qd8 48.Qa7 Nc5 49.Qe7 Qd7 50.Rxe5 Rxe5
51.Qxe5 Ne6 52.Qb8+ Kh7) 34.exf5 (34.exf5 {+0.96/24/1.5" Stockfish 17.1 REMOTE}
a5 35.f6 g6 36.Qg3 axb4 37.cxb4 c3 38.h4 Rxa4 39.Bxa4 bxa4 40.Re1 Qb5 41.Qxc3
Nb7 42.Qh3 Bxc5+ 43.bxc5 Rxc5 44.h5 g5 45.Qa3 Nd6 46.Qxa4 Qb6 47.Qa7 Qb4 48.Rae2
e4 49.Rd1 Rc4 50.Kh2 Rc8 51.Qd4 Qb5 52.Ree1 Re8 53.Rb1 Qa5 54.Rbc1) f6 (34... a5
{+0.92/21/1.6" Stockfish 17.1 REMOTE} 35.f6 g6 36.Qg3 axb4 37.cxb4 c3 38.h4 Rxa4
39.Bxa4 bxa4 40.Qe3 Qb5 41.h5 g5 42.Rxa4 Qc4 43.Qxe5 c2 44.Rc1 Ne4 45.Bxf8 Rxf8
46.Raa1 Rd8 47.b5 Nd2 48.b6 Nb3 49.b7 Nxc1 50.Rxc1) (34... f6 {+1.18/29/1.3"
Stockfish 17.1 REMOTE} 35.Bxd6 Bxd6 36.Qb6 Qe7 37.Be4 Rcb8 38.Qf2 Qd8 39.axb5
axb5 40.Kh2 Rxa2 41.Qxa2 Bc7 42.h4 Rb6 43.Kg3 Bb8 44.Qf2 Bd6 45.Kf3 Bc7 46.Qa2
Kh8 47.Qa7 Kg8 48.Qa8 Rb8 49.Qa6 Rb6 50.Qa7 Bb8 51.Qa8 Qc7 52.Ra6 Kh7 53.Ra5 Qd7
54.Ke2 Kg8 55.g4 Kf7 56.d6 Rxd6) 35.Bxd6 (35.Bxd6 {+1.18/26/1.4" Stockfish 17.1
REMOTE} Bxd6 36.Qb6 Qe7 37.Be4 Rcb8 38.Qf2 Qd8 39.axb5 axb5 40.Kh2 Rxa2 41.Qxa2
Bc7 42.h4 Rb6 43.Kg3 Bb8 44.Qf2 Bd6 45.Kg4 Bc7 46.d6 Bxd6 47.Qd2 Qe7 48.Ra8+
Kh7) Bxd6 (35... Bxd6 {+1.14/24/1.6" Stockfish 17.1 REMOTE} 36.Qb6 Qd7 37.axb5
Rab8 38.Qxa6 e4 39.Bxe4 Qe7 40.Re1 Qe5 41.Kf1 Re8 42.Rae2 Qf4+ 43.Bf3 Rxe2
44.Rxe2 Be5 45.d6 Bxc3 46.d7 Kh7 47.Qc6 Qc1+ 48.Kf2 Bd4+ 49.Kg3 Qg5+ 50.Kh2 Rd8
51.b6 Be5+ 52.Rxe5 Qf4+ 53.g3 Qxe5 54.b7 c3 55.Bg2 h5 56.h4) 36.axb5 (36.axb5
{+0.94/25/1.6" Stockfish 17.1 REMOTE} e4 37.b6 e3 38.Qf3 Rcb8 39.Qe4 Qe5 40.Qxe5
Bxe5 41.Ra3 Rxb6 42.Kf1 Rd8 43.Be4 g6 44.fxg6 f5 45.Bxf5 Rxd5 46.Rxa6 Rxa6
47.Rxa6 Rd2 48.Ra8+ Kg7 49.Ra7+ Kf8 50.Rf7+ Ke8 51.Bg4 Bxc3 52.Bh5 Rb2 53.Rf4
Bg7 54.Rxc4 Kf8 55.Bf3 Be5 56.Re4 Bc3 57.Rxe3 Bd4) axb5 (36... e4 {+1.03/22/1.7"
Stockfish 17.1 REMOTE} 37.b6 e3 38.Qf3 Rcb8 39.Qe4 Rxb6 40.Kf1 Qe5 41.Qxe5 Bxe5
42.Ra3 Rd8 43.Be4 g6 44.fxg6 f5 45.Bxf5 Rxd5 46.Rxa6 Rxa6 47.Rxa6 Rd2 48.Ra8+
Kg7 49.Ra7+ Kf8 50.Rf7+ Ke8 51.Bg4 Bxc3 52.Bh5 Rc2 53.g3 Bd4 54.h4 Bc3 55.b5
