Endgame that Stockfish DD does not evaluate correctly

[beachknight]

15:
8/8/3p1kb1/2pP4/1pPp2p1/1P1P2Pp/4BP1K/8 w - - 0 1

Analysis by Stockfish 180114IP 64:

45.Kg1 Kg7 46.Kh2 Bh7 47.Bf1 Kh6 48.Be2 Bf5 49.Kg1 Kg5 50.Kh2 Bh7 51.Kg1 Kh5 52.Kh2 Bf5 53.Kh1 Bg6 54.Kh2 Kg5 55.Bf1 Bh7 56.Be2 Bf5 57.Kg1 Bg6 58.Kh2 Be8 59.Bd1 Bd7 60.Be2 Kf5 61.Bd1 Ke5 62.Be2 Kf6 63.Kh1 Bc8 64.Kg1 Kg5 65.Kh2 Bd7 66.Bd1 Kf5 67.Be2 Kg5
-/+ (-0.78) Depth: 61/61 00:00:55 179MN
(Chess Engines Testing, 22.01.2014)

[beachknight]

15:
8/8/3p1kb1/2pP4/1pPp2p1/1P1P2Pp/4BP1K/8 w - - 0 1

Analysis by Stockfish 180114IP 64:

45.Kg1 Kg7 46.Kh2 Bh7 47.Bf1 Kh6 48.Be2 Bf5 49.Kg1 Kg5 50.Kh2 Bh7 51.Kg1 Kh5 52.Kh2 Bf5 53.Kh1 Bg6 54.Kh2 Kg5 55.Bf1 Bh7 56.Be2 Bf5 57.Kg1 Bg6 58.Kh2 Be8 59.Kg1 Bd7 60.Kh2 Kf5 61.Bd1 Ke5 62.Be2 Kf6 63.Kh1 Bc8 64.Kg1 Kg5 65.Kh2 Bd7 66.Bd1 Kh6 67.Be2 Bf5
-/+ (-0.78) Depth: 63/65 00:01:30 298MN
(Chess Engines Testing, 22.01.2014)

[beachknight]

15:
8/8/3p1kb1/2pP4/1pPp2p1/1P1P2Pp/4BP1K/8 w - - 0 1

Analysis by Stockfish 180114IP 64:

45.Kg1 Kf7 46.Bxg4 Bxd3 47.Bh5+ Kf6 48.Bd1 Be4 49.Kh2 Bf5 50.f4 Bb1 51.Kxh3 d3 52.g4 Bc2 53.g5+ Kg7 54.Bxc2 dxc2
-/+ (-0.82 --) Depth: 64/73 00:06:22 1394MN
(Chess Engines Testing, 22.01.2014)

[beachknight]

15:
8/8/3p1kb1/2pP4/1pPp2p1/1P1P2Pp/4BP1K/8 w - - 0 1

Analysis by Stockfish 180114IP 64:

45.Kg1 Kf7 46.Kh1 Kg7 47.Kh2 Bh7 48.Bf1 Kh6 49.Be2 Bf5 50.Kg1 Kg5 51.Kh2 Bh7 52.Kg1 Kh5 53.Kh2 Bf5 54.Kh1 Bg6 55.Kh2 Kg5 56.Kg1 Bh7 57.Kh2 Bg6
-/+ (-0.78) Depth: 65/77 00:08:22 1653MN
(Chess Engines Testing, 22.01.2014)

[beachknight]

15:
8/8/3p1kb1/2pP4/1pPp2p1/1P1P2Pp/4BP1K/8 w - - 0 1

Analysis by Stockfish 180114IP 64:

45.Kg1 Kf7 46.Kh1 Kg7 47.Kh2 Bh7 48.Bf1 Kh6 49.Be2 Bf5 50.Kg1 Kg5 51.Kh2 Bh7 52.Kg1 Kh5 53.Kh2 Bf5 54.Kh1 Bg6 55.Kh2 Kg5 56.Kg1 Bh7 57.Kh2 Bg6
-/+ (-0.78) Depth: 66/77 00:09:45 1938MN
(Chess Engines Testing, 22.01.2014)

[beachknight]

15:
8/8/3p1kb1/2pP4/1pPp2p1/1P1P2Pp/4BP1K/8 w - - 0 1

Analysis by Stockfish 180114IP 64:

