Chess Engine Leptir

[Eduard]

March 10, 2023:

Some of my friends have tested Leptir Analyzer in games. The analyzer engine is made for deep analyses or long timecontols, the search depth is too low for fast games. This is even more noticeable when playing with short books (less than 10 moves) and Ponder OFF. But ok, everyone can test as he wants.

Ipman has achieved the following results:
https://ipmanchess.yolasite.com/

i9 7980XE 18cores/36threads @4.2Ghz, Ponder OFF

Hash: 128Mb, 6men-syzygy (settings are of course for my 10s+1s games, so i didn't change them for 5min or 10min games).

10m+0s games, played with 1core, Noomen 3moves, using Cutechess Gui, concurrency=18

Code: Select all

First test using TC: 10s+1s
My Tournament  2023
                                 
1   Stockfish 15.1 avx512     +5  +46/=923/-31 50.75%  507.5/1000
2   Leptir Analyzer avx512    -5  +31/=923/-46 49.25%  492.5/1000

Code: Select all

Second TC: 5m+0s
My Tournament  2023
                                 
1   Stockfish 15.1 avx512     +1  +22/=960/-18 50.20%  502.0/1000
2   Leptir Analyzer avx512    -1  +18/=960/-22 49.80%  498.0/1000

Code: Select all

10min.games
My Tournament  2023
                                 
1   Stockfish 15.1 avx512     +5  +13/=481/-6 50.70%  253.5/500
2   Leptir Analyzer avx512    -5  +6/=481/-13 49.30%  246.5/500

Code: Select all

Engine                 Depth       Time   Games     Moves  Average Forfeit  Book Depth     MIDG   EARLY    ENDG    LATE
Leptir Analyzer avx512   35.94   63:30:20     500     31396    7.28     0     1488  2.98    31.13 | 32.58 | 37.81 | 46.46
Stockfish 15.1  avx512   41.13   62:07:53     500     31403    7.12     0     1487  2.97    34.83 | 36.39 | 43.90 | 54.60

Download all 5m+0s games and 10m+0s games:
https://filehorst.de/d/eapnbhrA

----------------------------------------------

Test on PlayChess by a friend:
i7-4720HQ CPU @ 2.60GHz 2594MHz, (4 cores, 8 threads), about 2700 kns.
Tournament CTG-Openings-book

Three days ago shared 1st place with +1 at the 5m+0s tournament
The day before yesterday sole winner with +1 at the 5m+0s tournament
Yesterday 50% points at the 5m+0s tournament

I myself would never play a tournament with Leptir Analyzer with such slow hardware. The hardware would be too weak for me personally because the engine has a slow search. I am all the more surprised about all the results.

[DrEinstein]

March 10, 2023:

Some of my friends have tested Leptir Analyzer in games. The analyzer engine is made for deep analyses or long timecontols, the search depth is too low for fast games. This is even more noticeable when playing with short books (less than 10 moves) and Ponder OFF. But ok, everyone can test as he wants.

Ipman has achieved the following results:
https://ipmanchess.yolasite.com/

i9 7980XE 18cores/36threads @4.2Ghz, Ponder OFF

Hash: 128Mb, 6men-syzygy (settings are of course for my 10s+1s games, so i didn't change them for 5min or 10min games).

10m+0s games, played with 1core, Noomen 3moves, using Cutechess Gui, concurrency=18

Code: Select all

First test using TC: 10s+1s
My Tournament  2023
                                 
1   Stockfish 15.1 avx512     +5  +46/=923/-31 50.75%  507.5/1000
2   Leptir Analyzer avx512    -5  +31/=923/-46 49.25%  492.5/1000

Code: Select all

Second TC: 5m+0s
My Tournament  2023
                                 
1   Stockfish 15.1 avx512     +1  +22/=960/-18 50.20%  502.0/1000
2   Leptir Analyzer avx512    -1  +18/=960/-22 49.80%  498.0/1000

Code: Select all

10min.games
My Tournament  2023
                                 
1   Stockfish 15.1 avx512     +5  +13/=481/-6 50.70%  253.5/500
2   Leptir Analyzer avx512    -5  +6/=481/-13 49.30%  246.5/500

