🚀 go-pugleaf

In article <8692@star.cs.vu.nl> henk@cs.vu.nl (Henk Smit) writes:
>etxtomp@eos.ericsson.se (Tommy Petersson) writes:
>>In article <9012110556.AA11184@mole.ai.mit.edu> cracraft@AI.MIT.EDU (Stuart Cracraft) writes:
>>>On a SUN **4/50** GNU Chess defeated a CRA-rated program
>>>of USCF 2265. The real rating for that one was probably 2220.
>>> <.....>

> Could someone explain to me how the USCF ratings relate to the ELO ratings ?
> I understand ELO ratings, and have an idea which strength they indicate, but
> something like 2265 USCF-points (or 10, or 1,000,000) means nothing to me.

It has been a while (like almost 20 years) so I could be wrong, although the
last time I remember the USCF tried ELO ratings before returning to the old
formulas.  I am not going to dig through my old Chess Life and Review books
or attempt to find my Tournament Director stuff containing the explanation,
because it might have changed, and what I have to say is much more general.

First, the "real rating" vs. "CRA-rated" has to do with ratings of computer
chess programs, which manufacturers naturally want to state to the highest
extent consistent with observable phenomena.  I have this bridge I can sell
you, cheap...

Second, ELO ratings stem from the work of Professor Arpad Elo and predate
the use of computers to calculate things easily.  He derived a set of
point differentials based upon a linear numerical rating of observed chess
playing strength.  While I never inspected the mathematics of his work, it
is copacetic that the difference in ratings should be the most significant
factor in deciding what the new rating should be, based on the result.  I
do remember that his table approximates a standard deviation and reaches a
zero-expected-result at about 700 or 750 points.

The USCF rating is a linear approximation of the ELO system, which made it
possible to compute ratings in the days when a 1K table would have to be
looked up by hand every time.  Technology sure has marched:  when was the
last time you saw a slide rule, much less used or bought one?

The USCF rating is generally consistent with the Elo rating within the
limits of linear approximation.  The expected-zero-result is truncated at
350 points.  When players of similar strength compete (as in the popular
class tournaments) any difference is probably one of perception.

There was an odd rating system used in USCF postal chess but I believe it
was made consistent with USCF ratings a few years ago.

Laird Heal                 
laird@chinet.chi.il.us
-- 
Laird J. Heal                           The Usenet is dead!
Here:  laird@chinet.chi.il.us		Long Live the Usenet!