Massive blunders

Why, if one player has a +10 advantage or so, and makes an error which does not have much effect on the game because the opponent has few major pieces left, does Stockfish classify it as such a big mistake that the advantage falls to 0? When the opponent moves, that move is a blunder and the advantage is restored, Could there have been three-fold repetition which the opponent failed to notice, and so did not claim a draw? Noticed in this game http://en.lichess.org/t6n05sif#2

no because it was a blunder... and leads to a mate. but opponent replied it with another blunder... check variation lines given by sf.

Hi everyone,

I believe Lichess either uses an old version of the Stockfish which may have a sort of a bug or its analysis engine incorrectly treats Stockfish's results in some particular cases.

For http://en.lichess.org/t6n05sif#35 , the Stockfish 5 x64 that I'm using on my laptop clearly shows the tremendous advantage of white after the move "18. Ne5+".

For instance, at depth 25 the score is +14.12 with PV:
Kd6 Nf7+ Kc6 c3 Nb5 g3 Nf5 a4 Nc7 Ne5+ Kd6 Nxc7 Kxc7 Bxh7 e6 Kc2 Ne7 Re1 Nd5 Bg5 Bd6 Bg6 Ne7 Bf7 Bxe5 Rxe5 Nc6 Ree1 e5 Bg6 Bg4 f4 exf4 Bxf4+ Kb6 Be3+ Kc7

It has nothing to do with the version of Stockfish.

Your Stockfish on your laptop will also show 18...Kd8 to be 0.00.

It's a known issue with SF where it counts the second occurrence of a position as 0.00 (here the position after 18...Kd8 would be the same as after 16...Kh4).

You probably just set up that position and analyzed, in which case SF is blind to the first occurrence of the position, and thus gives the correct huge positive score.

Input the whole PGN, and play through the game, and after 18.Ne5+ it will show 18...Kd8 as 0.00.

This has been discussed in other threads here, and I posted a link to the discussion about the bug in the fishcooking forum.

groups.google.com/forum/#!msg/fishcooking/ZAgj3bLIlEU/lTGSjHyl2gMJ

Basically, the SF developers weren't so concerned, because when SF is +15, it's not going to repeat a position anyway, so the bug doesn't affect gameplay, only analysis.

Also, when someone actually got around to trying to push a patch for the bug through the testing framework, it actually got rejected for resulting in a -1.5 ELO change (maybe due to the very slight increase in speed of detecting repetitions as draws when the second occurrence is counted as 0.00).

Either way, it's known behavior, and affects all versions of SF, unless you tweak the code and compile yourself :)

Which I do already. Maybe I should include that patch?

That would be really cool :)

I wouldn't mind having that change made in my project, either. I'm not following on how it was done, but I wouldn't mind granting commit permissions on my repo to anyone who can figure it out!
My repo @ http://github.com/ddugovic/stockfish

Perhaps, I'm doing something wrong, but I don't see 0 in SF while going through PGN:
http://screencast.com/t/11MXndTpeCs

Well, that screenshot doesn't tell me much about the analysis :)

So, you play through the PGN with SF analyzing, and after 18.Ne5+ is played, it doesn't show 18...Kd8 0.00?

I've tried it with several different SFs, and they all behave as expected.

Are you sure you're using SF and not some other engine for analysis? :)

Perhaps a screenshot of the actual analysis and not just the PGN would shed some more light :)

OneOfTheQ,

It's not a screenshot. It's video. Just click it to play :)