The Mathematics of Matchmaking: Why Glicko-2 and TrueSkill 2 Don't Force a 50% Win Rate

Players queue up for their tenth ranked match of the night, convinced the system is actively working against them. The prevailing theory in every competitive gaming community—from tactical shooters to digital card games—is that matchmaking algorithms are designed to force a 50% win rate, artificially feeding players unwinnable games to keep them addicted to the climb.

On the other side of the screen, the mathematicians and system designers who build these architectures maintain a completely different reality. To them, the algorithm has no interest in a player's emotional state or engagement metrics; it is a cold, Bayesian inference engine designed to do exactly one thing: reduce mathematical uncertainty about a player's true skill as rapidly as possible.[3]

The stakes of this disconnect are massive. Player trust in competitive integrity is the foundational currency of the multibillion-dollar esports industry. When a player believes the system is rigged, they churn out of the ecosystem. When a system is actually flawed, the competitive ladder collapses into chaos. To understand the chasm between player perception and mathematical reality, one has to look at the engines actually powering modern competitive ladders.[3]

The two most prominent architectures are Mark Glickman's Glicko-2 and Microsoft Research's TrueSkill 2. Both are evolutions of the Elo rating system originally designed by Arpad Elo for chess, but they introduce a critical variable that the original chess model never had to account for: mathematical uncertainty.[1][2]

In a traditional Elo system, a player has a single number representing their skill. If they win, it goes up; if they lose, it goes down. But in the model published by Mark Glickman in 2001, a player's rating is not a single number. It is a range. The Glicko-2 system tracks a core rating—the system's best guess at the player's skill, which defaults to 1500 for a new account—and a Rating Deviation (RD), which measures exactly how confident the system is in that guess.[2]

When a player first creates an account, their RD is massive, defaulting to a value of 350. The system has no idea if they are a future professional or someone who has never held a mouse. Because the uncertainty is high, the algorithm allows their rating to swing wildly after every match. A win might grant 100 points; a loss might deduct the same. The system is aggressively searching for the bracket where the player stops dominating.[2]

TrueSkill 2, developed by Microsoft Research for franchises like Halo and Gears of War, takes this concept of uncertainty and scales it across teams. Where Glicko-2 was optimized for one-on-one games, TrueSkill 2 tracks a variable called Sigma for every player in a multiplayer lobby.[1]

TrueSkill 2 does not just look at wins and losses. It ingests squad membership, individual kill counts, tendency to quit, and even performance in other game modes to shrink that Sigma value as fast as possible. By feeding the algorithm more granular data, the system's predictive power jumped significantly. "TrueSkill2 predicts historical match outcomes with 68% accuracy, compared to 52% accuracy for TrueSkill," the Microsoft researchers noted in their 2018 technical report.[1]

This is where the "forced 50% win rate" illusion is born. As a player grinds hundreds of matches, the system's uncertainty parameter—whether RD or Sigma—shrinks. The algorithm becomes mathematically certain of exactly how good the player is.[3]

Once that uncertainty approaches zero, the system stops awarding massive rating swings. It locks the player into a bracket where they are statistically perfectly matched against their opponents. And when two perfectly matched teams collide, the outcome is a literal coin flip. The 50% win rate is not a hardcoded objective; it is the natural mathematical byproduct of a system that has successfully identified a player's exact skill ceiling.[3]

Players interpret this stabilization as the system trapping them in a manufactured purgatory. They feel they are improving, but the algorithm is so confident in its historical assessment that it requires an overwhelming, sustained streak of victories to widen the uncertainty parameter again and allow upward mobility.[3]

To combat this stagnation, both Glicko-2 and TrueSkill 2 incorporate a volatility metric. In Glicko-2, if a player suddenly starts performing wildly outside their expected range—say, by beating several highly-rated opponents in a row—the system's volatility parameter spikes. This tells the algorithm that its previous certainty was wrong, widening the RD and allowing the player's rating to climb rapidly once more.[2]

Similarly, if a player logs off for six months, the system artificially inflates their uncertainty parameter. It assumes their skills have degraded or changed, meaning their first few matches back will result in larger rating swings until the system recalibrates.[2]

The friction stems from a difference in goals. Players want a progression system—a ladder they can climb through sheer persistence and time invested. Matchmaking algorithms are not progression systems; they are measurement tools. They are designed to find a player's exact skill level and keep them there until the player genuinely, measurably improves. The algorithm will keep searching for the perfect match, and the players will keep blaming the algorithm when it finds it.[3]