Originally posted by Paul Leggett
"So just grabbing 20 odd games and running them through a box is
only half of it. You need access to everything he has played else you
could be matching up to someting he played two years ago, went wrong, boxed it, found the correct path and stored it for a later day.
There is nothing wrong in that. "

The italics were added by me, becaus ...[text shortened]... f their losses in future games), but you guys have really clarified the issue.

Paul

Paul.
The way you can do this type of analysis is this:

1) select 20 games which are found to have 20 or more non multi-million game database moves. I use www.chesslive.de for instance.
2) select games vs high-quality opponents. I use 2200+ rateds on chess.com, but the ratings pool is a bit lower in figure here, so I use 2000+ RHP.
3) choose the most recent 20 games which fulfil the criteria. I know what GP means, about someone starting then stopping using an engine, but generally engine use has been found to increase over time, almost like a dependency or addiction.
4) analyse the games once out of book using a 3000+ Elo rated engine, like Houdini, Fritz, Rybka, Stockfish or Firebird
5) use either a fixed time for analysis per ply or fixed depth to ensure results are as even-handed as possible. Set the engine to look for top 3 or top 4 best moves.

To suggest that pre-arranged analysis of a particular opening could skew results toward a false positive assumes that both you & opponents are playing top engine choice moves for most of the games once out of book. I think this is highly unlikely, unless you've rapidly managed to generate something like a 60 million game private cyborg database & your opposition have almost a collusion to play these engine lines!

The non-database move sample size is typically 700-900 moves. This is then compared to benchmark data done in the same way, of both the OTB greats & pre computer-era CC World Championship finalists. The benchmark thresholds (including forced/obvious moves) is remarkable in it's consistency when you look at large sample sizes. Remember - individual games mean next to nothing. It's the results from many objectively chosen games that is the key.

I've personally tested 6 CC WC finals & 7 OTB FIDE WC matches. Other analysts have tested many more.

If someone with little credentials plays not stronger chess but crucially more engine-like chess on RHP than all the benchmark data, then you are left with the question why this may be the case!

Originally posted by Zygalski
Paul.
The way you can do this type of analysis is this:

1) select 20 games which are found to have 20 or more non multi-million game database moves. I use www.chesslive.de for instance.
2) select games vs high-quality opponents. I use 2200+ rateds on chess.com, but the ratings pool is a bit lower in figure here, so I use 2000+ RHP.
3) choose the mos P than all the benchmark data, then you are left with the question why this may be the case!

I'm sorry if I was vague, but I wasn't really referring to this kind of analysis- I was just surprised that we could use computer analysis in our opening prep, and then use the prep here, as long as the prep was done BEFORE the game and not as part of the game in progress.

I'm sure others will find the info useful, though, as people seem to love to talk about the topic!