Chess.com analysis engine suggests big blunder as the best move

Having hard time to understand, any ideas the reasoning behind it?
What should i change in analysis settings to avoid such confusion - calling a big blunder (leading to mate) the best move?

Thank you in advance

Initial position in question

next move

evaluation after 'the best move suggested Nxd7' -pretty straightforward forced mate in 6

Diamond members are supposed to be able to set Report run depth at 30. (There is a reporting settings gear icon under the depth=20 on the report.)

I would be curious to see if the depth change picks up the mate.

short version - at will take a lot of time )

long version - the depth to the mate is only 7 moves, my guess the problem with cutting-off the branch with rook sacrifice on g7, not with depth of analysis. Maybe with increasing depth of analysis starting tree became wider, I don't know, but for me any depth above default value is overkill. I'm only 1400 elo, all I want is to learn from my mistake, I can't extract useful knowledge from the analysis-depth-20 vs analysis-depth-30, if analysis-depth-20 works as it supposed to be

update
run depth=30 analysis, took 8 mins on 21-move game
engine spotted the blunder and there are a lot of changes compare to the depth by default, for example, my agreement with computer jump from 88 to 94, default depth marked 3 my moves as mistake, depth-30 remarked one of ‘mistake’ as the best move (you can see on my prev ss number 3 on retry section and now only 2).

I don't want to spend so much time on running analysis, but same time default setting in this game will teach me wrong, correct 'mistake' in Retry section, which turn out to be the best move, thats sad

I don't think it's sad at all. It's just a tool.

It's what you paid for when you bought the diamond membership.

Back before we had microscopes, nobody believed in germs because they couldn't see them. 

10 years ago, computer analysis was much more limited. 

I am surprised that the depth=30 analysis took so long. I thought that it would be faster on the chess.com servers.

It runs on chess.com servers (there is no CPU load on PC during analysis at all)
Out of curiosity ran analysis of 64-move game with depth=30, took 12 min. From 0% completion to 93% - 4min, 93%->100% - 8 min, it provides an impression of speed at start, but false one.

I paid for subscription mainly to see no ads and to support good project, the only perks I appreciate is unlimited puzzles. Today I’ve got some evidence that default analyzer makes rookie mistakes which can have negative impact on my learning. With Jonh’s (notmtwain) help I also found that with diamond membership I have a privilege !?? to spend 8-12 mins of my life (for each game) to fix this issue.

Maybe there is another solution for ordinary players, how to get decent analysis on chess.com fast?

Engines need time to get good results and in a lot of positions need to go deep to find the best continuation. The more complex the position, the longer and deeper the engine needs to look and sometimes will still miss things beyond its horizon.

Im only intermediate (about 900 - 1000 on a good day), so hesitant to criticise a chess engine. However, the analysis best move suggestion, many, many times is downright stupid. (yes I may be missing something 10 moves in advance), but the number of times its "Best move" gives a free piece for absolutely no apparent advantage (average 1, or 2, or 3 per game) is ridiculous.  It is so bad that I dont trust any of its suggestions now. Is there a way to change engines used for analysis - and identify this useless one so I never accidentally select it again? 

The Lichess analysis board shows that if you take the knight with the bishop white goes from +10.4 to +17.8 but I don't think that it's a blunder just not the best move because if you do anything else that isn't shown by the arrows you'll lose very quickly. 

schaunRooe, There are a lot rooms for improvement in bringing chess to a wider audience, observing recent chess.com actions, I would say they are on the right track, fingers crossed. On your topic there as a job position according to chess.com recent e-mail to provide human explanation of the moves. For example, this move leads to wining a rook in 5 moves, or this move provides knight outpost in 2 move.

I'm not sure this should be the first priorities, I’d probably firstly fix wrong labels in puzzle classification, for example puzzle label is rook on 7th and there is no rook and passed pawn (potential rook) on the board and do basic stuff with bots behavior and chat, for example 'teach' Beth Harmon bot to resign.

2ndBest lets imaging that AI at some point will ‘solve’ chess game, and verdict will be from the right white first move mate in 120 moves is coming. Black fights inevitable, it is only the matter of would they be matted in 120 moves or faster. Following your logic there will be no bad moves (for black) at all, why to worry about moves, if you are doomed from move one.

Will the chess game be ruined, definitely no. Humans are imperfect, even if you are on the verge of collapse, making this path longer increasing your chances to win. Common approach to evaluate move strength doesn’t take into account, lets say, how desperate or sensitive your position is, its pure math. In your example single move drop from +10.4 to +17.8 close to the whole queen for free, to my understanding in any reasonable chess universe such move must be called a blunder.

the blunder that is suggested as a best move might be a gambit or a sneaky sacrifice