Improved Computer Analysis

Today we installed our new Computer Game Analysis system. It uses an improved engine which makes analysis faster and more accurate (theoretically :D).

We'd love to get your feedback - what do you think of the analysis?

(NOTE: While we were installing this we got backlogged on about 5,000 games, so it's going to take some time for those to finish... thanks for your patience!)

Is there any way I can re-analyze games that I'd previously had analyzed with the old system?  It doesn't look like I can.

I am going to test it out thank you so much Erik :)

     I just entered a game to test it, although the computer may be stunned
that I actually lost.

cool! how many points did it improve by?

nope. sorry!

So before you were using fruit(?)...now you are using??? Stockfish? 

is it available for basic members?

 1 game a week for basic, and the engine is a bit wekaer but still 2200+ elo.

ok thank you

Thanks erik!

Thanks.  I send every game in for analysis, so I'll really look forward to this.

There are a few bugs. I just analyzed my most recent victory, and the computer said of my last move that "black wins", which is correct, but then goes on to say that play might have continued thusly, ending with "white has a decisive advantage", when he clearly doesn't as far as I can tell.

http://www.chess.com/home/computer_analysis.html?id=30286579&game_type=1

Hello Erik,

You might take a look at this game:

http://www.chess.com/echess/game.html?id=30225106

I blundered with 44. Rf6 but your computer evaluates the final position as slightly better for Black. In fact the final position is completely lost for White, Black can promote one of the d or f pawn without using his king.

That is indeed won for black. However, we know computers are terrible in endgames and evaluations there should be taken with a grain of salt.

Computers are weak in the endgames, where they don't recognize fortresses or missevaluate bishop of opposite colors endings. However, are they so weak to miss a clear win in a couple of moves?

I have Toga on a quite old office laptop and it finds the correct evaluation within 2-3 seconds.

Hmm, Crafty finds it in milliseconds, not to speak about Fritz or Rybka.

Doesn't the analysis computer on chess.com use tablebases as an aid?

This is not a problem with the engine, rather a problem with fixed depth analysis. Computers will get to 12 or 14 ply (whatever the value for diamond members) in less than a second. Since there are so few pieces it is easy to get to the higher plies and get the more accurate results that you are talking about. 

Indeed when restricted to 12 ply analysis, Glaurung 2.2 evalutes around 1 (slightly better for black), and after 3 seconds, allowing the ply count to go up, the correct value of close to -100 (value for checkmate) is reached. 

we've found a few bugs that we are fixing. please do help us if you find other oddities!