Rb2) (36... axb5 {+1.18/31/1.3" Stockfish 17.1 REMOTE} 37.Be4 Bc7 38.Kh2 Qd8
39.Qd2 Rxa2 40.Qxa2 Rb8 41.h4 Rb6 42.Qf2 Kh8 43.Kg3 Kh7 44.Kf3 Kg8 45.g3 Kh8
46.Qc5 Bd6 47.Qe3 Bc7 48.Kg2 Kh7 49.Qf2 Kg8 50.Kh3 Bd6 51.Ra7 Kh8 52.Qa2 Bb8
53.Ra8 Qc7 54.g4 Kg8) 37.Be4 (37.Be4 {+1.29/28/1.4" Stockfish 17.1 REMOTE} Bc7
38.Kh2 Qd8 39.Qd2 Kh7 40.h4 Rcb8 41.Kh3 Rxa2 42.Qxa2 Kh8 43.Qa6 Rb6 44.Qa8 Rb8
45.Qa2 Rb6 46.Kg4 Rb8 47.Qa7 Kh7 48.Ra6 Bb6 49.Qf7 Ra8 50.Qb7 Rxa6 51.Qxa6 Be3)
Rxa2 ?! (37... Bc7 {+1.18/22/1.5" Stockfish 17.1 REMOTE} 38.Kh2 Qd8 39.h4 Rab8
40.Kh3 Bb6 41.Qd2 Qd6 42.g4 Rd8 43.g5 hxg5 44.Qe2 gxh4 45.Qh5 Be3) (37... Rxa2
{+2.15/21/1.4" Stockfish 17.1 REMOTE} 38.Qxa2 Rd8 39.Qa6 Kh7 40.Kh2 Qh5 41.Qb6
Rd7 42.Qxb5 Rd8 43.Qb6 Rd7 44.Qf2 Rd8 45.Qe3 Rc8 46.g3 Qe8 47.Qb6) 38.Qxa2
(38.Qxa2 {+2.10/23/1.4" Stockfish 17.1 REMOTE} Kh8 39.Qa6 Rd8 40.Qc6 Qh5 41.Qb6
Qe8 42.Kh2 Bf8 43.Ra5 Bd6 44.Ra7 Kh7 45.Qb7 Qg8 46.Qc6 Bb8 47.Rb7 Qf8 48.h4 Bd6
49.Kh3 Kh8 50.Qxb5) Qd7 (38... Rb8 {+1.97/19/1.4" Stockfish 17.1 REMOTE} 39.Qa6
Qd8 40.Qc6 Bc7 41.Ra6 Rb6 42.Rxb6 Bxb6+ 43.Kf1 Bc7 44.Qxb5 Qa8 45.Qxc4 Qa1+
46.Ke2 Bd6 47.Kd2 Qb2+ 48.Bc2 e4 49.Qxe4 Be5 50.Qe3 Kh7 51.Qf3 Bd6 52.h4 Qa1
53.g3 Kg8 54.Qe2 Kf8 55.Qe3 Qb2 56.Qe6 Bxg3) (38... Qd7 {+2.19/26/1.4" Stockfish
17.1 REMOTE} 39.Qa7 Qd8 40.Ra6 Rb8 41.Kf1 Kh7 42.h4 Kg8 43.Ke2 Kh7 44.g3 Qf8
45.Kf3 Qd8 46.Kg2 Kg8 47.Rc6 Kf8 48.Kf2 Rc8 49.Qb6 Rb8 50.Qxd8+ Rxd8 51.Rb6 h5
52.Kf3 Rd7 53.Rxb5 Ra7) 39.Qa7 (39.Qa7 {+2.14/24/1.5" Stockfish 17.1 REMOTE} Qd8
40.Ra6 Rb8 41.Kf1 Kf8 42.Ke2 Kg8 43.h4 Kh7 44.Kf1 Kg8 45.g3 Bc7 46.Rc6 Bd6
47.Kg2 Kf8 48.Kf3 Kg8 49.Qa6 Bf8 50.Re6 Qd7 51.g4 Kh8 52.d6) Rc7 ?! (39... Qd8
{+1.99/21/1.7" Stockfish 17.1 REMOTE} 40.Ra6 Rb8 41.Kf1 Kh7 42.h4 Kg8 43.Rc6 Qf8
44.Qa6 Ra8 45.Qxb5 Qe7 46.Qb6 Ra1+ 47.Kf2 Qe8 48.Rxd6 Qh5 49.Rd8+ Kh7) (39...
Rc7 {+3.11/22/1.4" Stockfish 17.1 REMOTE} 40.Qb6 Qd8 41.Ra6 Rd7 42.Qxb5 Qe7
43.Qc6 Rd8 44.Kh2 Qf7 45.Qb6 Qe8 46.h4 Kh8 47.Kh3 Bb8 48.Qe6 Qf8 49.Qc6 Rc8
