"Fidelity" Website FAQ---General Considerations, Background, and Personal

This page answers some questions that have been posed in stories and forum comments, and gives some general scientific considerations, background, and personal vantage points.

1. Commenters here and generally have noted that use of such a program is "dangerous for honest players." Some steps in my papers, talks, and even the presentation of data in the New York Times article come from express awareness of this. My view is that the greater danger comes from unscientific practice and lack of statistical "street knowledge" such as Littlewood's Law. Most important:

Accusations of cheating that report move-matching statistics should not be entertained unless they are accompanied by logs of the analysis that is reported, and a statement of methodology sufficient to reproduce the procedure.
Letters noted on my site, and some other accusations well known to all, have contravened this principle.

2. The most important check on danger is that my full model provides error bars. Those are being field-tested first using simulated player-performances as reported on the blog I co-manage, and now with sufficiently many actual player performances.

3. I confirm the communication from me reported by Frederic Friedel here via the open-feedback link of his item, and it is reproduced accurately. The line with "essentially every move of every game" was a self-reference to my own webpage on the World Open 2006 case, in which I reported my reproduction of allegations reported in a column by Dylan McClain in April 2007.

I have agreed with the "one-bit" point stated by Frederic and others in communications going back to 2006. Nevertheless in Jan.-Feb. 2007 I noted that "the issue" as stated in this article was long sequences of moves, for instance the statement, "Later it turned out that all the moves that Topalov had played in this decisive phase are also the first choices of the popular chess programs." Incidentally my testing reported here does not reproduce this conclusion---this is an example where "issue" does not entail guilt. Also here, the issue was, "...that you could reproduce virtually all of [Clemens] Allwermann's moves with the chess program Fritz." Add the statistical results communicated in comments here in July 2007 to some publicly-known allegations in other cases, and there are already enough references to substantiate my assertion about "the issue" and "regrettable cases".

For contrast, in the 2011 German championship case, the issue for all we know is only three moves, and I can report that my stats on that player's game or games do not show anything. His IPR including the last game is 2345, without it, 2366 (mind you, with big error bars because he played only 162 relevant moves in those 9 games).

4. My statement about Karpov in the 1970's and the world's 40th player is based on the finding of no discernible rating inflation in my two papers from 2011, with independent kinds of data going back to the 1970's. Anatoly Karpov's ratings in 1976---1979, the four years of my intensive data, as given here from January lists, were 2695,2690,2725,2705,2725 with the last from January 1980, averaging 2708, almost exactly the 40th player on current lists. As noted in my comment here, I have taken IPR's for Karpov of 2722 in the 1978 match win over Viktor Korchnoi, and 2721 when co-winning Montreal 1979; I will soon do the 1974 Candidates' match over Korchnoi and other performances earlier in the 1970's.

More to come...