How we compute the ranking

Where the data comes from

Everything comes from the public standings of official UVS Games events, the organiser of Riftbound organised play. These are the same pages anyone can open on their locator: no private data, no account, no scraping of gated areas. We read entry lists, round-by-round pairings and results. We use no data that players have not already made public by entering an official tournament.

How the rating is built

The score is an Elo rating computed match by match across 1,583,335 games: every win moves the rating by how strong the opponent was, so beating a 1900 counts for more than beating a 1400. Everyone starts from the same value. Players enter the ranking after at least 5 official events — today 47,516 players: below that the number is still close to the starting value and means nothing.

Ratings are recomputed twice a day, at 06:50 and 20:50 UTC. Between runs the scores do not change, so reloading does not help. Every Sunday evening a snapshot of the ranking is stored: it powers the position arrows, which compare where you stood seven days ago.

Strength of field and games against the best

For every tournament we work out the average rating of who actually turned up: today we know it for 40,732 tournaments out of 51,127. That is the «Field» column, and it matters because the same score can be built in opposite ways — grinding eight-player events against unrated opposition, or playing 200-player regionals. Elo alone cannot tell them apart.

The «vs top 5%» column answers the same question match by match: how many wins against players rated above 1700, which today means 2,519 players out of 47,516. It is the difference between being strong and being the strongest at your local store.

Why there is a ± next to the rating

It is the uncertainty of our estimate, and it has two causes. The first is statistical: with few matches the number is a guess, with hundreds it is a measurement — the standard error goes from ±60 at thirty matches to ±13 at seven hundred. The second is time: a player's real strength changes, so a rating that has not moved in six months is worth less than yesterday's, even with the same number of matches. The two add in quadrature. A wide margin on someone with many matches means they have not played in a long time. It is not Glicko-2's RD — we use Elo, and this is the error of our own estimate, which is what we can honestly say.

The forecast on live tournaments

On live tournament pages we publish a win probability, computed by simulating the remaining rounds thousands of times. It is worth knowing what it is worth: across our history the player favoured by the rating wins 60.7% of matches, and the Brier score is 0.235 against the 0.25 a coin flip would get. The signal is real, but modest: it is there to size up a tournament, not to tell you how it ends.

What these numbers do not say

• We do not cover every tournament yet. Our archive grows daily, but tournaments we have not downloaded simply do not count: a player who is very active in a poorly covered country will look weaker than they are.
• Nationality is inferred, not declared. UVS does not publish players' countries: we work it out from where they play their local events. It works for anyone with a local circuit, and gets it wrong for people who moved or only play international events.
• We do not know what deck almost anyone played. Decklists are only published for players who make the cut, so we cannot say whether a result comes down to the player or the deck.
• A rating is a snapshot, not a promise. It measures how past tournaments went. Someone improving will look weaker than they are; someone who stopped playing will look stronger.