Stockfish-10.js

There seems to be a severe bug in version 10: the scores of the computer analysis of a game change rapidly when moving 1 ply ahead having nothing with mistakes. This creates a visual image of a jigsaw.
(example: lichess.org/P1jXqJIH/black)

Which moves do you think are wrongly considered to be errors?
Those are indeed masters playing, but in a 1+0 game it is normal to have quite a few errors.

Sorry for not getting a clear example. Here is a clear one (lichess.org/3FuxKxSV#82).
42.hxg6 the score drops from 13 to 9.7 even though this is the best move by Stockfish 10 on my computer. The same with 42...Kg7 (the score drops to 61 though this is the best move).
The 61 score is the score of the 2nd best-move 42...,fxg6??.

31...Ra5 is again the best move by Stockfish 10 (starting from low depths) but the js version gives a drop of 0.5 (which might again be the score of the 2nd best move)

One can find more examples of this phenomenon in other moves of this game.
Ofer

maybe the clearest example is 36.d6 which is best by Stockfish-10 but the scor drops from 6.3 to 4.0

I dwelled a little deeper into this issue. It seems that the problem does not stem from js version but from Stockfish version 10 itself.
For example, in this position (7k/p2qQ1p1/P1pP1r1p/1p3p2/4p3/2P4P/5PP1/3R2K1 b - - 2 2) Stockfish 10, depth=14 (single thread) approves 2…QxQ and it needs go further to prefer 2…Rf7 while Stockfish 8 gets the correct answer in lower depth.
Even worse, White’s answer to 2…Rf7 (7k/p2qQrp1/P1pP3p/1p3p2/4p3/2P4P/5PP1/3R2K1 w - - 3 3) by version 10, depth=12 is 3.Qe5?? while version 8 has no problem finding the correct move 3.QxR.