"Perfect game" by me! (41 moves, Caro Kann Exchange Variation)

Really perfect.

My game I won on time: http://en.lichess.org/GzZEEaAp/black#79

I know it's three months old, but I just came across this thread, and saw something in the initial post I'd like to comment on.

It's hardly clear that Houdini could improve on Stockfish's analysis. Of course each of the top engines will have some positions they play better than the others, but in general, it's actually the opposite: Stockfish is stronger than Houdini.

That's borne out in all the up-to-date rating lists, as well as the recent TCEC (and the almost completed special FRC event), which SF won relatively easily.

See a few rating lists for confirmation:

http://www.computerchess.org.uk/ccrl/404/
http://www.computerchess.org.uk/ccrl/4040/
http://fastgm.de/240+2.40i7.html

H4 is still very strong, and Komodo is not terribly far behind at slower time controls on big hardware, but for now, SF is king :)

That much is clear, of course :)

The interesting thing with giving engines more time, too, is that while there are diminishing returns from using more time, the curve doesn't flatten out nearly as quickly as one might intuitively think.

In other words, within some limits (less than weeks per move), engines keep getting measurably stronger with more time, which is an interesting result. For a long time many of us who follow those things had assumed that past some point like a couple hours a move, engines would stop getting measurably better, but that turns out not to be the case.

There are some fascinating discussions of that on the talkchess.com forums, for those interested :)

<Yes, but a real Houdini or Stockfish running on a normal computer could improve on the 1 second per move Stockfish we have on this site, because it would calculate deeper and thus find more subtle points. >

I've analyzed some of my games with my own stockfish running for 15 minutes and you find many inaccuracies that stockfish on this site simply ignores because of the limitation of resources ofcourse. so if you analyze that same game with your own computer it is highly likely that you find more mistakes on both sides but it is irrelevant.

The Stockfish analysis at Lichess has gotten terrible. It used to be very rare for someone to play a perfect game on here, so if someone played two perfect games in a row chances were good they were a cheater. Now not only is it common to play "perfect games", but someone can play a "perfect game" and still lose, as happened to my friend AdmiralA.

Let's look at your game, not to mention the fact that your opponent didn't play the Exchange Variation, which involves bringing the bishops out before knights and developing the light squared bishop to d3 and not e2, but at move 12. you have a -.25 advantage as black, then on move 13 white has a has the advantage again of .31. This is more than a half pawn advantage shift in your opponent's favor, so how was this played perfectly?

Also I have seen the Stockfish analysis here giving completely different lines for the same positions. This happened in two of my Budapest gambit games where White plays Qd4. Both times it called this an inaccuracy but one time the line began with Nc3 and the other e3.

I don't know how many seconds of analysis time they're using per move 1 or 2 seconds like people are saying. Have you ever done analysis using an engine on a home computer? The move evaluations can change radically within the first 30 seconds, and often it can take several minutes to arrive at a final conclusion.

The computer analysis on lichess should not be expected to be exhaustive and thorough. The computer analysis is simply meant to be roughly indicative of who has the advantage, and to supply basic commentary on standard mistakes.

If you want thorough analysis, the Download PGN button is quite accessible, so you can run local analysis.

Side note: The amount of AI servers has increased handsomely over time, and as a result we've been able to parallelise the processing and increase the amount of time spent calculating on each move.