Endgame

[Igor_Goldenberg]

When I analysed my last game with Houdini Engine it estimated a resulting endgame as almost equal. It was, however, already lost for black at the point.

8/5p1p/p1k1b1p1/1p1p2P1/8/PP2PP2/4BK1P/8 w - - 0 4

Here is the endgame

1.g5 Nc6 2.Nxc6 Rxc6 3.Rxc6 Kxc6 4.b4 Kd6 5.Bd3 Ke5 6.f4+ Kd6 7.Ke2 Bd7 8.Kd2 d4 9.exd4 Bc6 10.Ke3 Bd7 11.Be4 Bc8 12.h4 Ke6 13.Bc2 Bb7 14.Bb3+ Ke7 15.f5 f6 16.Kf4 fxg5+ 17.hxg5 gxf5 18.Kxf5 Kd6 19.Bg8 Bg2 20.Kf4 Bf1 21.Ke3 Bh3 22.Bxh7 Be6 23.Be4

[Kevin Bonham]

Interesting that strong programs can still be weak in such positions. Stockfish 4 on my computer took 30 seconds to have the swap-off on move 2 as the best option with an initial evaluation of +0.48. After playing the swap-off its evaluation rose to +0.9 in another 30 seconds.

[Max Illingworth]

Houdini is notorious for misevaluating endgame positions (giving +0.5 for lost endgames) - Komodo and particularly Stockfish are more reliable. With these engines, if the evaluation number is not changing at all after playing the best line for several moves, it's either a fortress, or the engine doesn't understand what is happening. :)

[Igor_Goldenberg]

Perhaps the length of seemingly unforced variation might still pose problem with some engines.