Here's 3MV- output in depth 47 after about one hour of pondering with 24 threads of a 12x3GHz CPU and 32G hash, Contempt 0AdminX wrote: ↑Mon Jan 20, 2020 5:48 pmI looked at that but felt it was a bit premature, I will take a second look at it however.peter wrote: ↑Mon Jan 20, 2020 5:37 pmBefore Louis gives his best move, here's mine: ...Bxh2+

r2qk2r/p3n1pp/2pb4/2pp3b/N3p1P1/BP1PP3/P1P2P1N/R2Q1RK1 b kq - 0 1

Analysis by Stockfish 11 64 POPCNT:

1. -+ (-9.85): 14...Bxh2+ 15.Kxh2 Ng6 16.Rh1 Ne5 17.gxh5 Qg5 18.Bxc5 Ng4+ 19.Qxg4 Qxg4 20.Rhg1 Qxh5+ 21.Kg2 0-0-0 22.Nc3 Rhe8 23.Kf1 exd3 24.cxd3 d4 25.Bxd4 Rxd4 26.exd4 Qh3+ 27.Rg2 Qxd3+ 28.Kg1 Qxc3 29.Rd1 Re7 30.Rg3 Qc2 31.Ra1 Qb2 32.Rf1 Qxa2 33.b4 Qe2 34.Rc1 Rf7 35.Rg2 Kb7 36.Rf1 Qe4 37.Rg3 Kc8 38.Re3 Qg4+ 39.Rg3 Qxd4 40.Re3 Qh4 41.Rg3 Qxb4 42.Re3 Qh4 43.Rg3 Qf6 44.Re3 h5 45.Kg2 Qg5+ 46.Rg3 Qd5+ 47.Kg1 Qd4 48.Rg2 h4

2. -+ (-5.26): 14...Nf5 15.f4 exf3 16.Qxf3 Qh4 17.Qg2 Bxh2+ 18.Qxh2 Qxg4+ 19.Qg2 Qxg2+ 20.Kxg2 Nxe3+ 21.Kh2 Nxc2 22.Bb2 Nxa1 23.Bxg7 Rg8 24.Bxa1 0-0-0 25.Nxc5 Rde8 26.Rf2 Rg6 27.Bd4 Rg4 28.Bf6 Rf8 29.Rf5 h6 30.Ne6 Rf7 31.Nd8 Rxf6 32.Rxf6 Kxd8 33.Rxc6 Rg6 34.Rc5 Bf3 35.Rc2 Kd7 36.Rf2 Bd1 37.d4 Ke6 38.Kh3 Rf6 39.Rd2 Bf3 40.Kg3 Be4 41.b4 Rg6+ 42.Kh4 Rg1 43.b5 Kf5 44.Rf2+ Kg6 45.Rd2 Rh1+ 46.Kg3 Kg5 47.Rh2

3. -/+ (-1.54): 14...Bg6 15.Bxc5 h5 16.g5 Nf5 17.Bxd6 Qxd6 18.dxe4 Nxe3 19.e5 Qxe5 20.Nf3 Qf4 21.fxe3 Qg3+ 22.Kh1 Qh3+ 23.Kg1 0-0 24.Qe2 Be4 25.Qg2 Rxf3 26.Rxf3 Qxf3 27.Qxf3 Bxf3 28.Nc5 Re8 29.g6 Rxe3 30.Rf1 Be4 31.Nxe4 Rxe4 32.Rf5 Rg4+ 33.Kf1 Rxg6 34.Rxh5 Rh6 35.Re5 Rh2 36.c4 dxc4 37.Rc5 Rxa2 38.Rxc4 Ra6 39.Kf2 Kf7 40.b4 Rb6 41.Ke3 Kf6 42.Rf4+ Ke5 43.Rg4 Rb7 44.Re4+ Kf5 45.Rc4 Rc7 46.Rc1 Ke5 47.Rc5+ Kd6 48.Rg5 Rb7 49.Rg4 Kd5 50.Rd4+ Ke6 51.Re4+ Kd7 52.Rg4 Kd6 53.Rg6+ Kd5 54.Kd3 Re7 55.Rg5+ Ke6 56.Rc5

BTW, to make LC0 "see" the real deal of 14...Bxh2+ more quickly than by stand- alone- pondering at starting position only, give in the moves of the output- line above up to 17...Qg5, that will make eval climb more quickly.

Here's LC0 (net 49921)- output after that (14...Bxh2+ 15.Kxh2 Ng6 16.Rh1 Ne5 17.gxh5 Qg5):

Analysis by Lc0 v0.23.2 2x 2080Ti syzygy:

18.Qe2 Nf3+ 19.Kh3 0-0 20.Qd1 Rf5 21.dxe4 dxe4 22.Qd7 Qxh5+ 23.Kg2 Rg5+ 24.Kf1 Qxh1+ 25.Ke2 Qxa1 26.Bc1 Qe5 27.Qd3 Ng1+ 28.Kf1 h6 29.f4 exf3 30.Qc4+ Kh8 31.Qg8+ Kxg8

-+ (-9.18) Depth: 24/51 00:01:24 1958kN

Pity, Backward with LC0 still is some more time- consuming than with SF as for moves, that it doesn't favour on its own. Give it a try and compare.