Sawtooth evaluation?

Just recently, when clicking on "computer analysis" after a game, instead of a relatively smooth progression (particularly for opening moves) the evaluation seems to jump each by about 0.2-0.3 pawns in favour of the player to move. For example:




I'm sure I saw something similar a while back - has the depth of the evaluation been dialled down or something?

could be, honestly sometimes the computer analysis makes mistakes even, when all it needs to see is like 4 moves deep. Also, opening theory is some really heavily analysed stuff, so the light evaluation that the computer gives is not that deeply calculated and can be misleaidng.

#2 @IsolatedPun-Youtube it's not that it makes mistakes, it's that the evaluation changes fairly significantly move to move, which I don't think happened to the same extent before.

Here's an example game from a few weeks ago with a smoother evaluation for the first few moves:

@FletchMK64 Thanks for reporting. This is a real bug in our Stockfish fork (related to Contempt). I submitted a fix for discussion: github.com/ddugovic/Stockfish/pull/535

@revoof Thanks, I have accepted this fix.

Hi - still seem to be seeing this behaviour - this was a particularly obvious example.

lichess.org/bTn7owXG/

Yeah. A fixed Stockfish version is released now, but so far only half of the servers are updated. Also still have to update client side Stockfish.