Pairing problems in long tournaments

The Australian draw has been used for decades in Australia and it works well for most tournaments which are short. But in longer tournaments, the aim of not playing an opponent again results in later games bypassing perhaps ten or more candidates for eventually a more uneven pairing, players differing markedly in their placings.
The solution adopted is to have one or more resets. The question is then how many resests and at what stage.
Barry argues that resets are just a mean towards an end : avoiding mismatches in pairings. The mismatch metric should not be via ratings, because a tournament gives opportunities for players to play better (or worse) than expected. The mismatch metric should be in terms of performance at that stage.
Barry has developed a version of AUPAIR which when considering pairings for round x+1 based on results up to round x, does two things: firstly flag a potentially bad pairing (set at win records differing by 2 or more) and secondly trying to select a better pairing. In the version sent to Paul this used the relatively easy way of going down the list, and the first bad pairing, just accept it without going any further. I am planning a better pairing method. In either version it is easy to impose a restriction that you will not play an opponent in two consecutive games, to avoid two top players tussling while an outsider creeps up. It would be harder to limit how many times a repeat might occur, and for the moment I suggest we try it and see if it is likely to be a problem — I suspect not.
There are several ways of trying to see how well this may work