See the model think.
Every number on KickServe comes from one engine: a dual Elo + Glicko-2 rating system. Instead of describing it, play with it. Drag the gap between two players and watch both models weigh in — then thin the evidence and see them disagree.
A point, and a point with a band
Both models rate every player from match results — but they answer slightly different questions. Elo gives a single number: fast to move, easy to read. Glicko-2 gives a number and a reliability band (its rating deviation, RD) — how sure we are. A hot streak on thin data widens the band, so the model hedges instead of overreacting.
Reacts quickly to every result. It carries no sense of its own certainty — a 3-match rating and a 300-match rating look equally confident.
Same rating, plus a confidence band that shrinks as evidence piles up. A wide band pulls win-probabilities toward a coin-flip.
Clay ≠ grass ≠ hard
We don't rate players once. We rate them on hard, clay and grass separately — plus an overall and a rolling last-52-week form window. The same player can be a top-2 clay threat and mid-pack on grass, and the surface ratings say so.
Illustrative: one clay specialist, three ratings. Each surface leader is normalized to 2350.
We show the market's fair price — vig removed
Betting odds bake in the bookmaker's margin (the vig), so the two prices add up to more than 100%. We strip that out to recover the market's own estimate of the true probability, then show it next to Elo and Glicko so you can see where model and market agree — or don't.
Where the numbers come from
Ratings are recomputed from full match results and refreshed daily. Live scores and win-probabilities update continuously through play, and every player page carries rating and ranking history going back years — all from the same engine. This project is independent and inspired by the open-analytics spirit of Tennis Abstract.
Is this betting advice?
Responsible gambling
Found something wrong?
◆ Who makes this
Joey Tennis Balls · a pen name
KickServe is built and run by one person, writing under a byline rather than a legal name. That is a preference about privacy and nothing more: there is no team behind the name, no agency, and no claim to credentials the work does not show on its own.
It started as a way to answer questions the official rankings could not — whether a player is better on clay than the table suggests, how much a result should really move a rating, and how confident a number deserves to be. The dual Elo and Glicko-2 model came out of testing those answers against real results rather than picking a system and trusting it.
Everything here is computed rather than compiled from elsewhere, and the method is written down on The Algo so it can be argued with. Where the data cannot support a claim, the page says so instead of quietly rounding it off — a title count that begins mid-career, or a peak ranking the record does not reach back far enough to see. Where a fact comes from someone else's reporting, it is attributed to them.
If you think something is wrong, that is genuinely worth hearing: editor@kick-serve.com.
With thanks to Ultimate Tennis Statistics
For the better part of a decade, Ultimate Tennis Statistics was the best tennis analytics site on the internet. Mileta Cekovic built it, opened the source, and set the standard for what a tennis database could be — surface Elo, a GOAT calculator you could argue with, and a records book that went far deeper than anyone else bothered to. It is offline now, which is a genuine loss to the sport.
Our Records pages and the Pressure panel on every player card follow its categories: deciding sets, tie-breaks, records after winning or losing the opening set, and how a player does against the top ten. Those are its questions, and they were the right ones. We have taken the definitions and none of the code — a win-loss record in deciding sets is a fact about tennis, while its own formulas are published under a NonCommercial licence and are not used here. The scoring on our GOAT calculator is our own, and simpler than theirs; the measure set follows goatranker.
If you ran UTS and would rather we did any of this differently, say the word.