Elo vs TrueSkill

TrueSkill was built by Microsoft Research for Xbox Live matchmaking, to solve a problem classic Elo was never designed for: rating individual players from the results of team and multiplayer games.

If your group plays 4v4, or eight-player free-for-alls, this is the comparison that actually matters to you.

What Elo does

Elo is a two-side system at heart. It takes two ratings, works out an expected score, and moves both by the gap between expectation and result. The full explanation is here.

Applying it to teams or to a six-way game means adding rules on top — averaging team ratings, or decomposing a multi-way result into pairs. Those rules work, but they are choices made by whoever built the implementation, not part of Elo itself.

What TrueSkill does

TrueSkill represents each player as a distribution rather than a number: a mean skill estimate plus a measure of how uncertain that estimate is. In that respect it is close in spirit to Glicko-2.

What makes it different is that it was designed from the start for matches with many participants. Given the finishing order of eight players, or the result of a 4v4, it updates every individual's skill estimate according to the part they plausibly played in that outcome. It is genuinely inferring individual ability from team results, rather than handing everyone on the team the same adjustment.

It also converges quickly — a handful of games gives a usable estimate, which is why it works for matchmaking where a new player needs a fair game immediately.

Side by side

Which matters for your group

If you mostly play one-on-one — darts, pool, chess, head-to-head card games — Elo is the natural fit and TrueSkill's machinery is answering a question you do not have.

If you regularly play team games or big free-for-alls, then you need something deliberate, and this is worth understanding rather than hoping the default copes.

Custom Elo Games' answer is to extend Elo rather than adopt TrueSkill. A result can have up to ten sides, and it is scored by breaking the finishing order into every head-to-head pair within it, running each pair through the standard Elo formula, and dividing the K-factor by one less than the number of sides — so playing a six-way exposes your rating to the same total movement as a one-on-one, rather than five times as much. Ties are allowed anywhere in the order, including an all-square draw.

For teams, a side competes at the average of its players' ratings, and every player on that side receives the whole of the change. That is a deliberate choice with a consequence worth knowing: it means a rating always reads as strength per person, but it does not try to work out which team-mate carried the game. TrueSkill does attempt exactly that.

The trade is transparency for inference. With the pairwise approach, anyone can verify a single head-to-head step themselves on the Elo calculator. With TrueSkill, you take the model's word for it. For a group of friends arguing about a leaderboard, we think being able to settle the argument is worth more than the extra precision.