Code: Select all

Engine                 Depth       Time   Games     Moves  Average Forfeit  Book Depth     MIDG   EARLY    ENDG    LATE
Leptir Analyzer avx512   35.94   63:30:20     500     31396    7.28     0     1488  2.98    31.13 | 32.58 | 37.81 | 46.46
Stockfish 15.1  avx512   41.13   62:07:53     500     31403    7.12     0     1487  2.97    34.83 | 36.39 | 43.90 | 54.60

Download all 5m+0s games and 10m+0s games:
https://filehorst.de/d/eapnbhrA

----------------------------------------------

Test on PlayChess by a friend:
i7-4720HQ CPU @ 2.60GHz 2594MHz, (4 cores, 8 threads), about 2700 kns.
Tournament CTG-Openings-book

Three days ago shared 1st place with +1 at the 5m+0s tournament
The day before yesterday sole winner with +1 at the 5m+0s tournament
Yesterday 50% points at the 5m+0s tournament

I myself would never play a tournament with Leptir Analyzer with such slow hardware. The hardware would be too weak for me personally because the engine has a slow search. I am all the more surprised about all the results.

Thanks Eduard, looks good and interesting to me. I have just downloaded the sources, will have a look on it and make my own compiles.

[AlexChess]

SIC TRANSIT GLORIA MUNDI...

March 30, 2023 Last on today Playchess.com tournament

...And since you are so interested in ProteusSF, don't forget to mention when you lost against it

[Chessqueen]

I would like to know if Leptir can find the move Bd3 that Lasker made against Gooding in 1908. Lasker Vs Gooding 1908 here on this position Emanuel Lasker played Bd2, was it a Blunder or any top engine above 2800 of today would have punished him severely?

NOTE: I do NOT know the story behind this game, but Bd2? is NOT the best move here, probably they had touched move and Lasker touched the Bishop and had to move it, but since Lasker was probably 80 to 100 elo stronger than Gooding he managed to beat him which if you pit two engines with elo difference between 80 thru 100 the stronger engine after forcing Bd2 ? can still manage to beat the weaker engine
[pgn][Event "Lasker Vs Gooding 1908"]
[Site "DESKTOP-OFQ3C0P"]
[Date "2023.03.10"]
[Round "?"]
[White "Leptir Speed-avx2"]
[Black "Leptir Speed-avx2"]
[Result "*"]
[BlackElo "2000"]
[Time "13:34:52"]
[WhiteElo "2000"]
[TimeControl "900+10"]
[SetUp "1"]
[FEN "r1bqk2r/ppp1npp1/4p3/3pPn1p/1b1P1Q2/2NB1N2/PPP2PPP/R1B1K2R w KQkq - 0 1"]
[Termination "unterminated"]
[PlyCount "0"]
[WhiteType "human"]
[BlackType "human"]

*[/pgn]

[Eduard]

SIC TRANSIT GLORIA MUNDI...

March 30, 2023 Last on today Playchess.com tournament

...And since you are so interested in ProteusSF, don't forget to mention when you lost against it

This is not my account He plays how he wants, I didn't tell him to play like that. I was surprised when he won the tournament twice with his hardware. that's great. But you and your friends are cheaters. I have clearly shown here that you and your friends are cheaters. Your test accounts ProteusSF-Crystal and ProteusSF-Sunrise cheated. If you don't stop it, I'll send everything documented to ChessBase. If you can play like that, can I too? Everyone is allowed to do it? I don't think ChessBase would be pleased. You should be kicked off the server.

Anyone who reads here can easily come to the conclusion that there was cheating here.

You also blocked my account in the Banksia forum. You edited one of my postings first (you're not allowed to do that) and manipulated it. After that, when I corrected that, you deleted the thread. And then you blocked my account too. You will be not bleased.

[AlexChess]

It seems that Eduard knows very well the tricks to manupulate ELO on Playchess.com

...For example suggesting to his testers to play a lot of games against the weaker engine running on poor Raspberry-P3 79 kN/s



et voilà!



Now he is convinced to be a so skilled author (random changing sources values) that he can freely criticize Vondele and Stockfish developers.

