How the rating works

Your rating is an estimate of how strong you are, not a score you accumulate. It is not points for showing up: playing more sets does not push it up on its own, and sitting out does not pull it down. Everyone starts at 1500, and from there the only thing that moves it is the result of a rated set.

The ladder runs Glicko-2. The short version is that it does the same job as the Elo rating most people have met before — predict who should win, then correct itself when it was wrong — but it also keeps track of how sure it is about each player, which is what the next two sections are about.

Before a set, the ladder works out how likely each side is to win from the two ratings. The result is then compared against that expectation, and the size of the correction is the size of the surprise.

• Beating someone rated far above you is a large surprise, so you gain a lot and they lose a lot.
• Beating someone far below you is what was supposed to happen, so almost nothing moves.
• An even matchup sits in between, and both sides move a moderate amount.

Two limits sit on top of that. A single set can never move a rating by more than 30 points, however lopsided the upset — one result should not undo a season. And it can never move it by less than 1: every confirmed set moves the winner up and the loser down, so a set you actually played never reads as nothing.

Alongside your rating, the ladder keeps a second number: how confident it is that the rating is right. A new account starts with that confidence very low, because 1500 is a guess rather than a measurement.

While it is low, results move your rating in large steps — that is the ladder finding your level quickly instead of making you grind toward it. As you play, the estimate tightens and the steps get smaller. Two players can therefore come out of the same set with very different changes: the newer account moves more, because there was more to learn from it.

Time loosens it again. After a long break the ladder becomes less sure you are still where you were, so a comeback set counts for more than it would have if you had never stopped.

Your rating is hidden, from you and from everyone else, until you have played 5 rated sets. A rating built on one or two results is mostly noise, and publishing it would invite people to read meaning into a number that does not have any yet. You are still matched and still rated during that time — only the display waits.

Ranks are bands of rating, so you move between them by playing rather than by claiming one.

A rank is only shown once you are past provisional, for the same reason the number is. The band the starting rating sits in is where most players settle.

The ladder keeps three separate ratings, and a set only ever moves one of them. They never mix: a bad run in practice cannot touch the number on your profile.

• Ranked — the main one, the one on the leaderboard.
• Practice — for sets you queue with practice mode on, for trying a character you have no business playing yet.
• Randitto — its own ladder, because being handed a random fighter measures something different from playing your main.

Friendlies are not rated at all. Neither are sets that end without a confirmed result: if a match is cancelled, or expires because nobody reported it, both ratings are left exactly where they were. Sets played at start.gg tournaments are not rated here either — this ladder only rates what happens on it.

A season is a rating period. When one ends, everyone is reset for the next, so a season is a fresh field rather than a permanent record — and a rating earned in one season is not comparable to one from another. Season 1 runs on Glicko-2; the preseason before it ran on plain Elo, which is why older numbers moved differently.