Free for non-commercial academic research and teaching. Public sample: DOI 10.5281/zenodo.22048730 · Dataset page: livetennisapi.com/data/academic · Full-dataset applications: research@livetennisapi.com
This repository holds the documentation and loaders for the dataset. The data itself lives on Zenodo (public sample) and is granted in full on application — nothing here to clone but knowledge.
- Match index — 173,571 completed professional matches, 2023-01-01 → 2026-08-13 (7-day rolling embargo), across ATP, WTA, Challenger and ITF, singles and doubles.
- Point-by-point score sequences for 170,527 of them (98.2%), in two stated
provenance classes:
observed— captured live by Live Tennis API's own collection with real UTC timestamps: 22,031 matches / 3,184,514 point states, from 2025-10-12 (full volume from February 2026), every tier down to ITF — where public point-level data has effectively never existed.reconstructed: rebuilt after the match, with no capture timestamps. It covers 2023 to 2025 and part of 2026. It is not in the public files and is shared per request under the programme.
- 32,678 players with stable IDs, tour, country, birth date and handedness.
Known limitations are stated plainly in the data dictionary and on the dataset page: the observed layer starts late 2025; tapes may start mid-match or end before the final point (each match declares its own coverage); no serve speed/direction or shot-level data; nothing before 2023.
Python:
import pandas as pd
matches = pd.read_csv("matches.csv.gz")
players = pd.read_csv("players.csv.gz")
points = pd.read_csv("points_sample_2026-06.csv.gz",
parse_dates=["timestamp_utc"])
# games_p1/games_p2 are JSON arrays (games won per set; last element = current set)
import json
points["games_p1"] = points["games_p1"].map(json.loads)
points["games_p2"] = points["games_p2"].map(json.loads)
# One row = one score STATE; a point's outcome is the transition between
# consecutive rows of the same match_id (rows are in capture order).R:
library(readr)
matches <- read_csv("matches.csv.gz")
points <- read_csv("points_sample_2026-06.csv.gz")notebooks/findings_retest.ipynb re-runs the four tests of
Tennis findings, re-tested on the public
sample. It fetches the sample through the concept DOI and imports the page's analysis code. Then it
sets each result beside the full record's. It needs Python 3.10 or newer and nothing to install.
Email research@livetennisapi.com from an institutional address (students: ORCID or a named supervisor) with one paragraph on your research and the data you need. Target turnaround: 3 business days. Conditions: non-commercial research and teaching only · no redistribution of raw data (published aggregates and findings are fine) · citation required · tell us when you publish.
Live Tennis API (2026). Tennis Point-by-Point Dataset — ATP, WTA, Challenger, ITF. Zenodo. https://doi.org/10.5281/zenodo.22048730
This DOI resolves to the latest version of the sample. See also the corrections to this sample.
@dataset{livetennisapi_2026_tennis_pbp,
author = {{Live Tennis API}},
title = {Tennis Point-by-Point Dataset --- ATP, WTA, Challenger, ITF},
year = 2026,
publisher = {Zenodo},
doi = {10.5281/zenodo.22048730},
url = {https://doi.org/10.5281/zenodo.22048730}
}Documentation and code snippets in this repository: MIT. The dataset itself: CC BY-NC 4.0 via Zenodo and the application process — this repository grants no data rights.