Not the solution but modified Rebel can find Th3 (Eigenmann)

[Eelco de Groot]

(Referring to this thread for background.)


[fen]7k/p1p3p1/1p1p3p/6B1/2P5/q1PP4/7R/5RK1 w - -[/fen]

Engine: Rebel MP 14.2 MOD (1 thread, 512 MB)
gemaakt door Fabien Letouzey, Thomas Gaksch, Jerry Donald Watson, Chris Whittington and Ed schroder

12/34 0:01 +0.01 1.Lc1 Dc5+ 2.Tff2 a5 3.Th3 a4 4.Le3 Da5
5.Tg2 Df5 6.Thg3 Dxd3 7.Ld4 Dd1+
8.Kh2 Dh5+ 9.Th3 (496.358) 379

13/34 0:02 +0.17++ 1.Lc1 Dc5+ 2.Tff2 a5 3.Th3 a4 4.Le3 Da5
5.Tg2 Df5 6.Thg3 Dxd3 7.Ld4 Dd1+
8.Kh2 c6 9.Lxb6 (767.805) 381

14/34 0:03 +0.01-- 1.Lc1 Da5 2.Tf8+ Kh7 3.Ld2 Dc5+
4.Kf1 a5 5.Tf3 a4 6.Ke2 a3 7.Tf7 (1.210.914) 386

14/42 0:07 +0.04 1.Th3 a5 2.Le3 a4 3.Tf8+ Kh7 4.Tf7 Da1+
5.Kg2 Db2+ 6.Kf3 Dxc3 7.Txc7 Dxd3
8.Tg3 Dd1+ 9.Kg2 Dc2+ 10.Kh3 Df5+
11.Kh2 De5 12.Tcxg7+ Dxg7 13.Txg7+ Kxg7
14.Lxb6 (2.962.124) 401

15/42 0:10 +0.20++ 1.Th3 a5 2.Le3 Dxc3 3.Tf8+ Kh7 4.Tf7 Kg8
5.Txc7 a4 6.Tg3 a3 7.Tcxg7+ Dxg7
8.Txg7+ Kxg7 9.Lxb6 Kg6 10.Kg2 Kf5 (4.183.788) 407

16/42 0:14 +0.36++ 1.Th3 a5 2.Le3 Dxc3 3.Tf8+ Kh7 4.Tf7 Kg8
5.Txc7 a4 6.Tg3 a3 7.Tcxg7+ Kf8
8.Kh2 a2 9.Lxh6 Db2+ 10.Tg2 De5+
11.T7g3+ Ke7 12.Txa2 (5.976.811) 418

17/50 0:24 +0.20-- 1.Th3 a5 2.Le3 Dxc3 3.Tf8+ Kh7 4.Tf7 Kh8
5.Txc7 a4 6.Tg3 a3 7.Tcxg7 a2 8.Tg8+ Kh7
9.T3g7+ Dxg7+ 10.Txg7+ Kxg7 11.Ld4+ Kg6
12.Kg2 Kf5 13.Kg3 h5 14.La1 (10.265.889) 427

17/50 0:33 +0.04-- 1.Th3 a5 2.Le3 Dxc3 3.Tf8+ Kh7 4.Tf7 Kh8
5.Txc7 a4 6.Tg3 a3 7.Tcxg7 a2 8.Tg8+ Kh7
9.T3g7+ Dxg7+ 10.Txg7+ Kxg7 11.Ld4+ Kg6
12.Kg2 Kf5 13.Kg3 h5 14.La1 (14.103.567) 419

17/50 0:35 +0.22 1.Lc1 Dc5+ 2.Tff2 De5 3.Thg2 a5
4.Te2 Df6 5.Lb2 Df3 6.c5 dxc5 7.d4 (14.749.604) 418

18/50 0:37 +0.22 1.Lc1 Dc5+ 2.Tff2 De5 3.Thg2 a5
4.Te2 Df6 5.Lb2 Df3 6.c5 dxc5 7.d4 (15.774.258) 418

19/51 0:50 +0.36 1.Lc1 Dc5+ 2.Tff2 De5 3.Thg2 Kh7
4.Ld2 Dc5 5.Tg3 a5 6.Le3 De5 7.Kh2 Dxc3
8.Tf7 De5 (21.078.180) 415

