average CAPS score

I started keeping track of CAPS scores and after 5 games (all in the 80s) the average CAPS score is 85.5 

when do you think the average CAPS score would accurately indicate whatever CAPS is supposed to indicate?  after 20 games , 50 games ?

My opinion is after using it a bit and seeing where it was good or bad, never. Seriously.

You might think someone could just run a few thousand games through CAPS analysis and compare resulting CAPS scores against the player ELOs of those games.   Shouldn't it be possible to prove a correlation if in fact there is one ?  If there is a correlation isn't it kind of basic statistics to figure out what sample size you'd need for different confidence levels of prediction ?

Maybe Danny Rensch and friends already did this ?

sorry bad edit ...

My point is this. I'm sure there is a correlation. I think it is probably a very weak one that isn't very predictive. And given what I have seen of how CAPS works in practice, they are fortunate it is even very weak.

do you feel the same way about computer analysis of individual chess moves ?   

I don't know, that's for sure ... maybe I over rate the quality of computer analysis in general  -  like somehow computers know for sure about what we are only guessing ?

Certainly computers are way stronger than we are in most positions. But when trying to rate the strength of humans, not all moves are equally predictive of strength. I had one game with a CAPS score that I think was in the mid 90s where I blundered 4 different times. I certainly wasn't playing at the implied strength.

You average in garbage, you get garbage results. GIGO.

so maybe CAPS is garbage but I don't get how computers can be accurate analysing individual moves and somehow not be good at estimating strength based on all moves played -  maybe if CAPS did a better job of amalgamating individual move assessments it would also be good ?

I'm not saying CAPS is good or bad at amalgamating individual move assessments because I don't know , I'm just saying there should be a connection between a bunch of good individual move assessments and an amalgamation of all those assessments.  no ?

I don't think CAPS is very accurate.  After I analyze my game on chess.com, I save the pgn file and then load it into the freeware Lucas Chess and have Stockfish analyze it.  Lucas Chess makes an estimate of the opening, middlegame, endgame, and overall rating and it's usually lower than CAPS.

For example, here a game where I was 1800+ and my opponent 2000+ and I, White, won.

Here's the chess.com max analysis.  Despite the fact I won and was never in trouble - an exchange sacrifice or a R for a N gave me the edge - CAPS said my game rating was about 1650 (CAPS 82.84) and Black's rating was about 2175 (CAPS 92.05) - based on the CAPS to ELO comparison chart (at the bottom of this message).

Here's the same game analyzed by Stockfish 10, 20-ply on Lucas Chess.  It says I played at 1810 insteads of CAPS 1650 and Black played at 1600 instead of 2175. The engine's calculations of avg. lost scores was also reduced for me and raised for Black. Admittedly, this games ELO numbers were driven down by me making safe moves instead of the "best" moves in a won but drawn-out middlegame, but the Lucas numbers fit the game a lot better.

Here's the CAPS to ELO table:

funny you should mention Lucas chess , I installed it yesterday and ran my first few analyses tonight .  yes I think it's much better than Max analysis on Chess.com  , more meaningful information AFAIK  -  I'll probably look at CAPS from time to time just for fun but won't place much importance on what it says about individual games or averages of individual games

thanks

I'm just curious, where did you find a download for this freeware? One download I attempted was really disliked by my security, and the other site I tried to download from gave me a corrupted file. 

(lucaschess.pythonanywhere.com and
lucas-chess.en.uptodown.com respectively.)

Sometimes my security system jumps at ghosts, but I don't really feel like taking that risk. 

I run it under WINE in a Linux virtual machine that is kind of sand boxed.  I appreciate your security concerns but can't comment on the code.  I chatted with the author of Lucas Chess a few years back on an Indian chess site and doubt he would have intentionally put anything bad in there but who knows what's really "under the hood" ?    

my average caps score is in the 90's plus. my elo should be 2300 orso. that has never happened in real life OTB.

wow , I guess CAPS really is CRAP ?

I have checked and you have a lot of games against very weak opponents.

CAPS seems to give an adecuate stimate for me when the analysed games are a mix of wins/defeats/draws against opponents of similar strength.

If I play against a 600 and deliver the scholar´s mate, then CAPS is of course useless.

Is there a way to see your own CAPS rating here? I know mine is bad, just not sure how bad? Am I missing it somewhere on my member page?