How ratings work

The leaderboard is an Elo ladder calculated from match results from rated tournaments. Nothing on it is entered by hand.

The short version

Everyone starts at 1000. Each rated match moves both players' ratings by an amount that depends on how surprising the result was.

What counts

At this time, only Monthly events are rated: events with "Monthly" in the name, which an admin can override. Other events still count for the metagame.

The formula

E = 1 / (1 + 10^((Rb - Ra) / 400))
new rating = old rating + K x (actual - expected)

actual is 1 for a win, 0.5 for a draw, 0 for a loss. A double loss scores 0 for both. Both updates use the ratings from before the match.

K factor

Every match uses K = 32. There are no placement matches: everyone enters the format on equal footing, so a first match moves a rating by the same K as any other. K is multiplied by the event's weight, currently 1.0 for all events. Admins can change these settings.

Recalculated, never edited

The whole ladder is replayed from match results whenever anything changes. Results are stored against handles, not people, so linking two handles to one player merges their ratings without changing any results.

Leaderboard

The leaderboard covers the rated events between two dates an admin sets, both included; with no end date it runs to the newest event. A player is ranked once they have played in a rated event between those dates. A bye is not playing. Everyone else who has played in an event is listed after the ranked players as unranked, so the whole community is on the page.

Anomalies

Bad data is recorded per rating run, not thrown away silently: a player matched against themselves or a repeated match is skipped; a game count that contradicts the result is flagged but still applied; a rating change larger than K is flagged as a bug.