Information about the true skill system

We use the TrueSkill™ (TS) system to estimate each player’s skill in every game. Its calculation is based on your performance in scored matches against other players, analyzed statistically. In the background, an expected skill (often called μ) is computed along with an uncertainty for that value (often called σ). The TS value itself corresponds to the skill level above which we can be 99.7% confident that your true skill lies (i.e., μ − 3·σ assuming a normal distribution). With each completed game, the system refines its estimate of your skill, so the TS converges over time.

Therefore, your TS will increase when your estimated skill rises (for example, by defeating a player with a higher expected skill—even if that player’s TS isn’t always higher) and/or when its uncertainty decreases (for instance, when results align with expectations and opponents’ skills are similar). Conversely, TS will decrease when your estimated skill drops (e.g., losing to a player with lower expected skill) or when uncertainty grows (for example, by playing against opponents with very different expected skills or having unexpected outcomes).

We offer an option to restrict game invitations based on a TS range, which typically prevents unexpected TS fluctuations after a match.

TrueSkill™ Documentation

Microsoft TrueSkill project page (see the detailed description and the original research paper “TrueSkill™: A Bayesian Skill Rating System” under “Publications”)
Moserware TrueSkill article