If I were running a 38-round event, with 50 players, I would use a
RR-guided Swiss.

Thanks for the answer! I din't know this system. It seems clever. But..

Pairings
for the top players will be very much like normal Swiss pairings - but
pairings for the lower ranked players will be a bit strange.

(Probably another dumb question) Doesn't we have just *one* correct
pairing (the top player's match) and twenty-four random ones?

The hard part will be finding a 50-player RR

I've found an easy way to build one in Wikipedia:
http://en.wikipedia.org/wiki/Round-r...ling_Algorithm

Frankly, it would also work (perhaps better, and certainly easier) to
simply assign pairing numbers by lot and play the first 38 lines in the
RR-table.

Sure. But I'd like to have: (1) every player, every match, playing
against a near opponent, so, if he'll win, he'll go up some positions,
and, if he'll lose, he'll go down; (2) not every time the same pairings
(but I'll accept, let's say, that two players meet 4 times).

Maybe a solution is to use the swiss system, dividing the tournament in
mini-tournaments:
- In every mini-tournament, two players can't fight each other more
than once; but, when a new tournament begins, I reset this counter.
- For the first round of the second (third, fourth...) mini-tournament
begins, the players are paired according to the points they conquered
in the previous mini-tournaments.
- For the next rounds, I have two choices: I can pair the players
either (1) according to the total points thay gained in the entire
tournament or (2) according to the points they gained in this
mini-tournament (in this case, players in better form are paired one
against the other).
- The number of rounds in every mini-tournament (except the last...)
could be the square root of the number players (i.e. every
mini-tournament would be good for a normal swiss system).

Kenneth Sloan
http://www.cis.uab.edu/sloan/

It seems that you study interesting topics!