50.Qb6 Kh7 51.g3 Qe8 52.Qe6 Qf8 53.Rc6) 40.Qb6 (40.Qb6 {+2.98/23/1.4" Stockfish
17.1 REMOTE} Qd8 41.Ra6 Rd7 42.Qxb5 Qe7 43.Qc6 Rd8 44.Kh2 Kh7 45.g3 Bb8 46.Qxc4
Rd6 47.Qc5 Qd7 48.Rxd6 Bxd6 49.Qc6 Qd8 50.h4 Qe7 51.Kh3 Bb8 52.Qa6 Bd6 53.Qb6 h5
54.b5 Bc5 55.Qc6 Qa7 56.Qe8 Bd6 57.Qxh5+ Kg8) Rb7 (40... Qd8 {+2.74/22/1.5"
Stockfish 17.1 REMOTE} 41.Ra6 Rd7 42.Qxb5 Qe7 43.Qc6 Rd8 44.Kh2 Qf7 45.Qb6 Qe8
46.h4 Kh7 47.Kh3 Bb8 48.Qc5 Qf7 49.Qxc4 Rd6 50.Ra1 Rd8 51.b5) (40... Rb7
{+3.09/21/1.2" Stockfish 17.1 REMOTE} 41.Ra8+ Kh7 42.Qc6 Qe7 43.Re8 Qc7 44.Qxc7
Rxc7 45.Rd8 Be7 46.Ra8 Bd6 47.Kf2 Rb7 48.Ra6 Bb8 49.Ke2 Ba7 50.Bf3 Rd7 51.Bh5
Bb8 52.Rb6 Rd8 53.Rxb5 e4 54.Rc5 g5 55.fxg6+ Kg7) 41.Ra8+ (41.Ra8+
{+2.92/23/1.3" Stockfish 17.1 REMOTE} Kh7 42.Qc6 Qe7 43.Re8 Qd7 44.Kf1 Ra7
45.Qxd7 Rxd7 46.Ra8 Rb7 47.Ra6 Bf8 48.h4 Kg8 49.d6 Rd7 50.Bd5+ Kh7 51.Be6 Rxd6
52.Ra8 Be7 53.Re8 h5 54.Rxe7 Rd3 55.Bf7 Rxc3 56.Bxh5 Rc1+ 57.Ke2) Kf7 ?! (41...
Kh7 {+2.63/22/1.4" Stockfish 17.1 REMOTE} 42.Qc6 Qe7 43.Re8 Qd7 44.Kf1 h5 45.Ra8
h4 46.Qxd7 Rxd7 47.Ke2 g5 48.fxg6+ Kh6 49.Bf5 Rg7 50.Ke3 e4 51.Re8 Re7 52.Rd8
Kg5 53.Rxd6 Kxf5 54.Re6 Rg7) (41... Kf7 {+3.27/23/1.3" Stockfish 17.1 REMOTE}
42.Qa6 Qc7 43.Kh2 Ke7 44.Kg3 Rb6 45.Qa5 Rb8 46.Ra7 Rb7 47.Qxc7+ Rxc7 48.Ra6 Rb7
49.Kh4 Kd7 50.Kh5 Bb8 51.Kg6 Kc8 52.Ra5 Bd6 53.Ra8+ Bb8 54.Ra6 Ba7 55.d6 Rd7
56.Ra5 Kd8 57.h4 Kc8 58.Rxb5 Rxd6 59.Kxg7 Rd3 60.Kxf6 Rxc3) 42.Qa6 (42.Qa6
{+3.25/22/1.3" Stockfish 17.1 REMOTE} Qc7 43.Kh2 Rb6 44.Qa5 Ke7 45.Kg3 Rb8
46.Ra7 Rb7 47.Qxc7+ Rxc7 48.Ra5 Rb7 49.Kg4 Kd8 50.Kh5 Kc8 51.Kg6 Kb8 52.Ra6 Rd7)
Qc7 (42... h5 {+2.87/20/1.4" Stockfish 17.1 REMOTE} 43.Kf2 h4 44.Kf3 Rc7 45.Qb6
Rb7 46.Qc6 Ra7 47.Qxd7+ Rxd7 48.Ra5 Bb8 49.Rxb5 Ba7 50.d6 Kg8 51.Rb7 Rxb7
52.Bxb7 Bb8 53.d7 Bc7 54.Bd5+ Kf8 55.Bxc4 Ke7 56.Be6 Bd8 57.c4 g6) (42... Qc7
{+3.23/24/1.3" Stockfish 17.1 REMOTE} 43.Qc6 Qe7 44.Kh2 Rb8 45.Ra6 Rd8 46.Qxb5
Kg8 47.Qxc4 Qf7 48.g3 Qh5 49.Qf1 Qg5 50.Qe1 h5 51.c4 h4 52.c5 hxg3+ 53.Kg2 Bf8
