Methodology
Racquet Rating is this site’s own rating for tennis players, published on a 0–1000 scale and computed from public match results. This page states what it measures, what it is built from, how the number is produced, and what it does not claim.
Everything below describes the model that produced the ratings this site publishes. Where a figure comes from the published data it is read from the data, so this page and the ratings cannot drift apart.
Definition
A single number summarising how a player has performed over a fixed recent window of match results.
Racquet Rating is a rating produced by this site. It is computed from publicly available tour-level match results using the model described on this page, and it is published on a 0–1000 scale.
It is not an official rating and carries no standing with any governing body. It is not the ATP ranking, it is not derived from ranking points, and this site is not affiliated with or endorsed by the ATP or the WTA. Where a Racquet Rating and an official ranking disagree, that is because they measure different things from different inputs, not because one of them is a correction of the other.
Pipeline
Source match results become canonical observations, which become component scores, which become one published number.
Step 1
Publicly available ATP tour-level main-draw match results, under the licence and at the exact source revision recorded on the Data sources page. Nothing is scraped from a live site and nothing is entered by hand.
Step 2
Each match becomes two observations, one from each player’s side, so a result is never stored from the winner’s perspective only. Matches with an unresolvable player identity, and matches a player played against themselves, are dropped rather than guessed at. The artifact publishes the counts, including what was dropped.
Step 3
Only observations inside the trailing rating window count toward a current rating. Everything outside it stays in the dataset and still informs the era baselines, but it does not enter the current calculation.
Step 4
The five components are each computed on a 0–100 scale from the observations in the window. Form Momentum is additionally damped toward neutral in proportion to how much evidence supports it.
Step 5
Era Normalization scores the player against frozen era baselines and tour-depth proxies; Match Context scores results against expectation given the round and the event. Both are components, weighted like the others — neither is applied afterwards as a multiplier.
Step 6
The five scores are combined with the fixed weights and multiplied by 10. This is the only arithmetic between the components and the published number, which is why the contributions shown on a profile add back up to it.
Step 7
A player below the evidence threshold is not given a number. A player above it is published with a status saying whether the number describes current form or past form.
Step 8
The result is written to a single versioned artifact, which is what this site reads. The site never recomputes a rating: every number on a profile is either published in that artifact or is published values multiplied by a published weight.
What the current artifact was built from
Scale
A weighted composite on a fixed scale — not a percentile, and not a ranking position.
Ratings are published on a 0–1000 scale. The number is a weighted composite of the five component scores — it is not a percentile, not a rank, and not a share of anything. A rating of 500 does not mean the median player, and no band of the scale carries a published label.
Where the published field currently sits
Model
Each is scored 0–100 and enters the rating at a fixed weight. These five are the whole model.
The composite
Racquet Rating = (Σ component score × component weight) × 10
Each component is scored 0–100. The five scores are combined with the fixed weights above, and the weighted score is multiplied by 10 to land on the published 0–1000 scale.
How well the player wins points and matches — serve and return points won, break points converted and saved, results against elite opponents, and the strength of the field they have faced.
It carries the most weight of any component, so it usually accounts for the largest share of a rating.
Recent trajectory: results over a 26-week window with exponential decay, a fatigue adjustment from minutes and sets played, and an allowance for a player returning after a long absence.
It is what makes the rating a current-form measure rather than a career résumé.
How evenly the player performs across hard, clay and grass — an entropy-like measure of the spread, with a small penalty for extreme specialisation and credit for consistency across surfaces.
It rewards a game that travels. It is a measure of balance, not of how good the player is on any one surface.
An adjustment against fixed era baselines and tour-depth proxies, so players from different periods remain comparable.
It is what allows one 0–1000 scale to span the whole dataset rather than only the current season.
How results compare with expectation once the round and the event are taken into account, so performance on bigger stages counts for more.
It carries the smallest weight of the five, so it moves a rating the least.
Provenance
One licensed public source of tour-level match results, pinned to an exact revision.
Ratings are computed from ATP tour-level main-draw match results taken from a single public dataset, at one recorded revision. The dataset, its author, its licence and the exact revision in use are listed on the Data sources page, which is also where the line between the source’s data and Racquet Rating’s own derived output is drawn.
Nothing else feeds the model. There are no manually entered results, no second dataset reconciled against the first, and no inputs beyond the match record — which is also why the model cannot see injuries, withdrawals or conditions except insofar as they changed a result.
Eligibility
A trailing window of match results, and a minimum amount of evidence inside it.
Racquet Rating requires at least 40 qualifying matches in the current 4-year rating window. Players without sufficient recent evidence are not assigned a current rating.
