How the Picklr Rating works
Every player has one number, the Picklr Rating. It starts at 1000 and moves after every game. This page explains exactly how — no hand-waving, no black box.
The short version: PR is a modified ELO. It borrows the idea of rating confidence from Glicko and the idea of using the score margin from TrueSkill 2. Beating someone better than you gains more than beating someone worse. Losing narrowly to someone far better costs almost nothing. And the system knows the difference between a rating it has seen fifty games of and one it has seen three.
Three things it does that plain ELO doesn't
1. The score matters, not just the win
In plain ELO, 11–2 and 11–9 are the same result. In pickleball they obviously aren't. PR projects the margin it expected from the two ratings, then compares it to what actually happened. Win by more than expected and the gain is scaled up; scrape through and it's scaled down.
That scaling is bounded — between 0.5× and 1.5×. A win never costs you rating no matter how ugly the scoreline, and one blowout can't swing a rating further than a game and a half's worth of normal movement.
2. It tracks how sure it is about you
Alongside your rating, PR keeps a second hidden number: how much it trusts that rating. A brand-new player's rating is barely more than a guess, so it moves fast. After enough games it settles and stops lurching on single results.
This also decays. Stop playing for a couple of months and the system becomes less certain about you again, so your first games back move your rating more. It isn't punishing you for being away — it just genuinely knows less about your current form than it did.
3. Singles and doubles are rated separately
They're different games, so they get different numbers. Being carried by a strong partner in doubles will never inflate your singles rating.
Doubles goes further: a team's strength isn't the average of the two players. A weaker partner pulls the team's effective rating down harder than a stronger one pulls it up, because in real doubles the weaker player gets targeted. Beating a pair carried by one excellent player is worth less than beating two solid ones with the same average.
A worked example
Two players meet. One is rated 1200, the other 1000. Both are settled, well past their placement games.
The underdog's side mirrors it. Winning as the 1000 is worth a lot; losing narrowly costs very little. This is the whole point: the rating chases what surprised the system, not what merely happened.
Your first five games
New ratings move at double speed for five games, then settle to normal. It's the fastest way to get you roughly where you belong without a long grind — and why your number can jump around early and then calm down. Until you've played those five, you show as unranked rather than carrying a number nobody should trust yet.
Why you sometimes can't see the PR change on a game. If anyone in a match is still in their placement games, the per-game rating change is hidden for that side. An unsettled rating produces noisy numbers, and showing them invites people to read meaning into noise.
Ranks
Ranks are fixed bands, not a curve — you're never demoted because someone else improved. Gold is the average player. Legend is roughly the top fraction of a percent.
Seasons
Leagues run in seasons. When one ends, ratings are pulled part of the way back toward 1000 rather than wiped — you keep most of what you earned, but the gap at the top compresses and the season is worth playing. The system also becomes less certain about everyone again, so early-season games move more.
A league that only plays part of the year can pause this entirely, so an off-season doesn't churn through empty seasons nobody played in.
What PR does not claim
It isn't comparable between leagues. A 1400 in one league and a 1400 in another haven't been measured against each other — each rating only means something relative to the people in that league. If two leagues have never played, their numbers are two separate conversations.
It isn't a global ladder, and isn't trying to be. Rating systems like DUPR exist to put everyone on one scale for tournaments and seeding, which is a genuinely different and harder problem. PR answers a smaller question: among the people you actually play, who's playing well right now. Plenty of players sensibly use both.
It can't see anything you don't log. Games that never get entered never count.
Under the hood
For anyone who wants the actual constants:
| Value | What it is |
|---|---|
| 1000 | Starting rating for every new player |
| 0 | Floor — ratings never go negative |
| 1100 | Rating scale: a gap this size means about a 91% win expectancy |
| 100 | Base step size per match |
| 0.5 – 1.5× | Range the score margin can scale a result by |
| ±150 | Largest possible single-match swing |
| 350 | Starting uncertainty; settles to about half that |
| 2× / 5 | Placement speed-up, and how many games it lasts |
The rating is computed on the server, not on your phone, so every player's number comes from the same code on the same data. The client and server implementations are held to each other automatically — a test suite compares them across hundreds of generated cases and fails the build if they ever disagree, so a preview you see before saving a game matches what actually gets written.