In my new idea I tested such an interesting position as a template:
[d]7r/p3k3/2p5/1pPp4/3P4/PP4P1/3P1PB1/2K5 w - - bm Kd1;
I set wrong and risky code only to solve this position and to avoid possibilities to find earlier a quasi-fortress blocking the tough solution here.
example of possible quasi-fortress in this position:
[d]8/p7/2p5/1pPp2k1/1P1P4/P7/1KBP4/7r b - -
It is a win, but it needs mutch higher depths. Why? I will take a look at this too.
Such a wrong position can block the correct move Kd1 for a very long time and search needs a very strong machine to reject it.
Now I set some better, logical and safe values (code) for my idea.
It is possible to do much more safe (~ the same ELO as the version 4.7) but I always remember about users of very old machines.
Version 4.8.3 should be soon. Give me a few days.
I am sorry for all who tested version 4.8.2 .
BTW. Newest Stockfish is very fast and waits for ideas of Sting.
For 5 .. 4 .. 3 .. 2 ..