How Does Tournament Seeding Work? Designing Who Meets Whom

A single elimination bracket is efficient, needing only one fewer match than the number of entrants to produce a winner, and it is a poor instrument for ranking. It identifies a champion with reasonable reliability and tells you very little about second place, since the runner-up is simply whoever the eventual winner beat last, and the genuine second best competitor may have been eliminated early by meeting the winner in round one. It also amplifies chance: if each match has any element of randomness, the probability that the strongest entrant survives several rounds falls multiplicatively, so a strong competitor with an eighty percent chance in each of five rounds wins only about a third of the time. Round robin formats, where everyone plays everyone, measure ranking far better and cost a number of matches that grows with the square of the field, which is why large events use group stages followed by a knockout, buying some of the reliability of one format and the brevity of the other.

The standard scheme ranks entrants and places them so that the highest ranked meet as late as possible, which pursues several goals at once:

• •Rewarding earlier performance, since a higher seed earns an easier early path, which gives competitors an incentive to perform in qualifying and in the regular season
• •Protecting the final, so that the two best entrants are on opposite sides of the bracket and can only meet in the last match
• •Producing an escalating schedule, with the strongest matches concentrated late, which is what broadcasters and spectators want
• •Reducing the chance of a strong competitor being eliminated by another strong one in round one, which most people regard as unfair to both
• •Byes, extra rest given to top seeds when the field is not a power of two, which is another form of reward and another source of complaint
• •In practice the seeds are assigned to fixed bracket positions, so the top seed is placed to meet the lowest, the second seed to meet the second lowest, and so on, which is standard seeding

Analysis of seeding has shown that the usual goals conflict formally. Standard seeding maximises the chance of a final between the top two, and it does not maximise the chance that the strongest competitor wins the tournament overall, because it can hand a middling seed an easy path while forcing strong seeds through each other earlier. Schemes designed to maximise the probability that the best entrant wins look different and produce brackets that seem unintuitive. Delayed confrontation, which keeps high seeds apart as long as possible, is not the same objective as monotonicity, the requirement that a better seed should never have a worse chance than a worse seed, and it is possible to construct cases where improving a competitor's seed lowers their probability of winning. There is also an incentive problem: if the reward for finishing higher in a group is a favourable bracket position, a team may benefit from losing a match deliberately, which has occurred visibly enough to cause disqualifications and rule changes.

Formats other than a seeded single elimination address different priorities. Double elimination gives every entrant a second chance and produces a more reliable result at roughly twice the matches, which is why it dominates competitive games and several combat sports. Swiss system pairing, used in chess and elsewhere, matches players with similar records each round without eliminating anyone, and after a number of rounds roughly equal to the logarithm of the field it separates the leaders reliably while everyone plays the same number of games. Group stages followed by a knockout sample more matches per entrant before the decisive phase. Where seeding must be computed rather than assigned by reputation, rating systems do the work, including the Elo system devised for chess and its descendants, which estimate strength from results and update after each game, and which are now used across sports, esports and matchmaking in online games, where the objective is not ranking at all but giving every player close contests.

A knockout bracket finds a champion cheaply and ranks badly, since the runner-up is only whoever lost the final and randomness compounds across rounds. Standard seeding places top entrants far apart to reward past performance, protect the final and escalate the schedule. Those goals conflict formally: maximising the chance of a top-two final is not the same as maximising the chance that the best competitor wins, and better seeding can occasionally lower a competitor's odds. Double elimination and Swiss pairing trade extra matches for reliability.