I'd really like to avoid comments about Eduard's childish posts. But after being kickeded off from Outskirts, BanksiaGui and ComputerSchach due to heavy offenses to almost all members there, now is disparaging me even on Talkchess...

PS: If you watch all my posts and even infos on my accounts, since the beginning I have clearly explained than on my own accounts the ELO is RELATIVE to my 4 cores, since I test all engines values and my engines play sometimes even together. Instead I never play against other ProteusSF testers so their ELOs are absolutely reliable. Anyway Playchess is absolutely not a rating list. If you want a statistical approach go to SPCC. But he doesn't understand...

[Dann Corbit]

For the test set, two problems have alternative solutions:
For this problem, Qd2 and Bc2 are permutions (examine the pvs):
[fen]1nrq1rk1/p4pp1/bp2pn1p/3p4/2PP1B2/P1PB2N1/4QPPP/1R2R1K1 w - - [/fen]
1nrq1rk1/p4pp1/bp2pn1p/3p4/2PP1B2/P1PB2N1/4QPPP/1R2R1K1 w - - acd 53/56; acs 3584; acn 164479853k; ce 0; bm Qe3; pv Qe3 Bxc4 Bc2 Ng4 Qd2 Qh4 Nh1 Nd7 f3 Ngf6 g3 Qh5 g4 Qh4 Bd6 Rfe8 Bg3 Qg5 Bf4 Qh4;
1nrq1rk1/p4pp1/bp2pn1p/3p4/2PP1B2/P1PB2N1/4QPPP/1R2R1K1 w - - acd 54/75; acs 3584; acn 164479853k; ce 53; bm Qd2 Bc2; pv Qd2 Bxc4 Bc2 Ne4 Bxe4 dxe4 Nxe4 Qd5 Nd6 Nc6 Bxh6 Bd3 Qxd3 Qxd6 Bg5 Qxa3 Bf6 Qd6 Re3 Ne7 Rb5 Ng6 Rg3 Rc7 Be5 Nxe5 Rxe5 Rfc8 Reg5 Kf8 Rxg7 Qf4 Rg8+ Ke7 Rxc8 Qc1+ Qf1 Qxf1+ Kxf1 Rxc8 Ke2 a5 Re3 b5 h4 Kf6 Kd2 Rb8 Rd3 a4 d5 exd5 Rxd5 Kg7 Rd4 Re8 Rb4 Rd8+ Kc1 Rd5 f3 a3;
1nrq1rk1/p4pp1/bp2pn1p/3p4/2PP1B2/P1PB2N1/4QPPP/1R2R1K1 w - - acd 54/75; acs 3584; acn 164479853k; ce 62; bm Qd2 Bc2; pv Bc2 Bxc4 Qd2 Ne4 Bxe4 dxe4 Nxe4 Qd5 Nd6 Nc6 Bxh6 Bd3 Qxd3 Qxd6 Bg5 Qxa3 Bf6 Qd6 Re3 Ne7 Rb5 Ng6 Rg3 Rc7 Be5 Nxe5 Rxe5 Rfc8 Reg5 Qf4 Rxg7+ Kf8 Rg8+ Ke7 Rxc8 Qc1+ Qf1 Qxf1+ Kxf1 Rxc8 Ke2 a5 Re3 b5 h4 Kf6 Kd2 Rb8 Rd3 a4 d5 exd5 Rxd5 Kg7 Rd4 Ra8 Rb4 a3 Rb1 Ra4;

