Some background on how pairing systems work.
Updated Tue Jul 12 11:08:09 EDT 2005 for tsh 2.960.
This section discusses the theory behind scheduling pairings for tournaments. For information about how to implement pairings in tsh, please refer to the section on configuring tsh.
Choosing a pairing system is a balancing act among several conflicting goals. A pairing system should be fair: ideally, players should end up ranked the same way whether you list them by performance rating or wins and spread. A pairing system should be fast: ideally, players should not have to wait to find out who their opponents are. A pairing system should be fun: ideally, the outcome of the tournament should not be determined until the end and in the last rounds contenders should be playing each other.
A fair or fun pairing system cannot be fast, because getting the pairings just right usually involves waiting until the last game finishes each round before deciding pairings for the next round. A round robin seems fairest (everyone faces as close to the same field as they could face, without playing themselves), but is rarely fun, because of the difficulty in predicting who will be in contention at the end of the tournament.
Some directors favour fairness (and simplicity) and divide their players into small groups so that they can all play round robins. This has the advantage that the director doesn’t have to do any work to figure out the pairings (but s/he wouldn’t have to anyway with this software), but the disadvantages that the prize pools and the potential ratings changes end up being relatively small. Most players prefer the larger ratings swings and prize pools that come with playing in larger groups.
Swiss pairings were invented to deal with the problem of tournaments where the number of players is vastly larger than the number of rounds, and they work well under such circumstances at preventing contenders from playing each other until late in the tournament, if ever. In ideal Swiss pairings, the top half of each group of players who have the same number of wins plays the bottom half. Portland (or Fontes) Swiss pairings improve on the logistics of Swiss pairings by pairing based on results from one round previous: players are arranged into win groups based on how many wins they had in round N–2 when pairing for round N.
With regular Swiss pairings, round N pairings cannot be determined until after all of round N’s games have reported in, which can result in games starting 70–75 minutes apart. With Fontes Swiss pairings, games can be scheduled to start hourly, allowing an extra game to be played in a typical daily schedule.
Each pairing command (except for RoundRobin) lets you specify how many repeat pairings a player is permitted. Early in a tournament, this should be zero, as no one should play a given opponent twice if it can be avoided. Later on in a long tournament (at least with typical North American group sizes), it should be greater than zero, so that the tournament is decided based on results between contenders rather than contenders competing to see which weaker players they can beat by how much. You should decide beforehand at what point you are going to allow repeat pairings. A good time to do so is after a break, so that people who end up playing each other two games in a row have a chance to stretch their legs first. You may also use a criterion such as “whenever pairing without repeats would pair the leader out of his quartile”, though since that criterion isn’t yet implemented in tsh (because no one’s asked), you’ll have to set that up manually.
Many tournaments end with a king-of-the-hill (KOTH) round, whether earlier rounds are paired round-robin or Swiss. The final KOTH round is customarily paired with no restrictions on repeats (enter a number of permitted repeats one greater than in the previous round), and not Fontes (i.e. based on the results of the immediately preceding, not second preceding, round).
David Gibson, the all-time top money winner in the history of Scrabble, has made a habit of clinching victory in major events without waiting for the final round. Because of this, players are said to be “Gibsonized” when after clinching, they are paired with lower-ranked players to avoid affecting the ranking of runners-up.
tsh will currently detect Gibsonization situations for
one or two players, if the config gibson
option is set.
One Gibson player will be given a bye if the number of players in the
division is odd, or require manual pairing otherwise
(with the highest-ranked
player not likely to finish in the money, which is hard for tsh
to determine).
Two Gibson players will be paired with each other.
If you are using Fontes pairing, some Gibson situations may occur that tsh cannot detect. For instance, suppose (as happened at BAT 2005) that you have two players three games ahead of the field after round 13 in a 15-round event, and round 14 is paired based on round 12. When tsh computes round 14 pairings, those two players may only have had a two-game lead (as of round 12) with three games to go. You should therefore always check in late rounds to make sure that a Gibson situation has not developed, and manually re-pair top players if necessary.
tsh Gibson detection is currently based solely on wins. At the 2004 NSC, the NSA used spread Gibsonization with thresholds of 500 points over one game, 800 over two and 900 over three. If these values become popular, tsh will adopt them in a future version.
I usually begin tournaments with a fixed three-round schedule that picks players for round-robin quads at random from each quartile, then pair a certain number of Swiss rounds (regular Swiss after session breaks, Fontes Swiss mid-session), allowing repeats at some point, then end with a KOTH allowing one more repeat than in the preceding round.
That does mean that after a break, two consecutive rounds are paired based on the same round’s standings, but since the program checks to avoid repeat pairings, this doesn’t cause trouble. If you use this scheme though, do not increase the number of repeat pairings permitted in either of the first two rounds of a session: always do so only in the third or fourth round. Increasing pairings at the wrong point in a session can lead to large numbers of back-to-back repeats and player complaints.
See also the section on Chew Pairings, currently under development.