Is chess.com's "Rating guess" accurate?

Recently I played a rapid game that I definitely feel proud of. 91 percent accuracy, no mistakes or blunders, and only one inaccuracy.

(Here's the game in case you're curious)

This is the rating that chess.com suggested I played at:

Now I'm wondering, how is the rating calculated? Is it just taken from the accuracy, or are there other factors applied that change the rating?

Also, those rated at the level it suggested we played at (1900-2000's and 1600-1700's), does this game look to be accurate to how games are played at your level, or does it differ? This doesn't really have an effect on how I play my chess, I'm just curious if this is truly accurate, because if it is, I may have some potential to rise far beyond my current rating.

Thanks, Flab.

That's currently a beta only feature and is going to be receiving some updates early in the year, which should make things a little more realistic. That said, there's not a lot of details on how that's being calculated, but is influenced by rating.

LucasChess has a similar Elo Game rating system.

I back loaded a few thousand games into LucasChess and determined all the ELO game ratings - takes a very long time - I did this with the intention of filtering out poor-quality games. Results ranged from 960 to 2600+ with a mean of around 1600 to 1900 (not determined exactly) - quite flat Bell Curve. When I played real people on chess.com my rating was 'randomly' in the range 900 - 1100 over a two year period.

With a bit of rough analysis I drew four conclusions:-

(a) Quality of gameplay is hugely inconsistent.

(b) Lower time constraints trash gameplay quality - and I mean trash.

(c) High quality gameplay tends to occur with early opponent opening mistakes.

(d) My own gameplay quality didn't seem to be correlated with game results!

Conclusions (a) to (c) would probably apply to other lower level players but for the most part I doubt there is anything surprising.

Conclusion (d) is a bit of a puzzle but there may be individualistic reasons for this!

However the overwhelming observation was the huge inconsistency in performance, a factor completely hidden by having a single rating measure. Indeed for on-line gaming I think rating statistics should include Standard Deviation(SD) measures - especially since these are very easy to calculate. Would also stop a lot of arguments as well!

I have argued elsewhere that chess.com's ELO system doesn't work at low levels, for on-line platforms - reasons not fully determined.  Not going to dig all this up again, but the analysis performed does not dissuade such a conclusion - giving that any such game rating system is benchmarked to a fixed 'known' engine rating for both players.  

no

Where do you see the rating estimator? I joined Beta to use this feature, but I can't find where it is.

I think you need a premium membership. 

Also, I don't think it's accurate; if it is, then give me the IM title

Chess: DonRajesh vs Charlieeden123 - 65811216625 - Chess.com

A premium membership can be a diamond, gold, or platinum right? Because I have diamond

Oh missed that. Idk then, maybe you have to be in Beta when the feature is released. Also, I just saw that its web-only.

Oh yeah, I am using the web, so it must be that I had to be in the club when they released that.

Also nice game

I believe in this case, that game is not a fair judgement. A short, 6 move game played at 100% accuracy would most likely give you a high score; you played book moves and then the best moves only. In this case, this game would give you a high rating estimate (based only on this game, not on others you played, as it specifically states) simply because the game was played at 100% accuracy.

I saw a game where both players had 3850 elo.

Here is why I believe the new performance rating is inaccurate; 89.9(rapid) 85.7(bullet) 83(rapid). Can you guess which one got a 1650 performance rating, with one blunder as the only bad move? 83. How about 89.9, with one inaccuracy and one miss? 1400. How about 85.7? 1100. I went on to check the win I got against the bot Li (2000 elo), which I had played like crp, but she blundered a tactic. I got 83.5 accuracy, 3 mistakes and one inaccuracy, but just because it was Li, I got a 1900 performance rating.

So if you get a lower performance rating than you expect yourself to get, don't get sad or worried, because it's not showing how good you are, just how good you are relative to your opponent. Look at your accuracy if you want to see how good you are.

I think it's cool, but for the moment, it's pretty inaccurate.

It said that the level 25 maximum bot was 2600 and it beat Magnus Carlsen, unless I just resigned to early or was barely a match till the point it ruins the rating, other times it seems way off; according to it, I'm anywhere between 100-1800 elo, I think it only rates the game as a whole and splits it not taking into account the player.

If you have free membership you get 1 game review per day. The rating pops up when you do a game review.

I kept a record of my rating and how well chess.com rated my games. Over 132 games it on average rated me 181 higher than my player based rating. 1235 to 1416

Actually the rating by the computer is much closer to my OTB ECF rating

No, it's a gimmick. For example, grab a PGN from a super GM classical OTB game, change the elos to 1000 for both players, and run it through the analysis. It will give you a higher rating than 1000, but nowhere near super GM elo strength.

Then go an change the elo a few times and re run the analysis each time. See how it changes.