Tennis Point-by-Point Dataset — ATP, WTA, Challenger, ITF (Public Sample)

Professional tennis data from Live Tennis API: a match index of 173,571 completed matches (2023-01-01 to 2026-08-13, 7-day embargo) across ATP, WTA, Challenger and ITF, with point-by-point score sequences for 98% of them. Point data comes in two stated provenance classes: a live-observed layer captured by our own collection with real UTC timestamps (22,031 matches / 3.18M point states; live observation began 2025-10-12 and reached full volume in February 2026), and a post-match reconstructed layer without capture timestamps covering earlier matches (not in this deposit). This record is a public sample of the full dataset, which is available free of charge for non-commercial academic research and teaching on application ([e-mail]). Files: matches.csv.gz — all 173,571 completed matches with tournament, tier, surface, round, best-of, scheduled time (UTC) and players. players.csv.gz — 32,678 players with stable IDs, country, birth date, handedness, and a crosswalk to Jeff Sackmann's open datasets. points_sample_2026-06.csv.gz — all 951,064 live-observed point states for the 5,380 observed-covered completed matches of June 2026 (77% of that month's completed matches), in capture order with UTC timestamps. README.md and DATA_DICTIONARY.md document every column, the provenance model (this sample contains live-observed rows only) and the known limitations. Limitations, stated plainly: the live-observed layer starts 2025-10-12 (roughly 10-13% of matches per tier so far); earlier point sequences exist only as reconstructions without timestamps; observed tapes may start mid-match or end before the final point; serve speed, serve direction and shot-level data are not included. Version 1.0.1 corrects the v1.0.0 description, which wrongly said the data was "recorded live since January 2023" — the match index starts 2023, live point observation started 2025-10-12. Data files are unchanged; README.md is updated accordingly. Access to the full dataset: email [e-mail] from an institutional address with one paragraph on your research and the data you need. Conditions: non-commercial research and teaching only, no redistribution of raw data, citation required. Version 1.0.1 corrects the v1.0.0 description, which wrongly said the data was "recorded live since January 2023" — the match index starts 2023, live point observation started 2025-10-12. Data files are unchanged; README.md is updated accordingly. Version 1.0.2 (2026-09-28) Version 1.0.2 corrects one column in two files. Every other data byte is unchanged from versions 1.0.0 and 1.0.1, whose data files were identical. The timestamps in versions 1.0.0 and 1.0.1 were labelled UTC but were written in Israel local time. The export script read a naive local timestamp from the database and wrote it out unchanged. The database and the public API were correct throughout; only the exported files carried the error. What moved: - points_sample_2026-06.csv.gz, column timestamp_utc: every one of the 951,064 rows moves back 3 hours (June is inside Israel's UTC+3 window). - matches.csv.gz, column scheduled_time_utc: 120,758 rows move back 3 hours and 52,813 move back 2 hours. The file spans January 2023 to August 2026 and crosses seven Israel offset changes; each row is corrected by the offset in force at its own instant. - players.csv.gz and DATA_DICTIONARY.md are byte for byte the version 1.0.0 files. README.md in this version is the 1.0.0 text with the words "corrected in 1.0.1" added on the two timestamp columns; it does not carry the 1.0.1 README rewording about when live point observation began. That correction stands and is restated in the paragraph above: the match index starts 2023, live point observation started 2025-10-12. Same scope, same rows, same counts as 1.0.0 and 1.0.1: 173,571 matches, 32,678 players, 951,064 observed point states. Scores, the point sequence, the server and every other column are identical. Any result that used only the sequence, the score or the provenance is unaffected. One row you can check. Match 9, scheduled_time_utc, read 2026-03-22 19:40:00 in versions 1.0.0 and 1.0.1. It reads 2026-03-22 17:40:00 here, which is what the API has served all along.