How Elo ratings work

A rating is one number that estimates how strong a player is. Record a game and that number moves, up for the over-performers and down for the under-performers, based on what happened on court. After a handful of games it settles near a player's real level, which keeps matchups close and makes the standings mean something.

PickleFriend uses an Elo system (the same family of math behind chess and many competitive games) adapted for doubles play.

The score moves your rating, not only the win

PickleFriend asks by how much?, not only did you win?

• A blowout (11–2) moves ratings more than a nail-biter (11–9).
• Beating a much stronger player or pairing moves your rating more than beating someone you were expected to beat.
• Losing a close one to a tougher opponent barely costs you anything.

A team that grinds out tight wins all night climbs differently than one that wins a couple of routs and loses the rest.

That's the default point margin method. A group can switch to win/loss, where winners always gain; see Rating method below.

Doubles is shared between partners

In a doubles game the rating change for a pairing is split between the two partners. Equal-rated partners split it evenly; the more their ratings differ, the larger the share the lower-rated partner takes.

When a stronger and a weaker player win together, the result tells you more about the weaker player, so their rating reacts more. How far the split can tilt toward the lower-rated player is tunable; see below.

Short and timed games count less

A full game to 11 is a full game's worth of evidence. A game the clock cut off partway is not. So every game's rating impact is scaled by how much of a full game was played:

• Rally scoring: the fraction is the winner's score over the target. A rally game to 21 that was called at 15–9 counts about 70% of a full game.
• Timed rounds (traditional scoring): the fraction is the game time (the Time set for the round, not counting any walk-on time) over your group's typical game duration (a Group Settings value). Fifteen-minute rounds in a group whose typical game runs 25 minutes count 60%.

On a short clock, that fraction is the same for every game in the round, whether or not it finished. An 11–7 that beat the buzzer and a 10–7 that didn't both count the same. That's deliberate: if finishing were worth more, a team could shrink a loss just by slowing play until time ran out. The discount belongs to the short format itself, and it never reorders anyone within a session; it only makes a short-format session count less than a full-length one, which is the trade you make by running a short clock.

A few edges:

• Games at or past full length count in full. A win-by-two game that runs past its target can count slightly more than a standard game.
• Low-scoring cut-offs get a margin assist. If the clock ended a game whose scoring was running well behind pace, both scores are projected forward so the margin isn't understated, but the game still counts by its time fraction.
• Because typical game duration sets the weight of every timed game, it's worth measuring rather than guessing. A value set to double what your games really take silently halves the weight of all your timed play.

What "no rating" means

A blank or no rating is not the same as a rating of zero.

• No rating means the player hasn't been rated yet: they're new, or they're in a group that doesn't use ratings. PickleFriend leaves them blank on purpose and supplies a sensible starting number the first time they record a game.
• Zero would be an actual position on the scale, the bottom. PickleFriend never quietly turns "unrated" into zero.

This is why a fresh player shows no number until their first game, then enters near the middle of the scale rather than at the floor.

The settings you can change

Every group can tune how its ratings behave. On a ladder, these settings are ladder-wide; they apply across every division.

Setting	What it controls
Scale	How far apart two ratings have to be to predict a given win chance. At the default (400, on the 1500-point scale) a ~200-point gap predicts the stronger team winning about 75% of the time. A smaller scale makes the same gap count for more, so it predicts more lopsided results; a larger one packs ratings closer together for the same odds. Scale pairs with the default rating, so Group Settings suggests a matching value and shows live examples like "1400 vs 1600 = 75%".
Base K-factor	How big each game's adjustment is: the volatility of the ratings. Turn it up and ratings respond fast to recent results; turn it down for slow, stable numbers. It's auto-scaled for doubles play, with a recommended value that tracks your scale.
Rating method	Point margin (the default) or Win/loss. Point margin compares the score with what was expected; win/loss makes sure winners always gain. See Rating method below.
Highest player responsibility (partner split)	A floor on the smallest share of a doubles rating change the higher-rated partner can take. Equal-rated partners always split 50/50; as the rating gap widens the stronger player's share shrinks (they were expected to win, so the result says less about them), but never below this floor. The default is 30%, which partly shields the stronger player's rating; raise it toward 50% to keep the split closer to even, even when partners are far apart. This one setting also feeds the Mixer: it supplies the default partner-balance weight the Mixer uses when estimating a team's combined strength to balance courts, so it's the single setting for "how much the stronger partner counts," shared by both the rating math and the mixer.

