Data Sources

Racquet Rating is built on publicly available tennis match data. This page records where that data comes from, the licence it carries, the exact revision in use, and which numbers on this site are the source’s versus Racquet Rating’s own.

Source dataset

Dataset: ATP match results, player master file and rankings
Author: Jeff Sackmann / Tennis Abstract
Repository: https://github.com/JeffSackmann/tennis_atp
Revision used: 712be0c5ade693cdab9e69c23a71a0edf5a23c44(thru 8 June 2026)
Licence: CC BY-NC-SA 4.0 — attribution required, non-commercial use only, ShareAlike
Coverage: ATP tour-level main draw, 1968–2026. WTA data is not currently used.

Required attribution

Match data by Jeff Sackmann / Tennis Abstract, licensed CC BY-NC-SA 4.0. Racquet Rating transforms this data into a canonical match schema and derives its own ratings from it. Racquet Rating is operated on a non-commercial basis.

Tennis databases, files, and algorithms by Jeff Sackmann / Tennis Abstract is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Based on a work at https://github.com/JeffSackmann.

Full data sources, licence and derivation details

What Racquet Rating takes from the source

Taken directly from the source data, unchanged in substance:

Match date, tournament, round and surface
Winner and loser, by the source’s own player identifiers
Match score and duration
Player rankings at the time of the match
Serve and return statistics, where the source records them (1991 onward)
Player biographical fields: name, country, date of birth, handedness

What Racquet Rating derives

The following are Racquet Rating’s own output, not source data, and should not be attributed to Jeff Sackmann or Tennis Abstract:

The Racquet Rating itself and its five components
Elo ratings and peak Elo
Rating history and rating rank
Derived per-match rates (serve points won, return points won, break-point rates)
Canonical player identifiers and URL slugs
Feature-completeness classifications

Transformations applied

Each source match becomes two player-centric observations — one win, one loss — so that losses are never discarded.
Tournament level and round codes are translated into the rating engine’s vocabulary.
Surfaces are normalised; a match with no recorded surface stays unknown and is never assigned one.
Scores are re-oriented per player, so a player who lost never appears to have won.
Placeholder identities and self-matches present in the source are filtered out rather than resolved.
Missing statistics are left missing. Nothing is imputed.

Every published rating traces back to specific source rows through a match key recorded alongside each observation.