54.Ra3 Qf4 55.Rf3 Qh4 56.Rxg3 Kh8 57.Qe3 Ra8 58.Rg4 Ra2+ 59.Kg1) 43.Qc6 (43.Qc6
{+3.12/23/1.4" Stockfish 17.1 REMOTE} Qd7 44.Kh2 Ke7 45.Kg3 Ra7 46.Qxd7+ Rxd7
47.Ra5 Rb7 48.Kg4 Kd8 49.Kh5 Bb8 50.Kg6 Ba7 51.d6 Rd7 52.h4 Kc8 53.Rxb5 Bf2
54.h5 Rxd6 55.Kxg7 Be3 56.Kg6 Kd7 57.Rb7+ Kc8) Qb6+ ?! (43... Qe7 {+2.83/22/1.6"
Stockfish 17.1 REMOTE} 44.Kh2 Rb8 45.Ra6 Rd8 46.Qxb5 Kg8 47.Qxc4 Qf8 48.g3 Kh7
49.h4 Ra8 50.Rc6 Ra1 51.Qe2 Qa8 52.Kh3 Ra2 53.Qd3 Qb8 54.b5 Qf8 55.Qe3 Qe7 56.b6
Qd7 57.c4 Ra3) (43... Qb6+ {+3.64/25/1.3" Stockfish 17.1 REMOTE} 44.Kh2 Rb8
45.Qxb6 Rxb6 46.Kg3 Rb7 47.Kg4 Bb8 48.Kh5 Rb6 49.Ra5 Rb7 50.Ra6 Rd7 51.Rb6 Ba7
52.Rxb5 Bf2 53.Rb8 Be1 54.b5 Ke7 55.Rc8 Rd8 56.Rxc4 Rd6 57.Kg6 Rb6 58.Kxg7)
44.Kf1 (44.Kh2 {+3.48/22/1.5" Stockfish 17.1 REMOTE} Rb8 45.Qxb6 Rxb6 46.Kg3 Ke7
47.Kg4 Bc7 48.Ra7 Kd8 49.Kh5 Rd6 50.Kg6 Bb6 51.Rxg7 Bf2 52.Rb7 Be1 53.Rxb5 Bxc3
54.Rc5 Bd2 55.Rxc4 Rb6 56.b5 Bg5 57.g3 Kd7 58.h4 Bd2 59.Rc6) (44.Kf1
{+3.09/23/1.5" Stockfish 17.1 REMOTE} Rb8 45.Qxb6 Rxb6 46.Kf2 Bb8 47.Kg3 Ke7
48.Kg4 Kd7 49.Kh5 Rb7 50.Kg6 Kc8 51.Ra6 Rd7 52.Ra5 Ba7 53.d6 Kd8 54.h4 Bb8
55.Rxb5 Bxd6 56.Rb6 Bc7 57.Rc6 Rd3 58.Kxg7 Rxc3 59.Kxf6 Kd7 60.b5) Rb8 (44...
Rb8 {+2.90/21/1.6" Stockfish 17.1 REMOTE} 45.Qxb6 Rxb6 46.Kf2 Rb8 47.Ra7+ Ke8
48.Kf3 Rc8 49.Ra5 Rb8 50.Kg4 Rb7 51.Kh5 Kd7 52.Kg6 Kc8 53.Ra8+ Bb8 54.h4 Rd7)
45.Ra6 ? (45.Qxb6 {+2.69/23/1.7" Stockfish 17.1 REMOTE} Rxb6 46.Kf2 Ke7 47.Ra7+
Kd8 48.Kf3 Bb8 49.Rxg7 Ra6 50.Rg8+ Kc7 51.Rf8 h5 52.h4 Ba7 53.Rf7+ Kd8 54.d6 Bg1
55.Rxf6 Ra3) (45.Ra6 {+0.84/25/1.5" Stockfish 17.1 REMOTE} Qe3 46.Qxd6 Re8
47.Qd7+ Re7 48.Qc6 Qxe4 49.d6 Qxf5+ 50.Kg1 Rd7 51.Qxb5 Qe6 52.Qc5 Kg6 53.Rc6 e4
54.Qxc4 Qxc4 55.Rxc4 Kf5 56.b5 Ke5 57.Rb4 Kxd6 58.Rd4+ Kc7 59.Rxe4 Kb6 60.c4 Kc5
61.Kf2 Rd2+ 62.Kg3 Rb2 63.Rg4 g5 64.Re4 h5 65.h4 Kd6) 1-0
User avatar
hgm
Posts: 28435
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: Hi... I'm working on engine research