20/51 1:13 +0.20-- 1.Lc1 Dc5+ 2.Tff2 De5 3.Thg2 a5
4.Te2 Dh5 5.Le3 Kg8 6.Ld4 g5 7.Te7 Dd1+
8.Kh2 Dh5+ 9.Kg3 Dh4+ 10.Kf3 Dh3+
11.Tg3 Df1+ 12.Lf2 Dxd3+ 13.Kg2 Dxc4
14.Te8+ (30.108.298) 410

20/52 1:51 +0.36++ 1.Lc1 Dc5+ 2.Tff2 De5 3.Thg2 a5
4.Ld2 Dc5 5.Tg3 a4 6.Le3 De5 7.Kg2 Dxc3
8.Tf8+ Kh7 9.Tf7 c5 10.d4 Kh8
11.Tfxg7 Db2+ 12.Kf1 Db1+ 13.Kf2 Dc2+
14.Kg1 (45.792.641) 409

21/53 2:04 +0.11-- 1.Lc1 Dc5+ 2.Tff2 De5 3.Thg2 Kh7
4.Ld2 De6 5.Te2 Dd7 6.Le3 a5 7.Ld4 g5
8.Kh2 Df7 9.Kh1 Dh5+ 10.Kg1 Kg8 (51.033.237) 409

21/53 3:25 +0.50 1.Th3 a5 2.Kg2 Kh7 3.Le3 a4 4.Tf7 Db2+
5.Lf2 Kg6 6.Txc7 a3 7.Tg3+ Kf6
8.Tgxg7 a2 9.Kf3 Dxf2+ 10.Kxf2 a1D
11.Tgf7+ Kg5 12.Tcd7 Dxc3 13.Txd6 Dd2+
14.Kg3 (85.207.025) 415

22/55 4:36 +0.41 1.Th3 a5 2.Kg2 Kh7 3.Le3 a4 4.Tf7 Db2+
5.Lf2 Kg6 6.Txc7 a3 7.Tg3+ Kf6
8.Tf3+ Kg6 9.Kh3 De2 10.Tg3+ Kh5
11.Kg2 g5 12.Te3 Db2 13.Kg3 (117.400.544) 424

23/69 9:03 +0.57++ 1.Th3 a5 2.Kg2 Kh7 3.Le3 a4 4.Tf7 Db2+
5.Lf2 Kg6 6.Txc7 a3 7.Tg3+ Kf6
8.Tgxg7 a2 9.Kh3 Dxf2 10.Tcf7+ Ke6
11.Te7+ Kf5 12.Tgf7+ Kg6 13.Txf2 a1D
14.Te6+ (238.234.183) 438

24/70 23:46 +0.64 1.Th3 a5 2.Kg2 Kh7 3.Le3 Dxc3 4.Tf7 Kh8
5.Tg3 Dxd3 6.Txc7 De4+ 7.Kf2 Dc2+
8.Kf3 Dd1+ 9.Kg2 Dc2+ 10.Lf2 De4+
11.Kf1 Db1+ 12.Le1 Df5+ 13.Kg1 De5
14.Lf2 (637.148.151) 446


Considering that from the previous versions, and there were many, I did not see a single score that at least looked promising, and sometimes the move Ld2 which does not win, this is still a modest progress. Credits go to Ed for all the NNUE tuning effort! I did not try default Rebel yet I must say. I only made some small changes, and they are not tested. NPS is still low, which is a pity but thanks to Ed and Chris' tips at least I have a project and a working compile!

Prefer the simple Rebel MP as a name myself, the credits are already in the names of several of the previous authors, beginning with Fabien of course. Only there is not enough room for all of them if you post the Shredder GUI output. It stops after the D from Donald.

With a small change, not including line 1027 history_good(best_move,board,ThreadId); (this line caused problems in my version of 14.1, but I think it is fixed now by using Unicode.)
Rebel will play Lxh6 for a long time but not with more than a drawscore. So there is a random factor here, but I decided to activate the line again because finding a score of about half a pawn is more important than Lxh6 with a drawscore.