The Deuce Rating, explained

Rating and rung are different things

Players mix these two up constantly. The difference takes thirty seconds to learn and it saves an argument later.

Your rung is the thing you are playing for. The rating is only a measurement taken alongside it. Deuce keeps the two apart on purpose, so that no player ever loses a rung to arithmetic they cannot see. Rung movement is governed entirely by the rules, applied the same way to everyone.

One more thing to fix in your head before the rest of this makes sense. Ratings live on the {{RATING_SCALE}} scale, and a player who arrives with no evidence whatsoever starts at {{RATING_INITIAL}}. Nobody should read that starting number as an opinion about their tennis. It is where the estimate begins before any evidence arrives.

The separation runs in the other direction too. Your rating is not a tiebreaker for rung order, and it never will be. When two players finish a season level, the five-step order published on the scoring page settles it, and every step in that order is something both players can recompute from results they can see.

The split shows up in two places players notice. A walkover moves rungs exactly as a played win does, and it leaves your rating where it was, because nobody learned anything about how you play tennis. A retirement does count, using whatever part of the match got played.

The algorithm, by name

The Deuce Rating is calculated with {{RATING_ALGORITHM}}.

We name it because hardly anybody does.

Of eleven tennis rating and league platforms audited in July 2026, one stated in public which algorithm produced its rating. None published a per-player confidence figure.

Source: Deuce Ladder competitive research, audit of eleven tennis rating and league platforms. Captured July 2026, next recheck October 2026. This is our own research and not a third party's, so here is the test we applied: a platform counted as naming its algorithm only if its public documentation gave a model by name, in writing, without a login. Calling a rating proprietary, advanced, or results-based did not count. The platform list is available if you ask for it, and any of these companies can change this about themselves tomorrow.

The rest call their rating proprietary, or advanced, or say nothing about it at all. Which is an odd thing to guard, when you think about what is actually valuable here. Nobody is going to beat us with a copy of our update rule, because the moat in this business has always been the match data and not the math applied to it. Publishing the math costs nothing, and what it buys is the one thing that makes a rating useful to anybody, which is your belief in it.

Deuce publishes more than the name. The parameter values, the update rule in plain language, and a worked example you can check with a calculator all sit on the rating methodology page, versioned, with a changelog entry for every revision. A named algorithm with hidden constants is half a disclosure.

That is the first of the two commitments this page makes. The second one is harder to build and it is the one nobody else does.

Confidence, published next to the number

Every Deuce Rating carries a confidence figure, and it appears wherever the rating appears. Same line, same visual unit. Not buried in a settings screen.

A rating built on two matches and a rating built on forty are not the same kind of object. A system that prints them in the same font is misleading you. The figure answers one question: how much should you trust this.

On your profile and in the standings, the two print as one unit. High confidence renders as {{RATING_EXAMPLE_HIGH}}, medium as {{RATING_EXAMPLE_MEDIUM}}, and low confidence as {{RATING_EXAMPLE_LOW}} with a provisional label attached, on the grounds that a range is the honest way to print a number the system is still guessing at. No view in the product gives you the rating without the figure beside it, and neither does the API.

It runs on three things: how many results you have, how recent they are, and how varied your opponents have been. That third one catches people out. A player with twenty matches against the same three opponents has less information sitting behind their rating than a player with twelve matches against twelve different opponents, and the figure reports the difference.

The band boundaries are numbers, not moods. Low becomes medium at {{CONFIDENCE_LOW_THRESHOLD}} and medium becomes high at {{CONFIDENCE_HIGH_THRESHOLD}}, both published on the methodology page. What we will not print is a match count, because a match count on its own does not get you there. Play eight matches against the same two opponents inside one month and you can still be sitting in the low band, correctly.

Provenance counts here too. A result both players confirmed, or one an organizer countersigned, is better evidence than one that confirmed itself on a timer. The five tiers and their definitions live on the scoring page, which owns them. At launch the tier shows up in your confidence figure rather than in the rating arithmetic itself, and the reason is on the methodology page: a rating movement you cannot recompute is a rating movement you will not believe. Provenance never touches your rung.