This problem I was not sure of, but Qh5 is clearly better:
[fen]r1b2rk1/pp1n1ppp/4p3/3pP3/1bPNnB1q/2NB2P1/PPQ2P1P/R3K2R b KQ - [/fen]
r1b2rk1/pp1n1ppp/4p3/3pP3/1bPNnB1q/2NB2P1/PPQ2P1P/R3K2R b KQ - acd 51/58; acs 2982; acn 130400337k; ce -28; bm Qh3; pv Qh3 O-O-O Nxc3 bxc3 Ba3+ Kb1 Nb6 Bf1 Qh5 Be2 Qg6 Bd3 f5 cxd5 Nxd5 Bc1 Bxc1 Rxc1 Bd7 Ka1 Qe8 Qb2 Bc6 Qa3 Nc7 Rhg1 Rc8 Qb4 Qd7 Qd6 Rfd8 Qxd7 Rxd7 f4 Bd5 Kb2 b5 Rgd1 Rb8 Be2 h6 a3 Rf7;
r1b2rk1/pp1n1ppp/4p3/3pP3/1bPNnB1q/2NB2P1/PPQ2P1P/R3K2R b KQ - acd 51/62; acs 2982; acn 130400337k; ce -40; bm Qe7; pv Qe7 O-O Bxc3 bxc3 Ndc5 Be2 f6 exf6 Rxf6 Be5 Rf8 Rae1 Bd7 Nb3 Rad8 Nxc5 Nxc5 f4 dxc4 Bxc4 b5 Be2 Bc6 Bd4 e5 fxe5 Ne6 Rxf8+ Rxf8 Qb3 Kh8 Rf1 Ng5 h4 Rxf1+ Kxf1 Nf3 Be3 Nxe5 Qb4 Qf6+ Qf4 a6 Qxf6 gxf6 Bd4 Bd5 Bxe5 fxe5 a3 h6 Ke1 Kh7 Kf2;
r1b2rk1/pp1n1ppp/4p3/3pP3/1bPNnB1q/2NB2P1/PPQ2P1P/R3K2R b KQ - acd 51/57; acs 2982; acn 130400337k; ce -5; bm Qh5; pv Qh5 O-O g5 cxd5 gxf4 dxe6 fxe6 Nxe4 Nxe5 gxf4 Qg4+ Kh1 Qxf4 f3 Kh8 Rad1 b6 Bb5 Rg8 Nc6 Nxc6 Bxc6 Rb8 Ng3 e5 Be4 Bh3 Qc7 Qh6 Qxe5+ Qg7 Qxg7+ Rxg7 Rf2 a5 Nh5 Rf7 Rc2 Bc5 b3 Rg8 Ng3 Rg5 Rd8+ Kg7 Rc1 Rf8 Rxf8 Kxf8;

For this problem, both Be2 g4 are solutions because the pvs are permutations:
[fen]r2qr1k1/pp2bpp1/2pp3p/4nbN1/2P4P/4BP2/PPPQ2P1/1K1R1B1R w - -[/fen]
r2qr1k1/pp2bpp1/2pp3p/4nbN1/2P4P/4BP2/PPPQ2P1/1K1R1B1R w - - acd 50/55; acs 3355; acn 154809279k; ce -11; bm a3; pv a3 Bf6 Be2 Qe7 Ne4 Bxe4 fxe4 Nd7 Bf4 Nc5 Bxd6 Qxe4 Bf3 Qxc4 Bxc5 Qxc5 h5 Bg5 Qd3 Re3 Qd4 Qxd4 Rxd4 Re7 Rhd1 g6 Rd7 Rae8 c3 Kg7 Kc2 Kf6 R7d6+ Re6 Rxe6+ Rxe6 a4 Ke7 Bg4 f5 Bf3 gxh5 Bxh5 Re5 Kb3 Ke6 Bf3;
r2qr1k1/pp2bpp1/2pp3p/4nbN1/2P4P/4BP2/PPPQ2P1/1K1R1B1R w - - acd 50/62; acs 3355; acn 154809279k; ce 27; bm Be2 g4; pv g4 hxg5 Be2 Bg6 hxg5 d5 f4 Bxg5 fxg5 Be4 Rh2 Ng6 Rf1 Qd6 Rhf2 Rad8 Rxf7 d4 Bg1 c5 R7f2 b6 Bh2 Qd7 b3 d3 cxd3 Bxd3+ Bxd3 Qxd3+ Qxd3 Rxd3 Kb2 Re4 Bb8 Rd7 Ka3 Re8 Bf4 a6 Bh2 Re4 Rc2 Kh7 Rff2 Rxg4 Rcd2 Rgd4 Rxd4 cxd4 Bg1;
r2qr1k1/pp2bpp1/2pp3p/4nbN1/2P4P/4BP2/PPPQ2P1/1K1R1B1R w - - acd 50/62; acs 3355; acn 154809279k; ce 30; bm Be2 g4; pv Be2 hxg5 hxg5 d5 g4 Bg6 f4 Bxg5 fxg5 Be4 Rh2 Ng6 Rf1 Qd6 Rhf2 Rad8 Rxf7 d4 Bg1 c5 R7f2 Qd7 Bh2 d3 cxd3 Bxd3+ Bxd3 Qxd3+ Qxd3 Rxd3 b3 b6 Kc1 Re7 Kb2 Re4 Bb8 a6 Kc1 b5 cxb5 axb5 Rd2 Rxd2 Kxd2 c4 Kc3 cxb3 Kxb3 Rxg4 Rf5;

