playBunny: The same applies but to a diluted degree if a player restricts themselves to a small group of opponents who may pr may not play more widely themselves. That's the issue that I think Fencer is looking at and it's a fair one from the game pool/rankings perspective.

That's a fair one, but IMO, that also isn't solved by only counting the first N games played against each other in a year. You're not getting a more accurate rating by giving more weight to games played in January than the games played in December.

If you want to restrict the influence of a large number of games a pair of people played against each other on the BKR, perhaps something can be done to "normalize" the results. Say you want to count all the games as "as most 14 games", and a pair of players, say A and B, have played 48 games, with 30 wins for A, and 18 for B. Then normalize that as 8.75 for A and 5.25 for B. (30 / 48 * 14 = 8.75 and 18 / 48 * 14 = 5.25). Of course, this is simpler said that done, as the in the current system not only the number of wins/losses matters, but also the order in which they were made, and what the ratings were at the moment games finished.