Publishing that figure is the second commitment on this page. Going by the audit above, none of the eleven platforms made it. A rating with no uncertainty attached invites exactly one behavior, and that is treating a thin number as a firm one. It holds up fine until the Saturday somebody turns out to play well above the number on their profile, and by then the argument is about the algorithm rather than about the two matches it had to work with.

Low confidence never buys an easier opponent

This is a design commitment, not a feature. It is written down on a public page so that it cannot be quietly traded away in a planning meeting eighteen months from now.

A low confidence figure never puts you in an easier level band, an easier bracket, or a softer playoff draw. Ever.

Say it out loud and the reason is obvious. If uncertainty produced easier opponents, the smart play for anybody who wanted to win a season would be to stay uncertain. Post few results. Play inconsistently. Spread your matches across three ladders so no single record adds up to anything. At that point you have paid people to be hard to measure. Every sandbagging problem in amateur tennis is a version of that, and it is the easiest way there is to wreck a rating system built with good intentions.

So the mechanics run the other way.

• Placement uses the strong end of your range, not the middle and not the weak end. If the system thinks you are somewhere between a 3.5 and a 4.5, you get placed as a 4.5 until results say otherwise. Being unmeasured costs you the benefit of the doubt and nothing else.
• Low confidence widens the range upward faster than downward. An unproven player is assumed capable, not assumed weak.
• Bracket and band assignment reads the conservative bound, which is the stronger end. Confidence changes how the number is displayed and how cautious the placement is. It never changes which direction the caution points.
• The rule is enforced by a test, not by good intentions. A test in the build fails if placement ever reads the middle of a player's range or the bottom of it. That is deliberate, because this is the rule most likely to get softened one day by a reasonable-sounding ticket about how new players feel in week one.
• You can always ask to be placed higher. Asking for a harder draw is not a thing anybody games a rating system to do, so that request gets no defenses built around it.

The tradeoff is real, so here it is. A genuinely new player with no history may find their first match or two harder than necessary. Rule 5.4 in the rules is the fix. Your first challenge in a ladder is unrestricted inside your band, so you can correct a placement downward with one played match instead of waiting for the system to notice. A first match that is slightly too hard and fixable in a week is a cost we will take. A system that pays people to stay unmeasured is not.

Decay is continuous, not a cliff

Ratings go stale. Nearly every system in the sport handles that with a fixed window: results inside the last twelve months count, results outside it do not, and your rating is a hard function of whatever is left.

There is a hole in that design, and competitive players find it inside a season. Say your rating is high and your form has dropped. You wait. You play nothing. The window rolls forward, your good results sit in it a while longer, and you come back with a rating you no longer deserve. If your rating is low the trick runs the other way: a bad patch drops out of the window on a date you can count to, and you plan your return around it. Anyone who has played league tennis for a few years has watched somebody do this, and could tell you roughly which month they will reappear.

Deuce decays continuously instead. Every result starts losing weight the day it is played. Nothing falls off on a date, so there is no date to wait for.

That leaves you with two effects to get used to. Time away lowers your confidence figure before it moves your rating much, and that is correct. The system knows less about you than it did last spring, a different statement from saying you got worse. Your rating can also move a little on a day you played nothing. Also correct. The evidence behind it is older than it was yesterday.

Rating decay is not rung decay. Rung decay is a ladder rule with a published rate under section 12 of the rules, and it exists to stop players parking on a good rung. Rating decay is a property of the measurement. The two run independently, and a declared absence pauses rung decay only. Telling us you are away stops the ladder from costing you rungs. It does not stop the evidence behind your rating from getting older, because that is not the kind of thing a notification can change.

What the rating carries between ladders

Your rating and your rung have different lifespans, and the difference matters most to the person planning a winter off.

The rating and its confidence figure travel. Across ladders, across clubs, across the off-season, and out the other side of a two-year gap. Your match record travels with them, because the record is the evidence the rating is built from. The rung does not travel at all. It is a position in one ladder's standings and it means nothing anywhere else, although the standings row itself stays live at its URL after the season closes, which is a different thing from following you around.

