"Best Move" Sequence Wrong - Suggested Engine Analysis Mistake?

Found a game analysis in which the suggested "best move sequence" of the engine makes no sense. Loaded the position seperately into the engine, and it gives very different best moves. Not a complex position either. Did you notice the engine giving bad move sequences as "best moves" before, or am I missing something?

Game: https://www.chess.com/analysis/game/live/82078861420?tab=review&move=31

Situation: On move 16, Qd6 - in the game analysis, it suggests knight to b6, which looks wrong, as the best move. Then in the follow up sequence it sacrifices the white queen, which seems totally unnecessary.

The strength the review is ran at will have an impact. I'm showing Qxd5 as the best reply to 16. ... Qd6

Shows Qxd5 for me in a seperate review, but as part of the "game review" process it shows the stupid knight move still ... I see no way to adjust the strength of that game review. So basically it's giving wrong advice, it's a quite bad move. And the position is not complex at all

My check was with the Maximum strength review via the website. It's possible a lower strength, or app review, gives a different suggestion.

Ah, I just realized what you mean. On the Fastest, Standard, and Deep review strength, after Nc6, it does suggest Nb6. That all comes done to how deeply the analysis is able to go at that level.

White is winning in both lines but apparently it takes the Maximum strength to evaluate Qd6 for black as being best.

Interesting yes, even on deep .. the position does not look complex, and Nb6 looks like an obvious blunder even to a beginner. Never saw such a blunder from the game review before, even at low depth, that's why I was so surprised.

The white queen is not sacrificed, you can win the black queen through a fork...

For clarity, I am referring to the sequence shown as "best moves" highlighted on the right. Why take the black rook with the white queen, for example. Just move the white queen out of the way. And why the first move of the black Knight to B6, which looses the black queen for a knight.

Here is the engine setting, how can it mess up such a simple move sequence so badly with supposed such high rating even on Standard? When I play a 2000 rated bot offline it never makes such blunders as that sequence shows. So low depth does not seem to be the full explanation?

When taking it to a seperate analysis board it shows the correct moves, for example for the white queen the following:

I just do not get how the Game Analysis Engine (top screenshot) can fail that badly in such a simple situation