Before a match, each player's rating implies an "expected score" — roughly, their odds of winning. A 1200 vs. a 1200 is a coin flip; a 1400 vs. a 1200 is roughly a 76% favorite. After the result, both ratings shift by the same amount in opposite directions — the winner gains exactly what the loser loses.

Beating a stronger opponent is worth more because your expected score was lower going in — upsets pay out closer to the full K=32, while a huge favorite winning as expected only nets a couple of points.

Ratings aren't a column in a database that drifts out of sync — every standing on this site is recomputed from scratch, in chronological order, from the full match history on every load. Fix a typo'd result or delete a bad game and every rating downstream of it updates immediately.

A rating earned months ago and never defended shouldn't camp the top of the board forever, so current rating (never your historical peak, never a past game's delta) drifts back toward 1200 the longer you've been away — after a 1-week grace period. Two things make it bite harder over time: the weekly decay rate itself accelerates the longer you're gone, and because it's a percentage of how far above 1200 you are, a big lead erodes faster than a small one.

One game snaps you out of decay entirely — it only ever affects what "current" means while you're on the sidelines.