Leave a ladder and the rating comes with you. Join a second one while the first is still running and both read the same number. One rating per discipline per player, never one per ladder, which is the whole reason the thing is worth publishing.

Rungs are the organizer's call between seasons. Rule 16.9 in the rules lets them reset the ladder completely, seed the new season from the old final standings, or something in between, and the ladder page says which before you sign up. Your rating takes no part in that decision. It carries straight through the off-season, still aging the whole time.

Singles and doubles are separate

You get a rating for each discipline. They are calculated independently and stored as two separate objects, not one number with a flag on it.

Blending them is the common shortcut, and it produces a number that describes nobody. The skills overlap without being the same skills. Every club has a singles player who is lost at the net, and a doubles specialist whose second serve does not survive a baseline rally. A blended rating misplaces both of them.

One limitation, and you should have it before you plan around anything. Your doubles rating is not published at launch. Doubles brings a problem singles does not have. Four people produce the result and it has to be attributed to each of them. Deuce records both partners and both opponents on every doubles result, and the attribution rule belongs to the {{RATING_ALGORITHM}} decision instead of being a separate system somebody invents later. Until that is settled and written up on the methodology page, doubles results are stored and no per-player doubles rating is displayed anywhere. A rating for the pair may well arrive before a rating for each partner, since a fixed pair is a stable thing to measure, and if it does the methodology page will say so before the product shows it. What we will not do is publish a per-player number we cannot yet explain, which would break the one promise this page is built on.

Doubles at launch runs as fixed pairs for the season, and the pair holds the rung, so partners climb and fall together. Rotating partners is on the roadmap rather than in the product.

Your doubles rating does not affect your singles ladder placement, and your singles rating does not affect your doubles placement.

How it lines up with NTRP, UTR, and WTN

You probably already have a number. Deuce reads it at signup and publishes an approximate mapping in both directions, so nobody has to guess what a Deuce Rating means in a system they already understand.

Seven inputs are accepted, and this is the full list: NTRP, UTR, WTN, an LTA rating, an ITN, a national federation ranking, or your club level. Every one of them is optional. Bring none of them and you answer a few plain questions instead, then the unrestricted first challenge corrects the placement inside one match.

The mapping is a published table of ranges, not a formula. The approximation is deliberate. The systems it maps to measure different things, on different scales, across different populations.

What each input is, and what Deuce does with it at signup.

A conversion table is where rating pages usually start overclaiming, so the small print goes in plain sentences.

The mapping is unofficial. It is Deuce's published estimate and nothing more. The USTA, UTR, and the ITF have not endorsed it, we do not call it equivalence, and it is better understood as a translation with error bars around every entry. It also runs in one direction more reliably than the other. Convert a Deuce Rating to NTRP and back and it will not always land where it started, because the ranges overlap, and anybody who tells you their conversion round-trips perfectly is mapping two scales that were never the same shape.

Whatever you bring seeds you and then stops mattering. Your first placement uses the number you arrived with; after a handful of confirmed results your Deuce Rating is driven by matches played on Deuce, which makes an imported rating a starting point rather than a permanent credential. And we cannot verify any of it. A self-reported NTRP is a claim, which is one reason placement reads the conservative end of your range, and one reason the unrestricted first challenge exists.

For organizers setting the level bands

Players can skip this section. Organizers use the number differently, and the difference takes about ten lines to explain.

You set the level bands on your ladder. The Deuce Rating is an input to that, not the decision. When you build a ladder, every player arrives with a rating, a confidence figure, and whatever they brought at signup, and you can sort on any of it. What you cannot do is hand the placement to the algorithm and blame it afterward. Hence the override.

You can move any player to any band, at any point, for any reason. The move and the reason are written to the ladder's public activity log, visible to that ladder's players. The visibility cuts both ways and it is meant to. It protects a player from being quietly reseeded, and it protects you from the accusation, because the record shows you wrote a reason at the time instead of reconstructing one in week nine when somebody finally asks.

