When most people think of tournaments, they imagine single-elimination brackets like March Madness. A bunch of competitors enter, play opponents along a predetermined path, and are eliminated if they lose. Last one standing is the champion. These brackets are simple, exciting, and popular. However, they are quite volatile. One bad game or unlucky bracket path can prematurely eliminate a top contender.
Round-robin is a common format used to address these flaws; it has low volatility and no elimination. Every competitor plays everyone else, and some aggregate score is used to determine the winner. However, they don’t scale well. The number of matches rises quadratically with the number of entrants; if you double the entrants, you quadruple the matches. From a spectator’s perspective, round-robin is far less exciting as each match has equal importance. Unlike elimination brackets, round-robin lacks the escalating tension into a climactic final match.
Many tournaments try to get the best of both by using round-robin pools to seed a single-elimination bracket. This hybrid approach is everywhere, from the NFL playoffs to Olympic badminton. However, this can lead to incentive-misalignment issues where competitors intentionally try to lose (or “not strategyproof” in mechanism design jargon). One famous incident was in 2012 Olympic badminton, where two teams both tried to lose in order to secure a favorable bracket path. They were both disqualified: unjustly, in my opinion. Intentionally losing may be unsportsmanlike, but I’d argue they were trying their best to win the real goal of a gold medal. Don’t hate the player, hate the tournament design.
Double-elimination brackets are a great solution to all of these issues. It’s exactly what it sounds like; an elimination bracket where you are out after you lose twice. This extra room for error creates more consistency. Strong entrants progress more reliably, and the final match is more likely to be a close match between two top competitors rather than a blowout. Initially popularized by the fighting game community, double elimination brackets have been adopted by other games like chess (shoutouts to Nintendude).
Other tournament formats abound, like Swiss-system (popular in chess and MtG) or gauntlets (a lopsided elimination bracket). Interestingly, it seems like any game can use any tournament format. Why?
Modularity
Tournaments work interchangeably with games due to a key assumption: two players enter, one player leaves. This abstraction allows tournament formats to work for any game that has this property. However, not all games are built this way. Free-for-all games like battle royales or Teamfight Tactics have more than two players per match. Wagering games like poker or Marvel Snap have outcomes of varying magnitudes. Some one-on-one games like Chess can end in a draw, with no defined winner.
Usually there is a way to wrap extra rules around the game to force a win/loss. If games are quick (ex: fighting games), you can play a quick extra round. If games are slow (ex: chess), you can tiebreak with other heuristics, like the number of games won as black. As a last resort, you can always flip a coin.
However, these wrappers can twist strategic incentives. For example, Teamfight Tactics are an 8-player free-for-all, and some tournaments allow the top 4 to move on while the bottom 4 are eliminated. There is no difference between placing 1st or 4th. This skews the strategy towards the median, with consistent, low-risk strategies shooting for 3rd or 4th crowding out riskier all-or-nothing plays.
If the game can’t fit well into this “two players enter, one player leaves” abstraction, then they need to invent a new tournament format. Poker has its own unique Multi-Table Tournament format, where tables are merged as players are eliminated until there is only one table left. Battle Royales like Fortnite usually use a point system, wherein higher placements and more kills award more points. Since these games don’t fit neatly into the standard assumptions, they can’t use the tournament designs that have been developed over the years.
I’ve been musing about this issue because Marvel Snap falls into the latter camp. Since Snap involves variable wagers, it does not create a binary win/loss. The Conquest game mode does wrap the base game into a heads-up poker format that does force a binary, but it often creates short-stack scenarios that are a drag to play out. Poker is a closer analogy, and perhaps something akin to multi-table tournaments would work. But Snap is exclusively one-vs-one, whereas poker features 10-player tables, so additional modifications would be needed to make it work. I love Snap’s innovative wagering system, but unfortunately it is not modular to most competitive games, and so many systems such as tournaments and Elo need bespoke modifications.