Accuracy versus game rating

A very odd discrepancy in this game, which makes me wonder what on Earth the "game rating" in the engine review is doing.

It was a long bullet game ending in checkmate by me (after never having a disadvantage, and having been in a winning position for half the game). According to the review, I made 1 mistake and my opponent made 3 mistakes and 1 blunder. I had 31 best moves (about 50%) versus my opponents 20. My accuracy number was higher at 76.7 versus 70.3. But the weird thing is the game ratings were 1650 for me and 2000 for my opponent!

Hello @Elroch . I Just analyzed your Game on The phone app And It was completely different! The Ratings were actually accurate.

Interesting! Could you clarify what you mean by "accurate"? That they were the other way round?

[EDIT: this is hilarious - when I looked back at my homepage, the game had entirely different accuracy numbers to the above - ~89 for me - which had surely come from your review replacing mine. Then when I did the review again, it replaced those with the data I posted in the OP here!
Can you possibly repeat the review and post a screengrab of the review "highlights" (i.e. like the OP) here?].

With accurate i mean that you had ~89 And your opponent ~81. The Game Ratings were correct there.

My Screengrab Feature ist a bit Bugged bit i'll try

I Just analyzed again And i had the Same result Then Last Time. 2100 For you And 2000 For cendremagician

Maybe it comes from different Analysis depth.

I sadly can't send a Picture but i'll resume It For you: Accuracy 89,4 / 82,5 brilliant 0 / 0 great 1 / 1 best 29 / 24 mistake 0 / 1 Miss 0 / 0 blunder 0 / 1 Rating 2100 / 2000

Thank you very much, @Cyberghost-13. This appears to be a quite glaring bug in the Review tool,

The "game rating" feature can be very misleading because it doesn't measure who played better, but rather estimates the Elo level that would typically make the moves you played in the specific positions you faced. It's common for the player defending a difficult, losing position to get a higher game rating because they were forced to find complex defensive moves, while the player with the winning advantage had a much simpler, more straightforward task.

I almost always see the winner having the higher game rating, and when Cyberghost analysed my game he found that. My results from the analysis tool were very odd and indicative of a bug.

It's almost certainly a bug.

Not that the bug makes the whole rating estimate thingy any worse. It's just a bad joke regardless.

I believe the estimated rating is more of a gimmick to tease you to play more. If I got it right you can review the exact same game, but with different player ratings (edit the pgn tags). You will get different estimated ratings. Ie the estimated rating depends on the players’ rating.

Yes. But i still think It has also to do with Analysis depth.

It's perfectly normal to have games having a rating that is +/- 600 points from your rating. What is not normal is to have 400 points less than your opponent in a game where they never had the advantage, you played no mistakes, and they got checkmated! (That was the case, although the faulty review claimed one "mistake" for me).

I have just redone the review from the analysis tool, with the players anonymized, and it screws it up in a different way. Just look at the evaluation graph and try to make sense of this nonsense!

Note that three different reviews have now found 0, 1 and 3 mistakes for me, always more for my opponent (including one blunder).

It seems that this latest analysis might be different because it doesn't know it's a bullet game. I think the three "mistakes" are actually correctly identified. There were clearly better moves each time. But it is ABSURD to rate my ending as "?!" and the opponent's with a green tick - I entered it with a small advantage, never had a disadvantage and ended it with a checkmate! And he entered it with a small disadvantage, never got an advantage, and got mated!