Post by hgm »

syzygy wrote: Wed Dec 31, 2025 9:58 amWhat counts is the average performance, i.e. maximum Elo.
Says who? It seems to me that you are assuming this as an axiom, which of course is the easiest way to prove anything...
chrisw
Posts: 4757
Joined: Tue Apr 03, 2012 4:28 pm
Location: Midi-Pyrénées
Full name: Christopher Whittington

Re: Hi... I'm working on engine research

Post by chrisw »

hgm wrote: Thu Jan 01, 2026 8:59 am
syzygy wrote: Wed Dec 31, 2025 9:58 amWhat counts is the average performance, i.e. maximum Elo.
Says who? It seems to me that you are assuming this as an axiom, which of course is the easiest way to prove anything...
Yes, you’re correct. Chess development can be likened to the man who only has only one tool, a hammer. And only one way of measuring, win or loss. SPSA is very much hit engine with hammer, measure win loss rate, keep if wins, try another hammer blow if losing. NN training is a little better, apply a bunch of positions, calculate tiny weight adjustment to bring outputs a little better in line, apply another bunch of positions, repeat. In both cases the measuring tool is win/loss (Elo) and the adjustment a hammer.
So, yes, development is guided by win/loss (elo), so Elo is what counts, says the hammerer.
Uri Blass
Posts: 11139
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: Hi... I'm working on engine research