The window. A current rating is computed only from matches inside the trailing 4-year window ending at the artifact’s as-of date. Older matches stay in the dataset and still inform the frozen era baselines that Era Normalization scores against, but they do not enter a current rating. This is what makes the rating a measure of a recent period rather than of a career.
The 40-match threshold. A player with fewer than 40 qualifying matches in that window is not given a number. This is a data-sufficiency rule: below it there is not enough evidence in the window for the components to be computed from, so no rating is published. It is a decision about how much evidence the model requires before it will publish, not a statistical confidence guarantee, and no margin of error is claimed on either side of it.
A player under the threshold is not rated zero and is not ranked last. They are shown with no rating and the reason, and their match record, last match date and other published data remain visible. This says nothing about how good the player is.
159
143
16
632
States
Four different reasons a page may not show you a number. They are not interchangeable.
Active
A rating was computed, and the player has competed within 365 days of the artifact's as-of date. The rating describes current form.
Stale
A rating was computed from a sufficient evidence window, but the player's last match is more than 365 days before the as-of date. The number is real and traceable but describes past form, not present form. last_match_date states how old it is.
Not currently rated
The player appears in the rating window but has fewer than the minimum canonical observations required to publish a number. This is a statement about evidence volume, not about the player's ability, and it is NOT the same as being unranked or rated zero. Historical and peak data may still exist.
Not yet computed
No rating could be produced because the calculation did not complete -- the engine raised, or the pipeline has not yet run for this player. This is a pipeline state, not a statement about evidence.
Data not yet qualified
WTA match data has not completed source qualification, so no WTA ratings are published. Racquet Rating does not publish ratings from an unverified source. This is not a per-player status and it is not the same as a player having too little evidence: no rating has been computed for any player on an unqualified tour, so there is nothing to publish or to withhold. ATP is the only tour currently qualified.
Data unavailable
If the published artifact cannot be read, surfaces that would normally show a rating say so explicitly rather than showing a blank, a zero or an empty table. A rating that exists is still there; the site simply could not load it for that request. This is the only one of these four states that carries no information about a player or a tour.
Scope
ATP only. The WTA has not completed source qualification.
ATP is the only tour currently qualified for Racquet Rating publication. WTA match data has not completed source qualification, so no WTA ratings are published. Racquet Rating does not publish ratings from an unverified source.
There are no unpublished or hidden WTA ratings. No WTA rating has been computed, because the WTA source data has not been qualified, and this site does not publish ratings from a source it has not verified. Where the site offers a WTA view it says the tour is not qualified rather than showing an empty result, because an empty result would read as a claim that no WTA player qualifies.
Boundaries
The rating answers one question from one kind of evidence. These are the questions it does not answer.
For readers who want to check it
How to verify a published rating against the published components. Nothing here is needed to read a rating — the definitions above are complete on their own.
Every match produces two canonical observations, one from each player’s side, rather than a single winner-perspective row. A player’s record in the window is therefore read the same way whether they won or lost, and a score is never displayed from only the winner’s point of view.
The rating is a linear combination, so each component’s contribution in rating points is exactly its score × its weight × 10, and the five contributions sum to the published rating. The “Why this rating?” section on every rated profile prints that sum next to the published number so it can be checked on the page.
The check is performed at full precision on the published values and holds to within 0.5 rating points — the same tolerance the publication gate uses, so the publish-time and read-time checks agree on what counts as explainable. The artifact rounds component scores and ratings before writing them, which is the only reason the difference is not exactly zero.
Weights: Performance Index 35%, Form Momentum 25%, Surface Balance 15%, Era Normalization 15%, Match Context 10%. Ratings are published on the 0–1000 scale to 1 decimal place. Component scores are published on a 0–100 scale and are a different quantity from a contribution in rating points; both are labelled wherever both appear.
A value outside the scale is treated as a contract violation and is not rendered, rather than being rescaled to fit.
Elo is one of the seven inputs to Performance Index. It is on its own scale, it is not a sixth component of the Racquet Rating, and it is not comparable with the component scores.
Where a profile prints an Elo figure it is labelled as an input and is shown on its own scale. It is published because it is checkable, not because it is a rating.
The site reads one versioned artifact and never recomputes a rating in the browser or on the server. The artifact currently served is model v1.1.0, schema 2.0.0, built from match data through 2026-05-25.
A rating is marked stale rather than withdrawn when the player stops competing, so an old number is still traceable to the matches behind it and is never presented as current form.
Keep exploring
Jump into player pages and comparisons to see how this framework appears in real score profiles.