Level Change Simulator — why your number moved

Set the four levels, say how settled your rating is and how the match went, and this estimates how far a level system would move you — and why the move is that size. A new rating shifts roughly three and a half times as far as a settled one on the same result. It is a model, not any platform’s algorithm: every constant is printed here, because no booking app publishes its own.

The match

Your side

3.0
How settled your rating is

Systems of this kind hold a confidence figure alongside your level and use it to damp the change. A new account moves several times as far as a settled one on the same result.

3.0

The other pair

2.7
2.6

The result

How it went
The margin
Method

How the model works

This is a model of how questionnaire-seeded, reliability-damped level systems behave. It is not any platform's algorithm, because no platform publishes one — and every constant it uses is printed here so you can argue with it.

Your side's level is the mean of you and your partner; the other side's is the mean of the two opponents. The expected result is 1 ÷ (1 + 10 ^ ((opponents − you) ÷ 0.86)), which is the Elo curve moved onto a level scale. The 0.86 is the level-scale analogue of Elo's 400: if 0.1 of a level behaves like roughly 40 rating points then a whole level is roughly Elo's reference gap, and dividing by 0.86 rather than 1.0 makes the curve slightly steeper, which is closer to how these systems price a half-level mismatch.

The change is then base × reliability × margin × (result − expected), with result 1 for a win and 0 for a loss. Base is 0.075 of a level. Reliability is 1.6 for a new rating, 1.0 for a settling one and 0.45 for an established one — a new account moves roughly three and a half times as far as a settled one on the identical result, which is the single biggest reason two people can play the same match and see different-sized changes. Margin is 0.75 for a close result and 1.2 for a clear one.

The answer is deliberately shown as a range, from 0.7 to 1.3 times that figure, rather than a single decimal. The real systems are private; printing one number to two decimal places would claim knowledge this page does not have. Base, reliability, margin, the spread and the 0.86 are all our figures, chosen to reproduce the behaviour these systems visibly have. None of them is published anywhere, by anybody, including us until now.

Where this comes from

  1. How the Playtomic level system works Playtomic player help centre
  2. The Glicko system — rating uncertainty, and why a new rating should move further Mark E. Glickman, Harvard University

Rivals is not affiliated with any padel federation and nothing here is coached, reviewed or official. Where a number is our own judgement rather than a published figure, the note above says so and cites nothing. Method last checked 21 August 2026.

Why your level moved that far

Two mechanisms explain almost every “why did it move like that” complaint, and neither of them is a conspiracy. The first is the team average: these systems rate the pair, not the person, so a night where a strong partner carried you against a weak pair can leave your number flat or lower even though you played well. The second is expectation — you are paid for the difference between what happened and what the system already believed would happen, which is why a comfortable win over a weaker pair is worth almost nothing.

The third mechanism is damping, and it is the one people notice last. Platforms like Playtomic describe a reliability percentage that damps changes as you play more — the lower the percentage, the more your level moves on a single result (their help center sets this out). A new account is allowed to swing because the estimate is a guess; a settled one hardly moves, because it is defending an estimate with evidence behind it.

The arithmetic on this page is Elo’s, rescaled for levels: 0.1 of a level behaves like roughly 40 rating points, so a whole level is roughly the 400-point reference gap Elo uses, and the expectation curve here divides the level gap by 0.86 instead of 1.0 to sit slightly steeper. That is a modelling choice, printed openly so you can disagree with it — which is more than any platform offers about its own.

Questions

Why did my padel level drop after a win?

Two mechanisms do it, and both are in the math rather than in anyone’s bad intentions. First, these systems rate the pair: your team level is the average of you and your partner, so winning alongside a much stronger partner against a weaker pair can be read as a result your side was always going to get. Second, you are only paid for the gap between the result and the expectation — if the model had you at 90% and you won, there was almost nothing left to award, and any nudge from the other side of the calculation can take you below where you started.

Why does a new account swing so much?

Because the system knows nothing about you yet and says so with a reliability figure. While that figure is low, each result is allowed to move your level several times as far as it would move a settled one — the estimate is being dragged toward the truth as fast as the evidence allows. Once it has enough matches, the same win or loss barely registers. Wild early swings are the system working, not a sign that it is broken.

Can I raise my level by beating weaker players?

Barely, and it gets worse the wider the gap. Expectation-based systems pay for surprise, so a win the model already predicted is worth close to nothing, and once your expected win chance is up near 99% the award rounds to zero. Beating players well below you is a way to keep a level, not a way to climb one. Climbing needs matches against pairs the model does not expect you to beat.

Is this the real algorithm?

No, and nobody outside those companies has it — none of them publish the formula. This page is a model of the shape those systems describe in their own help pages: a level seeded by a questionnaire, an expectation built from the two team averages, and a reliability figure that damps the change as you play more. Here it is in full, so you can check it by hand: E = 1 / (1 + 10^((opponents − team) / 0.86)), and the change is 0.075 × reliability (1.6 new, 1.0 settling, 0.45 established) × margin (0.75 close, 1.2 clear) × (result − E), where result is 1 for a win and 0 for a loss; the page then shows 70% to 130% of that figure. Worked example: a settling 3.0 pair beats a 2.7 and a 2.6 clearly, so E = 0.72 and the estimated change is +0.02 to +0.03. The constants are ours and the output is a range for exactly that reason — if your app moved you by a different amount, your app is right about itself.

How is a crew-local rating different?

It has nothing to protect and nothing to hide. A public matchmaking level has to place you against strangers, which is why it needs a questionnaire, a reliability figure and a private formula. A rating scoped to one crew is built only from matches those people actually played against each other, so every move can be shown in full: who you beat, what was expected, how far the number went and why. That is the version people stop arguing with.

More tools

Read next

A rating your crew fully understands

Rivals shows the whole calculation for every match: the expectation, the margin, and the exact points each of the four players moved. No hidden reliability figure, no algorithm to argue with.

Free for a crew of four. Nothing inside is metered.