The DUPR algorithm produces results that feel wrong on a regular basis. You win and drop. You lose and climb. You play well for a month and the number will not budge. None of that is a bug. Once you see how the algorithm actually decides, the strange outcomes turn into a system you can work with.
This is the mechanics guide. For the plain-English overview of the rating first, start with what DUPR is, then come back here for how it is calculated.
DUPR is a modified Elo system
DUPR borrows its core from Elo, the same idea used to rate chess players. Before a match it looks at all four players' ratings and predicts a score. After the match it compares what happened to that prediction. Beat the prediction and your rating rises. Fall short of it and your rating falls. The raw win or loss barely matters on its own. What matters is the gap between expected and actual.
That is why a heavy favourite can lose rating in a win. If the model expected an 11-2 result and you only won 11-8, you underperformed the prediction, and the algorithm nudges you down.
The four things that move your rating
Every logged match is weighed on four factors.
Factor | How much it weighs | What it means |
|---|---|---|
Performance vs expectation | Heaviest | Did you beat the score the algorithm predicted from all four ratings? |
Match type | High | Tournament and club results count more than a self-posted rec score. |
Recency | Medium | Recent matches weigh more. Old results fade. |
Match volume | Builds over time | More logged matches make the rating more confident and stable. |
- Performance versus expectation. The biggest lever. Did you beat the predicted score?
- Match type. Tournament and club results carry more weight than a self-posted rec game.
- Recency. Recent matches count for more, and old results fade.
- Match volume. The more you log, the more confident and stable the rating becomes.
Margin of victory counts
DUPR reads the score, not just the result. Every point matters. Beating a matched team 11-3 moves you more than scraping past them 11-9, because the algorithm treats the wider margin as stronger evidence. This is why running the score up against players near your level is legitimately good for your rating, and why coasting once you are ahead leaves points on the table.
The July 2025 update: you can move either way
For years the rule of thumb was win up, lose down. In July 2025 DUPR changed the algorithm to a point-level, performance-versus-expectation model, and that rule no longer holds. In DUPR's own words, if you score more than expected your rating can rise even when you lose.
So a 3.5 who is predicted to lose 5-11 to a 4.0, and instead loses 9-11, can gain rating from that loss. The flip side is the favourite who wins by too little and drops. It is the same principle in both directions: the algorithm scores your performance against the prediction. We break down the change in our look at DUPR's new features, and it is the reason most older advice about the system is now wrong.
The 2025 reset, and what the data showed
Alongside the algorithm change, DUPR ran a reset. Players could earn a fresh rating from a clean window of recent matches, with a minimum of eight games against at least two partners, and keep whichever was higher, the old rating or the new one. It was a chance to shed a stale number dragged down by years of old results.
The interesting part was the outcome. When DUPR published the numbers, the vast majority of players barely moved.
How much ratings moved in the reset | Share of players |
|---|---|
Less than 0.3 points | About 90 percent |
More than 1.0 point | Under 0.1 percent |
The takeaway is reassuring: for most people the existing rating was already close to right. We covered the full results in what the DUPR reset showed.
Singles, doubles, and the partner effect
DUPR keeps separate ratings for singles and doubles and never mixes them. In doubles it also has to account for your partner. The team's predicted score leans on the average of both partners' ratings, so playing with a stronger partner raises the bar you have to clear to gain. The hardest problem the algorithm cannot fully solve is telling, from a team score alone, who actually controlled the match. It is the main honest limitation of the system, which we get into in is DUPR accurate.
Why the algorithm sometimes ignores your results
If your rating will not move no matter how you play, the cause is usually not the algorithm misjudging you. It is your reliability score. When DUPR is not yet confident in your rating, it moves the number cautiously, so even strong results barely register. Get past roughly 60 percent reliability and the same results move you properly. This is important enough that we gave it its own guide: the DUPR reliability score.
What this means for your rating
Put the mechanics to work and the strategy is simple. Play opponents near or slightly above you so a good performance beats the prediction. Keep scores decisive. Play formats that report to DUPR. Play often enough to stay reliable and recent. That is the whole game, and we lay out the plan in how to improve your DUPR rating. If you are still working out where you stand, our guide to a good DUPR rating puts the number in context.
The algorithm is not judging whether you won. It is judging whether you did better than expected. Everything else about DUPR follows from that one idea.
Common questions
Why did my DUPR drop after a win?
You won by less than the algorithm predicted. Against a much weaker team, a narrow win counts as underperformance, so the rating can fall even though you took the match.
Can my DUPR really go up after a loss?
Yes, since the July 2025 update. If you lose by less than expected, for example keeping it close against a much stronger team, that overperformance can raise your rating.
How often does DUPR update?
Your rating recalculates after matches are logged, and the reliability side of the system updates weekly. Do not read too much into a single day.
Does the reset mean my old rating was wrong?
For most people, no. About 90 percent of players moved less than 0.3 points in the reset, which shows existing ratings were already close to accurate.