Suggestion: Played like a ....

Can we have a feature in the analysis, where you can see the performance rating. For example, if I played like a 1200, but I didn't know that, I could go to the feature and it would show me and my opponent's performance.

Thanks!

Agreed

Agreed

+1

Performance rating off one game isn't a meaningful metric. The site could use it's CAPS information in combination with the current rating to form some kind of estimate, but that's probably not very accurate.

Yes it might not be accurate, but still it can give an idea on how good/bad a player played.

Thanks for your opinion Martin, that's ok, I just wanted to know how well some players played and my performance. It was just an idea.

+1

That's basically what the accuracy value is.

I'm sure there are some kinds of algorithms that could potentially give some kind of performance rating, but again, don't know how useful they would be on a single game basis. Seems like the margin of error might be kind of high.

Maybe something like Kenneth Regan's IPR and z-score algorithms could do it and be an approximation. And as mentioned, the original implementation of CAPS was related, though I think its main benefit is over a collection of games, less so for individual ones. 

+Infinity

Well, as @Martin_Stahl said, that's literally accuracy and CAPS

I like this idea a lot..   would also like clarification why accuracy ratings are different, sometimes significantly, betwwen the mobile app and the pc app.?????????

?

I think the apps either calculate locally or have a fixed depth, which may be different than what you have configured on the site.

The bottom line is that a performance metric in terms of rating level is *much* more complex than doing a statistical analysis of excellent vs good moves vs inaccuracies vs blunders within a single game. It's like trying to measure a car's speed off of a single photograph -- basing an estimate of speed based on how blurry or sharp the photo comes out (given that you know the exposure time,, etc) would only give the roughest of inaccurate measurements. 

???

This would be great, but I think we'd need more data points than one game for it to be meaningful.

Expected win share for a game based on moves seems doable, but that too would need a ton of data. It's tough to know "what % of 1500s would see this tactic, or that."

However, it would be neat to track, say, the average points per game for players who had 1% higher accuracy than their opponents, or 2%, etc. We could then reverse the ELO format for that e.g. if 10% more means you get .8 points per game (for example) then that'd mean you played at roughly 280 points higher (400 * log base 10 of 5.)

But of course things might be different for 30% accuracy vs. 20% as opposed to 80% vs. 70!

+1000

I don't know how to make a puzzle, but this should make the point.

Some moves are easier than a "Played like .. " might suggest, others are harder, and since a game is a collection of moves, we would also need a way to take averages. Do certain moves/positions weigh more heavily than others, and how could this be determined? Unless there is a way to resolve this, I'll stick with the CAPS score that we can get.