How the rating works, and how we prove it

The method

• Everyone starts at 3.000 on a 2–8 scale. Underneath it's an Elo rating: beat someone rated above you and you gain more; lose to someone below you and you drop more.
• One full point means something specific: a player rated 1.000 higher is expected to win about 76% of the time. In doubles, each side is the average of its two players.
• Only matches run by an Athlos referee move the rating. Self-reported scores never count.
• Your first 10 matches move the rating fast (K 64), then it settles (K 24). After a season reset, 3 placement matches use K 48.
• Where it was played matters: a rated tournament counts in full, a casual match counts 50%, and a club ladder match counts 10% toward the global rating (clubs keep their own ladder).
• Winning big counts a bit more, up to 1.5× for a lopsided win. Points the loser scores never add to the winner's gain.
• Playing the same people over and over counts less: within 30 days, the 2nd, 3rd and 4th+ meetings count 75%, 50%, 25%. That stops two friends farming points off each other.
• Singles, doubles and mixed doubles each also get their own rating.

Not in it yet: adjusting for suspicious results, partner and opponent context, and video. When they arrive, they'll be listed here first, and the monthly score will show whether they helped.

How we score it

• Before a rated match moves any rating, we log the Athlos chance that side 1 wins. If every player has a DUPR for that format, we log their DUPRs too. The chance uses only ratings from before the result; the result never feeds its own prediction.
• Each month we compute the Brier score: the average of (chance − result)², where the result is 1 for a win and 0 for a loss. Lower is better. Always saying 50% scores 0.25; a perfect predictor scores 0.
• DUPR doesn't publish a win-chance formula, so we give it its best case: we try every scale and use the one that scores best on that month's matches. Athlos gets no such help; its scale is fixed in advance.
• Two comparisons: matches where every player's DUPR was checked by Athlos, and matches that include self-reported DUPRs (shown separately, because a self-reported number can be wrong).
• We only compare once there are 200+ matches with both ratings. A gap under 0.005 is called a tie: at a few hundred matches, that can be luck.
• Voided matches are removed. A published month is never edited.

Monthly reports