45.Kg1 Kf7 46.Kh1 Kg7 47.Kh2 Bh7 48.Bf1 Kh6 49.Be2 Bf5 50.Kg1 Kg5 51.Kh2 Bh7 52.Kg1 Kh5 53.Kh2 Bf5 54.Kh1 Bg6 55.Kh2 Kg5 56.Kh1 Be8 57.Kh2 Bd7 58.Bd1 Kg6 59.Be2 Bf5 60.Kg1 Kf7 61.Kh2 Kf6 62.Kh1 Bh7 63.Kg1 Kg5 64.Kh2 Bg6
-/+ (-0.78) Depth: 68/77 00:12:39 2524MN
(Chess Engines Testing, 22.01.2014)

[beachknight]

15:
8/8/3p1kb1/2pP4/1pPp2p1/1P1P2Pp/4BP1K/8 w - - 0 1

Analysis by Stockfish 180114IP 64:

45.Kg1 Kf7 46.Kh1 Kg7 47.Kh2 Bh7 48.Bf1 Kh6 49.Be2 Bf5 50.Kg1 Kg5 51.Kh2 Bh7 52.Kg1 Kh5 53.Kh2 Bf5 54.Kh1 Bg6 55.Kh2 Kg5 56.Kh1 Be8 57.Kh2 Bd7 58.Bd1 Kg6 59.Be2 Bf5 60.Kg1 Kf7 61.Kh2 Kf6 62.Kg1 Ke5 63.Kh2 Bg6 64.Kg1 Bh7 65.Kh2 Kf6 66.Kg1 Kg5 67.Kh2 Bg6
-/+ (-0.78) Depth: 71/77 00:13:29 2696MN
(Chess Engines Testing, 22.01.2014)

[beachknight]

15:
8/8/3p1kb1/2pP4/1pPp2p1/1P1P2Pp/4BP1K/8 w - - 0 1

Analysis by Stockfish 180114IP 64:

45.Kg1 Kf7 46.Kh1 Kg7 47.Kh2 Bh7 48.Bf1 Kh6 49.Be2 Bf5 50.Kg1 Kg5 51.Kh2 Bh7 52.Kg1 Kh5 53.Kh2 Bf5 54.Kh1 Bg6 55.Kh2 Kg5 56.Kh1 Be8 57.Kh2 Bd7 58.Bd1 Kg6 59.Be2 Bf5 60.Kg1 Kf7 61.Kh2 Kf6 62.Kh1 Kg5 63.Kh2
-/+ (-0.78) Depth: 72/77 00:19:08 3891MN
(Chess Engines Testing, 22.01.2014)

[Lyudmil Tsvetkov]

Thanks Miroslav.
I have played many similar positions with many engines, but as I have never looked at the eval, I did not know they evaluated it in this way.

Here the problem is not the h3 black passer at all, but rather the ability of the weaker side to build a fortress. White does not even need its f2 pawn, and it is still a draw because of the fortress: b,c,d, g files are closed with blocked pawns, a4 is controlled by b3, e4 by d3, f4 by g3 and h4 by g3, so that the black king can not enter enemy territory, without which the game is drawn. White can play only Kh2-g1-h2-g1, etc.

I think Louis posted a while ago some positions where Critter recognized such simple endgame fortress patterns (at least for pawn endings), but Stockfish probably does not have that code.

I wonder if Stockfish recognizes instantly the below position as a draw:

[d]8/8/3p1k2/2pP4/1pPp2p1/1P1P2Pp/8/6K1 w - - 0 1

Same structure, only pawns. If it recognizes this as a draw, there are big chances it will also recognize the structure with bishops present. Problem is if this is measurable in engine-engine matches, as similar techniques are mostly used by humans, but I think the rule is helpful. Thanks again for the game.

[Lyudmil Tsvetkov]

15:
8/8/3p1kb1/2pP4/1pPp2p1/1P1P2Pp/4BP1K/8 w - - 0 1

Analysis by Stockfish 180114IP 64:

45.Kg1 Kf7 46.Kh1 Kg7 47.Kh2 Bh7 48.Bf1 Kh6 49.Be2 Bf5 50.Kg1 Kg5 51.Kh2 Bh7 52.Kg1 Kh5 53.Kh2 Bf5 54.Kh1 Bg6 55.Kh2 Kg5 56.Kh1 Be8 57.Kh2 Bd7 58.Bd1 Kg6 59.Be2 Bf5 60.Kg1 Kf7 61.Kh2 Kf6 62.Kg1 Ke5 63.Kh2 Bg6 64.Kg1 Bh7 65.Kh2 Kf6 66.Kg1 Kg5 67.Kh2 Bg6
-/+ (-0.78) Depth: 71/77 00:13:29 2696MN
(Chess Engines Testing, 22.01.2014)

It seems that Arjun guessed wrong: Stockfish 1801 has a more realistic eval even in this endgame.

Thanks, Harun, for the output.