The defaults are tuned for everyday play. These controls exist for the organizer who wants a faster-moving ladder, a 1–10 scale, or a gentler curve for a brand-new beginner group.

Rating method

Group Settings → Ratings → Rating method picks how one game moves the ratings.

• Point margin (default). Ratings move by how the score compares with what was expected. If a strong team was expected to win 11–4 and only won 11–9, they did worse than expected and can lose a little, while the underdogs who lost narrowly can gain a little.
• Win/loss (winners always gain). Winners always go up and losers always go down. The margin still decides by how much: winning by more than a typical win moves ratings more, a narrow win less. An expected win moves ratings a little; an upset moves them a lot.

Win/loss feels fairer to some players, but it uses less of each score, so ratings wobble more from game to game and, in regular (side-out) scoring, take longer to settle after a new player joins. On real games the two predict results about equally well, so the choice is mostly about what feels fair to your players. Pick win/loss if "we won and still dropped" causes arguments, for example on a ladder where every win should count. Keep point margin if your players are happy for the score to do the talking. Your base K-factor means the same thing under either method. Switching only affects new games; Recalculate All Ratings replays your group's past games under the method you've chosen.

Auto-calibration: matching the scale to your numbers

Default rating, scale, and base K-factor are linked. If you change the size of the numbers (seeding players on a 1–10 scale, or importing ratings from another system), the scale and K-factor need to move with them, or the math drifts: a 200-point gap means one thing on a 1500-point scale and something very different on a 1–10 scale.

You don't have to work that out by hand. In Group Settings → Ratings:

1. Set your default rating to the middle of your players' range.
2. The Scale and Base K-factor fields each show a blue, underlined Recommended value derived from that default. Click either number to fill it in.

The Scale field also shows live win-probability examples, like "1400 vs 1600 = 75%", so you can sanity-check the spread before you save. If you set a custom rating, or you're about to enter scores while some players sit outside the standard rating band, PickleFriend shows a tip that points you back to this same calibrator.

Starting point and the default rating

Every player who hasn't played a rated game yet shows no rating; that blank is intentional (see What "no rating" means above). The moment a score is recorded, PickleFriend supplies a starting number and begins updating from there.

That starting number is the group's default rating, which is 1500 by default and can be adjusted in Group Settings → Ratings. A higher or lower number doesn't change the math; it shifts where "the middle of the pack" sits on the scale. If your group uses a small 1–10 scale you'd set a default near 5; if you're mirroring a DUPR-adjacent scale you'd set it somewhere in that range instead.

How ratings display

The number you see is rounded depending on the size of the rating:

Rating magnitude	Decimal places	Example
Less than 10	3	4.207
10 to 99	2	42.07
100 or higher	1	1500.3

This keeps small-scale ratings readable (a 1–10 group needs three decimal places to show any movement at all) without cluttering large-scale ratings with meaningless extra digits. On the ladder standings table, ratings are always shown to 2 decimal places regardless of scale, because that view is designed for comparison and the extra precision helps.

Individual ratings vs. team ratings

Everything above describes how PickleFriend tracks individual ratings: one number per player, updated after every game. That's the default for all groups, including ladder leagues.

If your group uses fixed partnerships and you want the team to be the rated entity rather than the individual players, there's a separate Team Rating Mode. In that mode, ratings are stored per partnership, standings show team numbers, and the mixer uses the team rating to balance courts. Individual players still accumulate a secondary per-player rating, but it doesn't drive anything.

See Team ratings for how that mode works.

What's next

• Placement games: new players settle in over their first several games.
• Organizer score entry: ratings only move when scores get recorded.
• Start ratings: how movers get seeded when joining or moving between divisions.