A couple of traps, then. A low confidence figure on a new joiner is not a reason to place them down: the system already places at the top of the plausible range, and undoing that by hand is how a band acquires a sandbagger. And a small ladder produces correlated evidence, because everybody plays roughly the same handful of opponents, so ratings inside a ten-player ladder can drift together relative to the outside world. The confidence figure reports that. It does not fix it. The known-limitations section on the methodology page is blunter about this than a marketing page would be, and it is the section to read before you use ratings to seed a bracket.

Who can see your rating

Anyone. Better stated here than discovered in week three.

Every Deuce ladder has a public standings page, readable with no account and no login, and your rating and its confidence figure sit on it. So does your match record. Section 1.1 of the privacy page covers what a standings page shows, and it shows you the row before you join anything.

There are limits. The raw number you brought at signup is not published: your level band is public, the NTRP or UTR you typed in is not. And under section 5.2 of the privacy page you can take your display name out of search results, which replaces it with an initial and a surname in the public standings. Your row itself stays, because deleting a row would renumber the rungs of everyone below you and make the standings wrong for the whole ladder.

There is no way to hide your rating from the ladder you are playing in. A ladder is a public record of who beat whom, and a hidden rating inside one would be a number your next opponent has to take on trust without being allowed to look at it. That trade is not available.

What the rating is not for

Publishing a method means publishing where it stops working.

It is not your rung. Rungs move by the rules. The rating is a measurement that sits beside them.

It is not a qualification. A Deuce Rating does not enter you into anything, does not certify you at a level, and is not accepted by any governing body. If your league needs an NTRP, you still need an NTRP.

It is not a prediction for one match. It estimates a standard and states how sure it is. On any given Saturday the lower-rated player wins often enough, more or less the reason anybody bothers playing the match.

It is not tuned to make you feel better. A rating that only ever goes up would be a score rather than a measurement of anything. The confidence figure is published for that reason: it lets a number be honest and uncertain at once. More useful than a number that is precise and wrong in a flattering direction.

Check the math yourself

Everything above is the explanation. The proof sits one click away, on its own page, because most people reading this do not want the arithmetic and the few who do want every line of it.

The rating methodology page carries the parameter values, the update rule written out three ways, a worked example you can reconcile with a calculator, how the confidence figure is computed, the known limitations stated as limitations, and a version number with a changelog. It is written for the club committee member who has been asked to approve a number their members will argue about.

If that is you, forward that page instead of this one. And if a number in it does not reconcile, tell us. We will either fix the page or show you the query that produced it.

If your own rating looks wrong. Start with the match record. Every confirmed match stores the two ratings that went in, the two that came out, and the version of the methodology that produced them, so the answer to "why did that move like that" is on the record instead of in a support agent's head. If the record and the number disagree, that is a bug and we want it. If they agree and you still think the number is wrong, the honest answer is usually that the confidence figure beside it is already saying the same thing.

If we change the algorithm later. Two commitments, and a club deciding whether to adopt this should hold us to both. A parameter change or a model change is a methodology revision: it gets a version number, a date, and an entry in the changelog stating whether existing ratings were recomputed and, if they were, how far the typical rating moved. And no revision ever quietly rewrites history. Either the old ratings stand and the new method applies from a stated date, or history is recomputed and the changelog says so in the same release. What will not happen is a rating that changes overnight with no published reason, which is how a platform earns the accusation that the math was adjusted to suit somebody.

Where to go next

Playing. Bring whatever rating you have, or none, and get placed. Your first challenge is unrestricted, so a bad placement costs you one match. Find a ladder → More for players →

Organizing. You set the level bands. The Deuce Rating helps you place people, and you can override it, with the reason written to the public log. Start a ladder → More for organizers →

Checking. The full specification, with the parameters and the worked example, is on the methodology page.

Reading further. How a ladder works, how results enter the system, what a standings page shows, the terms used here, and the questions we get asked.