Estimated Rating - bug or by design?

I uploaded game 16 of the Karpov - Kasparov world championship match in 1985, and then did a game review.  This was the result:

All in all not too surprising.

Then I started tinkering with the ratings in the PGN file.  I ran the same game review with them rated 2220/2200, 1720/1700, 1420/1400, 1120/1100, and finally 520/500.  This was how it was graded with them as 500 players.

We know that estimated rating is based in part on the player's listed rating, so that's not a surprise.  What is interesting is that the counts for brilliant/great/best etc. also change, as do the grading of opening/middlegame/ending.

In terms of the report card, Karpov at 520 appears to have his two misses upgraded to mistakes, two mistakes upgraded to inaccuracies, three inaccuracies upgraded to good, and two goods upgraded to excellent.  So his 0-0-16-2-6-8-3-3-2-0 becomes 0-0-16-4-7-8-3-2-0-0.

Assuming this all makes sense, is this how the grading system is supposed to work?  That accuracy remains consistent, but everything else changes according to the players' ratings?

I have no idea how its supposed to work.

I played a game as an 800 rated blitz player that it scored me at 2500 and my opponent (a WGM) at 2800. I pointed out this was pretty ludicrous, and when I ran it again after I was told a change had been made, apparently I'd only played like a 1750 and she at 2100.

If this is more than a gimmick and a useful guide to help us improve our chess (it is the coach's comments after all) I'm really not sure how to make use of it.

For me, an objective rating of my play no matter who the opponent is would be a more useful guide, but then I guess that is your accuracy score so you already have that?

When I have said this in other discussions on the matter people have said that it is harder to be more accurate against a higher rated opponent.

I'm not sure that is true. You still have to find the best moves for the position whoever you're playing against. Or am I failing to grasp something fundamental about this whole engine analysis business?

it is how u played according to ur skill level

I believe Review takes rating into account on some classifications.

The review takes your skill level into account. It isn't just going to say that a 520 played a game like a 2000. It is all in accordance to the skill level of the account playing.

are you sure?

it makes sense to me....if Magnus Carlsen finds a way to force checkmate in 2 by sacrificing his rook, it isn't particularly brilliant...if anything, people would be surprised if he missed the checkmate. If, on the other hand, a player with 500 elo finds the same tactic, I think most would agree it's a brilliant tactic for that level.

As I ran the games with the ratings going down from 2700 to 500, the "estimated rating" dropped more or less in line. I don't have the numbers handy, but generally speaking Kasparov's estimated rating was consistently 600 or so points above whatever rating I had put in the PGN, and Karpov's was 300-400 points higher than the starting rating.

So the overall progression shows clearly that it does depend on the starting rating.

I was a little surprised that that also affected whether a move was a mistake or an inaccuracy, but I probably should have expected it.

I don't think most of the commenters read the question carefully, and if I understand it correctly, the question is about why all the statistics change, except the accuracy, which is constant regardless of rating.

I think the game review behavior makes sense. Statistics like whether a move is a miss or brilliant is slightly subjective, and the estimated rating naturally depends on the player's current rating. However, there needs to be some metric to actually tell you how well the game was played with no other factors considered, and I think that the word "accuracy" very well describes that.

interesting experiment. I love that you took the time to look at this. I hadn't realized that the system was designed to take in to account ratings when defining certain rating move classes for the game (ie blunder, inaccuracy, brilliant, etc etc).

Although, to be honest, I'm a little insulted now that I realize I get "great moves" on moves that are all-but forced (moves where alternatives are all blunders but the "great" move is the one "forced"-type move) -- they can't possibly meet anyone's definition of a "great move" -- but now you've made me realize, that for someone at my level choosing amongst options that include a single good move but several blunders is still "great" in comparison with the alternative.

I'm not sure if I'm explaining this clearly enough, but, yeah, ouch. that realization is a bit depressing! Time to go back to playing with my crayons, I suppose if I fail to eat any or poke my eye out with one the teacher will tell me how amazing I'm doing....

But I'm happy to see people asking questions and testing out the system like this; that's a pretty cool discovery in terms of how those moves are defined.

