How Elo, Glicko, and TrueSkill Power Fair Matchmaking in Esports

On September 26 at 19:00, the Chinese team won the first esports gold medal at the Hangzhou Asian Games by defeating Malaysia 2‑0, and the article emphasizes that beyond team skill and strategy, the matchmaking mechanism is a key factor in ensuring fair and exciting competition.

1. Elo Rating System

The Elo system, originally created for chess, assigns each player a numerical score representing their skill level. Before a match, the system predicts each player's win probability based on the rating difference; equal ratings yield roughly a 50% chance for each side, while a higher rating increases the expected win rate.

After a game, a player's Elo rating is updated according to the actual result. A lower‑rated player who defeats a higher‑rated opponent gains more points, whereas a higher‑rated player who wins as expected gains only a small amount. The update formula incorporates a weight factor (often 16, 32, or 64) that limits how much a single match can change a rating.

In practice, players start with an initial rating (e.g., 1500). Over time, the system continuously refines each rating to more accurately reflect true skill, making Elo a common foundation for player matching and ranking in many competitive games, though it has limitations for team‑based titles.

2. Glicko System

To address Elo’s shortcomings—especially handling players who are inactive for long periods—Mark Glickman introduced the Glicko system. In addition to a skill mean, Glicko tracks a rating deviation (RD) that measures uncertainty about a player’s true skill.

When a player has not played for a while, their RD increases, indicating greater uncertainty. Upon returning and winning games, the player’s rating can rise quickly as the system gains confidence in their skill level. Glicko also includes a volatility parameter (tau) that reflects how rapidly a player’s skill may change.

Glicko matches players whose ratings and RDs are similar, and it can produce more accurate rankings by accounting for uncertainty, encouraging regular play to keep RD low.

3. TrueSkill Algorithm

Developed by Microsoft Research, TrueSkill extends these ideas for multiplayer and team games. Each player is modeled with a Gaussian distribution defined by a mean skill value and a standard deviation representing uncertainty.

During a match, each player’s performance is treated as a random variable drawn from their skill distribution. After the game, Bayesian inference updates both the mean and the standard deviation based on the outcome, using factor graphs and the sum‑product algorithm. TrueSkill also supports a newer version, TrueSkill 2, which adds further enhancements.

All three rating systems aim to accurately estimate player skill and pair opponents of comparable ability, thereby fostering a fair and competitive gaming environment.

Next time you step onto the King of Glory battlefield, remember that sophisticated mathematical algorithms are working behind the scenes to keep each match fun and challenging! Author: Wang Haihua