[Eduard]

It seems that Eduard knows very well the tricks to manupulate ELO on Playchess.com

...For example suggesting to his testers to play a lot of games against the weaker engine running on poor Raspberry-P3 79 kN/s



et voilà!



Now he is convinced to be a so skilled author (random changing sources values) that he can freely criticize Vondele and Stockfish developers.

I'd really like to avoid comments about Eduard's childish posts. But after being kickeded off from Outskirts, BanksiaGui and ComputerSchach due to heavy offenses to almost all members there, now is disparaging me even on Talkchess...

PS: If you watch all my posts and even infos on my accounts, since the beginning I have clearly explained than on my own accounts the ELO is RELATIVE to my 4 cores, since I test all engines values and my engines play sometimes even together. Instead I never play against other ProteusSF testers so their ELOs are absolutely reliable. Anyway Playchess is absolutely not a rating list. If you want a statistical approach go to SPCC. But he doesn't understand...

Liar! You even fake pictures to portray one of the most honest players on PlayChesas as a cheater. I will contact the moderator

And you don't answer my questions.



if you scroll down



There are the real pictures of A2-a3, that's the order he played and not the one in your picture. But you deleted that in the Banksia forum. And then you locked me out when I got it right.

I will now contact the moderator of the CCC forum.

[Eduard]

For the test set, two problems have alternative solutions:
For this problem, Qd2 and Bc2 are permutions (examine the pvs):
[fen]1nrq1rk1/p4pp1/bp2pn1p/3p4/2PP1B2/P1PB2N1/4QPPP/1R2R1K1 w - - [/fen]
1nrq1rk1/p4pp1/bp2pn1p/3p4/2PP1B2/P1PB2N1/4QPPP/1R2R1K1 w - - acd 53/56; acs 3584; acn 164479853k; ce 0; bm Qe3; pv Qe3 Bxc4 Bc2 Ng4 Qd2 Qh4 Nh1 Nd7 f3 Ngf6 g3 Qh5 g4 Qh4 Bd6 Rfe8 Bg3 Qg5 Bf4 Qh4;
1nrq1rk1/p4pp1/bp2pn1p/3p4/2PP1B2/P1PB2N1/4QPPP/1R2R1K1 w - - acd 54/75; acs 3584; acn 164479853k; ce 53; bm Qd2 Bc2; pv Qd2 Bxc4 Bc2 Ne4 Bxe4 dxe4 Nxe4 Qd5 Nd6 Nc6 Bxh6 Bd3 Qxd3 Qxd6 Bg5 Qxa3 Bf6 Qd6 Re3 Ne7 Rb5 Ng6 Rg3 Rc7 Be5 Nxe5 Rxe5 Rfc8 Reg5 Kf8 Rxg7 Qf4 Rg8+ Ke7 Rxc8 Qc1+ Qf1 Qxf1+ Kxf1 Rxc8 Ke2 a5 Re3 b5 h4 Kf6 Kd2 Rb8 Rd3 a4 d5 exd5 Rxd5 Kg7 Rd4 Re8 Rb4 Rd8+ Kc1 Rd5 f3 a3;
1nrq1rk1/p4pp1/bp2pn1p/3p4/2PP1B2/P1PB2N1/4QPPP/1R2R1K1 w - - acd 54/75; acs 3584; acn 164479853k; ce 62; bm Qd2 Bc2; pv Bc2 Bxc4 Qd2 Ne4 Bxe4 dxe4 Nxe4 Qd5 Nd6 Nc6 Bxh6 Bd3 Qxd3 Qxd6 Bg5 Qxa3 Bf6 Qd6 Re3 Ne7 Rb5 Ng6 Rg3 Rc7 Be5 Nxe5 Rxe5 Rfc8 Reg5 Qf4 Rxg7+ Kf8 Rg8+ Ke7 Rxc8 Qc1+ Qf1 Qxf1+ Kxf1 Rxc8 Ke2 a5 Re3 b5 h4 Kf6 Kd2 Rb8 Rd3 a4 d5 exd5 Rxd5 Kg7 Rd4 Ra8 Rb4 a3 Rb1 Ra4;

