How accurate is accuracy?

So this question stems from an argument I had with my friend. 

He's played fewer than 10 games on chess.com and has a 2100 rating. I was doubtful from the start since he's never actually played chess seriously (taken lessons, tournaments, etc.) and I, a 1400, have beaten him before. So I look into his blitz games against some 1900s and all of them have 99.2 or 99.1% accuracy and I'm thinking he must have cheated. He claims he just played really well. Are my suspicions unfounded?  How accurate is accuracy? Please share your thoughts below.

besides accuracy..........playing less than 10 games does not yield an established Rating.   your friend has to play many more games to establish credibility.

We must accept he is an amazing genius online. Otherwise your fren is a groveling cheat

Ask @st0ckfish

Ask @Lc0

or ask @stockfish

Just ask him to play in real life. If he consistently sucks, he's been cheating.

it's a crude measurement tool

That is not true. I am no where near Carlsen's strength and I routinely get accuracy scores in the mid 90's. I also get them in the 60's - 80's.  

The problem is how the system determines the precision. When running GM's OTB games on chess engines, a match of just over 60% is common. So you have to find an explanation of why fans and experts reach more than 80% and 90% in casual games on this site.

The reasons are not complicated. Unless we are talking about a more or less forced line, the chess engine needs time to dig deeper by calculating variants and setting an appropriate positional value. As a consequence, the initial –fast– analysis sometimes praises a positional blunder and sometimes tags a deep positional idea as a blunder.

Another reason is how the system determines which are the optimal moves for purposes of establishing an accuracy percentage. And here the system gives optimal value to moves that are good but not the best possible.

Why were you even arguing with your friend XD

I don't think the accuracy is very good, as in one of my games my opponent blundered and I didn't see it, and then I lost the game, and he still had 98.3 accuracy. weird.

I am in a similar situation. I recently joined chess.com and only played like 12-14 games. My rating is usually around 2260 - 2300, but I have already reached my highest rating peak ever. Once I play more and lose more games it will settle in the middle probably.

This example should tell you much about accuracy per the formula used here. No blunders. 1 inaccuracy and 1 mistake for losing side, 2 inaccuracies for the winning side, 91.7 accuracy for the losing side, and 43.0 accuracy for the winning side. The side with the horrible accuracy had the advantage from start to finish.

my high is 97.9 in over 40 moves 

My accuracy is usually in the 90's and I'm only rated 1500+.

The truth is at a higher level of play, here on this site at least, a lot of games are subpar, or should I say a lot of games follow basic and familiar positions. You only have to follow some basic principles and you'll score high in the way of accuracy. Play a basic opening, develop classically, simple exchanges, stalling and lingering, these all score perfectly because they certainly aren't mistakes.

It's the wild games that will really test a player's accuracy.

When my opponents play against me, their accuracy score drops. I really put 'em through the wringer.

Another example is when you force your opponents into a defensive position; they're only play the only moves they can play, and with that they earn a high accuracy score.

That's not to say there's something wrong with the feature. It's doing what it should be doing. It's a statistical tool much like the analysis engine itself. It's for advanced analysis of games, and needs to 've understood to be used correctly. 

Are you insane? It says White had the advantage from start to finish, and white had the 91.7. And white probably lost because he/she blundered in the end or resigned for no reason. This is not an applicable example when we're talking about a 99 CAPS.