http://my.chesszen.com/status/gmgames.php?name=karpov&button=Submit
karpov's games analyzed on chesszen played on standard time ctrl
as see rating fluctuates from 1931 to 3383, but averages to 2761
[quote]
ELO Estimation, Accuracy
Why is there a large variation in the estimated ratings from game to game?
The large variations in the estimated rating from game to game is a natural reflection of the fluctuations in human's level of play. Remember you beat a strong opponent one day but lost to a weaker player the next day.
How accurate are these estimated ELO ratings?
Since human's level of play vary from game to game, you need to take average of multiple games for a more accurate rating. If we define the 'true rating' as the long term average of several estimates, there is a 90% chance for the estimated rating of a single game is within 500 points of the true rating. The estimated rating is calculated to be within 250, 150, and 100 points of the true rating for an average of 4, 10 and 20 games respecively.[/quote]
so i think the number 2000 came about since that alleged match vs fischer was some informal blitz / semi blitz match, and rating from one such game cant be used to determine a player's actual rating (which is derived from many games played on tournament time controls)
I found this game on chesszen site which allegedly claim to be played between karpov (white) fischer (black). it is just mentioned that game is top secret and there is no information on when or where the game was actually played
PGN
http://my.chesszen.com/status/index.php?refnum=5046-3872-7921
judging from lower ELO compared to most of their other games it was probably an informal game played with non standard / rapid time control