Engine Blind Spot

[yanquis1972]

i'm not sure what you're talking about. single core rybka 2.3.2a sees g6 loses very quickly

[+2.06] d=13 21...g6 22.Qh4 Qd4 23.e5 f6 24.Bxg6 Qc5 25.Rhg1 Kh8 26.exd6 Ra7 (0:00:07) 733kN
[+0.61] d=12 21...g6 22.Qh4 Kh7 23.Qe7 e5 24.Rhg1 Qa5 25.f5 Qd8 (0:00:03) 328kN
[+0.52] d=11 21...g6 22.Qh4 Kh7 23.Qe7 e5 24.Rhg1 Qa5 25.f5 (0:00:02) 252kN
[+0.39] d=10 21...g6 22.Qh4 f6 23.Rdg1 Bb7 24.Rxg6 Kh8 25.Qe1 (0:00:01) 154kN
[+0.20] d=9 21...g6 22.Qh4 f6 23.Qg3 Kh7 24.e5 (0:00:01) 107kN
[+0.14] d=8 21...g6 22.Qh4 Qc5 23.e5 Qc6 (0:00:00) 35kN
[+0.00] d=7 21...g6 22.Qh4 Qc5 23.e5 (0:00:00) 23kN
[-0.14] d=6 21...g6 22.Qh4 Qd4 (0:00:00) 9kN
[-0.47] d=5 21...g6 22.Qh4 (0:00:00) 4kN

[Martin Brown]

i'm not sure what you're talking about. single core rybka 2.3.2a sees g6 loses very quickly

[+2.06] d=13 21...g6 22.Qh4 Qd4 23.e5 f6 24.Bxg6 Qc5 25.Rhg1 Kh8 26.exd6 Ra7 (0:00:07) 733kN
[+0.61] d=12 21...g6 22.Qh4 Kh7 23.Qe7 e5 24.Rhg1 Qa5 25.f5 Qd8 (0:00:03) 328kN
[+0.52] d=11 21...g6 22.Qh4 Kh7 23.Qe7 e5 24.Rhg1 Qa5 25.f5 (0:00:02) 252kN
[+0.39] d=10 21...g6 22.Qh4 f6 23.Rdg1 Bb7 24.Rxg6 Kh8 25.Qe1 (0:00:01) 154kN
[+0.20] d=9 21...g6 22.Qh4 f6 23.Qg3 Kh7 24.e5 (0:00:01) 107kN
[+0.14] d=8 21...g6 22.Qh4 Qc5 23.e5 Qc6 (0:00:00) 35kN
[+0.00] d=7 21...g6 22.Qh4 Qc5 23.e5 (0:00:00) 23kN
[-0.14] d=6 21...g6 22.Qh4 Qd4 (0:00:00) 9kN
[-0.47] d=5 21...g6 22.Qh4 (0:00:00) 4kN

I clearly didn't phrase this observation clearly. g6 and e5 are both losing moves for black but e5 is distinctly superior.

The strongest refutation of g6 by far is by 23. Qf6 Play out the first two moves of the PV that Rybka thinks is +2.06 at d=13 to see what I mean. Rybka, Shredder and most other engines fail to see that 23. Qf6 is devastating. It takes more than an hour for it to see this line from the start position. Of the engines I tried only Stockfish sees it correctly.

It is a simple position so it is a bit of a surprise at least to me.

Perhaps it would be less ambiguous to cast the problem as find the win for white after black has played 22. ... g6 Rybka here sees:

Ply 12: +1.87 22. Qh4 Qd4 23. e5 after 5 mins
Ply 13: +4.01 22. Qh4 f6 23. Rhg1 Kh8
Play Qh4 and watch how quickly it sees the right main line and 23. Qf6 and strangely even with the cache preloaded with the results of this search it still reverts to 23. e5 etc when you step back a ply.

BTW how do you get an engine to write the PV to a log file?

[yanquis1972]

you should just be able to right click in the analysis pane & click copy, clip, etc depending on the GUI.

anyway, i now understand what you're saying. however every single modern (ie latest cycle) engine i've tried switches off g6 in seconds. (this includes rybka 2.3.2)