This problem I was not sure of, but Qh5 is clearly better:
[fen]r1b2rk1/pp1n1ppp/4p3/3pP3/1bPNnB1q/2NB2P1/PPQ2P1P/R3K2R b KQ - [/fen]
r1b2rk1/pp1n1ppp/4p3/3pP3/1bPNnB1q/2NB2P1/PPQ2P1P/R3K2R b KQ - acd 51/58; acs 2982; acn 130400337k; ce -28; bm Qh3; pv Qh3 O-O-O Nxc3 bxc3 Ba3+ Kb1 Nb6 Bf1 Qh5 Be2 Qg6 Bd3 f5 cxd5 Nxd5 Bc1 Bxc1 Rxc1 Bd7 Ka1 Qe8 Qb2 Bc6 Qa3 Nc7 Rhg1 Rc8 Qb4 Qd7 Qd6 Rfd8 Qxd7 Rxd7 f4 Bd5 Kb2 b5 Rgd1 Rb8 Be2 h6 a3 Rf7;
r1b2rk1/pp1n1ppp/4p3/3pP3/1bPNnB1q/2NB2P1/PPQ2P1P/R3K2R b KQ - acd 51/62; acs 2982; acn 130400337k; ce -40; bm Qe7; pv Qe7 O-O Bxc3 bxc3 Ndc5 Be2 f6 exf6 Rxf6 Be5 Rf8 Rae1 Bd7 Nb3 Rad8 Nxc5 Nxc5 f4 dxc4 Bxc4 b5 Be2 Bc6 Bd4 e5 fxe5 Ne6 Rxf8+ Rxf8 Qb3 Kh8 Rf1 Ng5 h4 Rxf1+ Kxf1 Nf3 Be3 Nxe5 Qb4 Qf6+ Qf4 a6 Qxf6 gxf6 Bd4 Bd5 Bxe5 fxe5 a3 h6 Ke1 Kh7 Kf2;
r1b2rk1/pp1n1ppp/4p3/3pP3/1bPNnB1q/2NB2P1/PPQ2P1P/R3K2R b KQ - acd 51/57; acs 2982; acn 130400337k; ce -5; bm Qh5; pv Qh5 O-O g5 cxd5 gxf4 dxe6 fxe6 Nxe4 Nxe5 gxf4 Qg4+ Kh1 Qxf4 f3 Kh8 Rad1 b6 Bb5 Rg8 Nc6 Nxc6 Bxc6 Rb8 Ng3 e5 Be4 Bh3 Qc7 Qh6 Qxe5+ Qg7 Qxg7+ Rxg7 Rf2 a5 Nh5 Rf7 Rc2 Bc5 b3 Rg8 Ng3 Rg5 Rd8+ Kg7 Rc1 Rf8 Rxf8 Kxf8;