This 1200 blundered a fork on move 12 and resigned a couple of moves later. His opponent was rated 2450.

No one can seriously tell me he played like a 2100.

I believe this was because he played the opening near flawlessly. But also, I realized that if I put this PGN in: 

[Event "Live Chess"]
[Site "Chess.com"]
[Date "2023.03.08"]
[Round "-"]
[White "Hikaru"]
[Black "alexrustemov"]
[Result "1-0"]
[CurrentPosition "2Rr2k1/pp1P1pp1/6n1/1q4P1/1P2p1Qb/P7/5P2/2BR2K1 b - -"]
[Timezone "UTC"]
[ECO "D12"]
[ECOUrl "https://www.chess.com/openings/Slav-Defense-Modern-Quiet-Schallopp-Defense...10.Nxe4-Nxe4-11.Qxe4-O-O"]
[UTCDate "2023.03.08"]
[UTCTime "20:11:06"]
[WhiteElo "3193"]
[BlackElo "100"]
[TimeControl "180"]
[Termination "Hikaru won by resignation"]
[StartTime "20:11:06"]
[EndDate "2023.03.08"]
[EndTime "20:14:57"]
[Link "https://www.chess.com/game/live/72045320373"]
[WhiteUrl "https://images.chesscomfiles.com/uploads/v1/user/15448422.90503d66.50x50o.856e9b278561.jpeg"]
[WhiteCountry "2"]
[WhiteTitle ""]
[BlackUrl "https://images.chesscomfiles.com/uploads/v1/user/76468406.c90f04b8.50x50o.73189932e0d1.jpeg"]
[BlackCountry "116"]
[BlackTitle ""]

1. d4 d5 2. Nf3 Nf6 3. e3 c6 4. c4 Bf5 5. Nc3 e6 6. Bd3 Bxd3 7. Qxd3 Nbd7 8. O-O
Be7 9. e4 dxe4 10. Nxe4 Nxe4 11. Qxe4 O-O 12. b3 Re8 13. a3 Qa5 14. Bb2 Rad8 15.
Rfd1 Qh5 16. Rd3 Qg6 17. Qe2 Bf6 18. Re3 Nf8 19. Re1 $6 Re7 20. h3 Red7 21. Re4 $6
h5 22. Ne5 $6 Qf5 $4 23. Nxd7 Rxd7 24. Rd1 Ng6 25. g4 $6 hxg4 26. hxg4 Qg5 $6 27.
Bc1 Qa5 28. b4 $2 Qa4 $2 29. d5 $6 cxd5 $6 30. cxd5 e5 $6 31. Rc4 Rd8 32. d6 Bh4 33. g5
Qd7 34. Qg4 Qb5 35. d7 e4 36. Rc8 {<br /><br />Game may have continued...} (36.
Rc8 Bxf2+ 37. Kxf2 Nf8 38. Be3 Qxd7 39. Qxd7 Rxd7 40. Rxd7 {+13.2}) 1-0

it tells me that White (Hikaru) played like a 2300 and his opponent like a 1950. With the original ratings, it estimated White to be 3350 and Black to be 3000.

I can get that estimated rating partially depends on your own rating... but on your opponent's rating too? That seems a stretch to me.

Yes. me too. 

This is a little out of date but explains the definitions and that the expectancy values are rating based https://support.chess.com/article/2965-how-are-moves-classified-what-is-a-blunder-or-brilliant-and-etc

Well, good luck with coming up with something that makes sense to every day users.

It seems I've been asking too many questions and reporting too many bugs.

@Eadwig2 I could always be mis-remembering, but a while back I think there was a report that those messages were being sent out in error..... I'm pretty sure you can just re-join with no consequences if you really were booted (as the message suggests).

Well, ironically, the whole thing seems bugged. Yes, you can indeed just re-join any time, it even gives you the link to do so!

So I did, and I got booted again a few minutes later. So I thought I better take the hint - at least for now.

That's very weird. I'm sorry that's happening to you, I can't imagine that's not a bug in itself.....

that makes sense now, it used to be that I only played bots (which don't increase your rating) and I was rated whatever chess.com started me with, but as I started playing games with normal people that increase my rating, I now never see a brilliant anymore.