Post by Uri Blass »

hgm wrote: Thu Jan 01, 2026 8:59 am
syzygy wrote: Wed Dec 31, 2025 9:58 amWhat counts is the average performance, i.e. maximum Elo.
Says who? It seems to me that you are assuming this as an axiom, which of course is the easiest way to prove anything...
For me different things are important and I would like to buy an engine that can force mate with a smaller number of moves in games with odds.
Stockfish is not the best and old Wasp can mate faster against stockfish in all my queen odds tests.
FireDragon761138
Posts: 18
Joined: Sun Dec 28, 2025 7:25 am
Full name: Aaron Munn

Re: Hi... I'm working on engine research

Post by FireDragon761138 »

chrisw wrote: Thu Jan 01, 2026 10:09 am
hgm wrote: Thu Jan 01, 2026 8:59 am
syzygy wrote: Wed Dec 31, 2025 9:58 amWhat counts is the average performance, i.e. maximum Elo.
Says who? It seems to me that you are assuming this as an axiom, which of course is the easiest way to prove anything...
Yes, you’re correct. Chess development can be likened to the man who only has only one tool, a hammer. And only one way of measuring, win or loss. SPSA is very much hit engine with hammer, measure win loss rate, keep if wins, try another hammer blow if losing. NN training is a little better, apply a bunch of positions, calculate tiny weight adjustment to bring outputs a little better in line, apply another bunch of positions, repeat. In both cases the measuring tool is win/loss (Elo) and the adjustment a hammer.
So, yes, development is guided by win/loss (elo), so Elo is what counts, says the hammerer.
Evaluation (neural network) is only part of what makes an engine efficient. The quality of the code for search is also important.

I'm not arguaing that Stockfish isn't coded well. I'm arguing that the philosophy behind it's evaluation may not be ideal.
FireDragon761138
Posts: 18
Joined: Sun Dec 28, 2025 7:25 am
Full name: Aaron Munn

Re: Hi... I'm working on engine research

Post by FireDragon761138 »

I might want to experiment with a fatter network for Project Theoria. Using the default search of Stockfish, the Lc0 trained network goes quite deep, sometimes 20 moves or more. A fatter network would be slower, but might be able to draw out more nuance about a position and encode more knowledge within its weights, shifting the balance from search to evaluation.
syzygy
Posts: 5825
Joined: Tue Feb 28, 2012 11:56 pm

Re: Hi... I'm working on engine research

Post by syzygy »

hgm wrote: Thu Jan 01, 2026 8:59 am
syzygy wrote: Wed Dec 31, 2025 9:58 amWhat counts is the average performance, i.e. maximum Elo.
Says who? It seems to me that you are assuming this as an axiom, which of course is the easiest way to prove anything...
Says I, and I explained why. In chess you win some, you lose some. Every engine will make mistakes in some positions. If you focus on not making any mistakes, you end up with an engine that plays hundreds of Elo weaker than the strongest engine, which means it will just make MORE mistakes, at least in practical play.

I understand that you don't care about practical play but about doing well in artificial positions that trick most engines but(/because) they will never show up in a real game. That is fine!
syzygy
Posts: 5825
Joined: Tue Feb 28, 2012 11:56 pm

Re: Hi... I'm working on engine research

Post by syzygy »

chrisw wrote: Thu Jan 01, 2026 10:09 am
hgm wrote: Thu Jan 01, 2026 8:59 am
syzygy wrote: Wed Dec 31, 2025 9:58 amWhat counts is the average performance, i.e. maximum Elo.
Says who? It seems to me that you are assuming this as an axiom, which of course is the easiest way to prove anything...
Yes, you’re correct. Chess development can be likened to the man who only has only one tool, a hammer. And only one way of measuring, win or loss. SPSA is very much hit engine with hammer, measure win loss rate, keep if wins, try another hammer blow if losing. NN training is a little better, apply a bunch of positions, calculate tiny weight adjustment to bring outputs a little better in line, apply another bunch of positions, repeat. In both cases the measuring tool is win/loss (Elo) and the adjustment a hammer.
So, yes, development is guided by win/loss (elo), so Elo is what counts, says the hammerer.
Note that HGM took my sentence out of its proper context. The argument was that the goal of engine development should be to make an engine that makes no mistakes whatsoever ("reliability in any conceivable situation"). But playing it safe is not going to result in an engine that will play particularly well.

So what's wrong with playing well?