For this problem, both Be2 g4 are solutions because the pvs are permutations:
[fen]r2qr1k1/pp2bpp1/2pp3p/4nbN1/2P4P/4BP2/PPPQ2P1/1K1R1B1R w - -[/fen]
r2qr1k1/pp2bpp1/2pp3p/4nbN1/2P4P/4BP2/PPPQ2P1/1K1R1B1R w - - acd 50/55; acs 3355; acn 154809279k; ce -11; bm a3; pv a3 Bf6 Be2 Qe7 Ne4 Bxe4 fxe4 Nd7 Bf4 Nc5 Bxd6 Qxe4 Bf3 Qxc4 Bxc5 Qxc5 h5 Bg5 Qd3 Re3 Qd4 Qxd4 Rxd4 Re7 Rhd1 g6 Rd7 Rae8 c3 Kg7 Kc2 Kf6 R7d6+ Re6 Rxe6+ Rxe6 a4 Ke7 Bg4 f5 Bf3 gxh5 Bxh5 Re5 Kb3 Ke6 Bf3;
r2qr1k1/pp2bpp1/2pp3p/4nbN1/2P4P/4BP2/PPPQ2P1/1K1R1B1R w - - acd 50/62; acs 3355; acn 154809279k; ce 27; bm Be2 g4; pv g4 hxg5 Be2 Bg6 hxg5 d5 f4 Bxg5 fxg5 Be4 Rh2 Ng6 Rf1 Qd6 Rhf2 Rad8 Rxf7 d4 Bg1 c5 R7f2 b6 Bh2 Qd7 b3 d3 cxd3 Bxd3+ Bxd3 Qxd3+ Qxd3 Rxd3 Kb2 Re4 Bb8 Rd7 Ka3 Re8 Bf4 a6 Bh2 Re4 Rc2 Kh7 Rff2 Rxg4 Rcd2 Rgd4 Rxd4 cxd4 Bg1;
r2qr1k1/pp2bpp1/2pp3p/4nbN1/2P4P/4BP2/PPPQ2P1/1K1R1B1R w - - acd 50/62; acs 3355; acn 154809279k; ce 30; bm Be2 g4; pv Be2 hxg5 hxg5 d5 g4 Bg6 f4 Bxg5 fxg5 Be4 Rh2 Ng6 Rf1 Qd6 Rhf2 Rad8 Rxf7 d4 Bg1 c5 R7f2 Qd7 Bh2 d3 cxd3 Bxd3+ Bxd3 Qxd3+ Qxd3 Rxd3 b3 b6 Kc1 Re7 Kb2 Re4 Bb8 a6 Kc1 b5 cxb5 axb5 Rd2 Rxd2 Kxd2 c4 Kc3 cxb3 Kxb3 Rxg4 Rf5;

Thanks for this post, Dann!

I agree with you.
The positions Qd2 or Bc2 and Be2 or g4 seem to have equally good solutions. I took both positions from other test sets. I will remove these two positions for my next test set (I plan to implement new positions). In my tests almost all Stockfish clone engines and Stockfish dev have favored the moves Qd2 and Be2 and got a point in the test for them.

The queen move Qh5 is different. Qh5 is better than Qh3. If Black plays the best line after Qh3, then White is slightly better. But Black can make a lot of mistakes, see analysis below what happened dozens of times in practice. The position after Qh3 is harder to defend.

My analysis:

[pgn][Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "New game"]
[Black "?"]
[Result "*"]
[SetUp "1"]
[FEN "r1b2rk1/pp1n1ppp/4p3/3pP3/1bPNnB1q/2NB2P1/PPQ2P1P/R3K2R b KQ - 0 1"]
[PlyCount "22"]

{[#]} 1... Qh3 (1... Qh5 2. O-O g5 3. cxd5 $11 Bxc3 (3... gxf4 4. dxe6) 4. bxc3 exd5) 2. O-O-O Nxc3 3. bxc3 Ba3+ 4. Kb1 Nb6 5. Bf1 Qh5 6. Be2 Qg6 $14 (6... Qh3 7. Nf3 Bd7 (7... dxc4 8. Nh4 Nd5 9. Rd4 Be7 10. Bg5 Bxg5 11. Bf1 Qxf1+ 12. Rxf1 Be7 13. Rfd1 $16) (7... f6 8. exf6 gxf6 9. g4 $18) 8. Ng5 $16) 7. Bd3 f5 8. cxd5 Nxd5 9. Bc1 Bxc1 10. Rxc1 Bd7 11. Ka1 Rac8 12. Qb3 $14 *[/pgn]

[Sopel]

The hardware would be too weak for me personally because the engine has a slow search.

That's very interesting reasoning. I like it because it clearly tells people that you have no idea about how chess engines work. Do you have any data that would show this engine's strength scales differently with hardware than Stockfish? Because you are basically saying that it's worse at short time control compared to Stockfish, just because it reaches lower depths.

I am all the more surprised about all the results.

You didn't expect a Stockfish clone to be about as strong as Stockfish?