Even if it's getting higher with long and slow enough Forward- Backward, but it has to be about 30, 40 moves long till SF sees the sure win for Black and to keep such long distance in hash takes much hardware time yet too.
And the nearer the tbs- positions come, the more SF- search tends to explode as for time to depth once in while again.
Again at starting position with 30 threads of 4.3GHz CPU, this time with 32G hash, 6men Syzygys as before too:
1r3rk1/p4pp1/8/2p1p1P1/8/P1P2B2/1P4PP/1bB1K2R w K - 0 1
Analysis by Stockfish dev-20260101-28844fc6:
20.Kf2 e4 21.Lh5 Ld3 22.Te1 g6 23.Lg4 f5 24.gxf6 Txf6+ 25.Kg3 a5 26.b4 axb4 27.axb4 cxb4 28.cxb4 Txb4 29.Lf4 Tb3 30.Lg5 Tfb6 31.Kf4 Tb1 32.Txb1 Txb1 33.h4 Kg7 34.Ke3 Tb3 35.Lh3 Ta3 36.Kd4 Ta4+ 37.Kd5 e3 38.Lxe3 Txh4 39.Lf2 Te4 40.Ld4+ Kf7 41.g3 Lc2 42.Le5 Te1 43.Kd4 g5 44.Lc7 Kg6 45.Lc8 Ta1 46.Ld6 Ta4+ 47.Ke5
Tiefe: 58/106 00:14:16 24741MN, tb=20094864
-/+ (-1.39 --)
And still too low yet.
Anyhow just after copying that, SF changes its mind to:
1r3rk1/p4pp1/8/2p1p1P1/8/P1P2B2/1P4PP/1bB1K2R w K - 0 1
Analysis by Stockfish dev-20260101-28844fc6:
20.Ld5
Tiefe: 58/118 00:21:14 36238MN, tb=28256165
-/+ (-1.23 ++)
That won't stay up probably for long, but I'm just extending edit- time already again. 20.h4 and 20.Be2 was up before copying the longer line too, at theese occasions I always made time it would have been kept at standalone- pondering probably shorter by going into the alternative lines. Guess I'll do so this time again, but won't show new output then again, if it doesn't bring new insights.