lichess.org
Donate

Stockfish, the endgame, and blunders

en.lichess.org/VKTpsNpi/black#93

Not sure what's going on with Stockfish here. I played 47 ... Ke7 and allowed a possible draw. The evaluation plummets from ca. - 40 to 0; this however doesn't count as a blunder in the analysis!

Stockfish actually suggests this move as being the best. However, if Black plays 47 ...Rd3, Black immediately has a winning endgame again. Stockfish apparently can't even see 1 move ahead in this relatively simple position ... are the analyses in more complicated positions to be trusted at all?
Ke7 is certainly not a blunder for black. After 47 ... Ke7 48. b5 Kd6 Stockfish rates the position at - 46. The reason why it is showing zero after 47. ... Ke7 and 48. b5 is not clear to me either. Seems like a mistake in the evaluation function.
According to tablebases, both moves win, but both require very accurate play to win. I have not looked at Stockfish's output in this position, but it probably has difficulties to spot an only move deeper in the search tree. However, in offline analysis with a little bit more time, it quickly finds that both moves are winning.
It actually seems to be a couple separate issues.

First, server-side analysis that you got from requesting a computer analysis of the game mistakenly evaluated the position as 0.00.

I'm not at all sure how that happened; at low (<=20) search depths my local SF 8 doesn't necessarily see that black is completely winning with Ke7, but the eval is never lower than 0.8

So, that first problem is unexplained, but definitely something strange happened with the server-side analysis.

Now, the SF8 you can run in the analysis board DOES see that Ke7 wins.

There's just a small glitch: if you only have the analysis set to show one line, it apparently now shows the evaluation from the server-side analysis, not the evaluation from your current analysis.

If you set multiple lines to anything >1, so that you get the small eval box for each line, you'll see that the SF 8 in the analysis board correctly shows a winning eval for both Rd3 and Ke7, but the big eval box still just shows the 0.00 from server-side analysis.

Seems like that's just a display glitch, hopefully easily fixed.
Did you let enough time for stockfish to assess the position? and maybe it uses different depths for board analysis than for game analysis.

This topic has been archived and can no longer be replied to.