diff --git a/CHANGELOG.md b/CHANGELOG.md
index 3ad0d10..444be7f 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,15 @@ All notable changes to the specification are recorded here.
The API surface is versioned as `v1`. Changes within `v1` are **additive only**;
removing a field or changing its type would require `v2`.
+## [1.13.89] - 2026-11-20
+
+### Added
+- **The changelog as an Atom feed and a JSON Feed.** `changelog.atom` (Atom 1.0) and `changelog.json` (JSON Feed 1.1) carry every entry of this changelog. Each one has the version, the date, the items of each section, a summary and the client library versions the entry names. Both are written from the same parse as `changelog.html`. The build fails when either feed breaks its format or differs from the page. An entry's id is its anchor on the page, so a rebuild never shows an old entry as new. The changelog page and the docs home link both feeds in their head, and llms.txt lists them. Nothing on the wire changed.
+
+### Fixed
+- **The changelog page shows all 115 entries.** Fifteen headings from 1.1.0 to 1.9.3 were written with an em dash. The page folded them into the 1.9.4 entry as plain text. Each is now its own entry, and the build stops on a heading it cannot read.
+- **Bold and code that wrap onto a second line render in full.** Six older entries showed raw asterisks and backticks there. Two links written in angle brackets showed the brackets. Both now render as on the rest of the page.
+
## [1.13.88] - 2026-11-19
### Changed
@@ -742,7 +751,7 @@ Documentation only. No field, endpoint or wire value changed.
`synthetic: true` by design because the live book attaches at match start;
`timestamp` is our observation clock (UTC), not a venue publication time.
-## [1.9.3] — 2026-09-04
+## [1.9.3] - 2026-09-04
### Clarified
- **`Match.tournament_id` null cases** restated with the real causes and a
@@ -752,7 +761,7 @@ Documentation only. No field, endpoint or wire value changed.
2026 backlog was backfilled the same day (2,547 rows); ITF null rate went
from about 25% to about 2.5%. No wire change.
-## [1.9.2] — 2026-09-04
+## [1.9.2] - 2026-09-04
### Added (spec catch-up — every one of these has been live on the API; the document lagged)
- **`GET /matches?status=cancelled`** is now in the `status` enum, with what it
@@ -774,7 +783,7 @@ Documentation only. No field, endpoint or wire value changed.
listings, so `has_more` is the only end-of-data signal there.
Asked by a Basic customer on Discord; no wire change.
-## [1.9.1] — 2026-09-02
+## [1.9.1] - 2026-09-02
### Clarified
- **`Player.ranking` / `Player.ranking_points`** now say what they are: the
@@ -784,7 +793,7 @@ Documentation only. No field, endpoint or wire value changed.
record even on historical matches (`/rankings?as_of=` is the as-of surface).
Asked by a Basic customer; no wire change.
-## [1.9.0] — 2026-09-02
+## [1.9.0] - 2026-09-02
### Added
- **`has_analysis` and `has_market` on `Match`** — two booleans on every row of
@@ -800,7 +809,7 @@ Documentation only. No field, endpoint or wire value changed.
`/matches/{matchId}/statistics` already uses in its `200`. New `error` codes
`no_analysis`, `no_market` documented on the `Error` schema.
-## [1.8.0] — 2026-09-01
+## [1.8.0] - 2026-09-01
### Added
- **`GET /history/archive/matches/{archiveId}/tape`** (`getArchiveTape`) — the
@@ -862,7 +871,7 @@ Documentation only. No field, endpoint or wire value changed.
corpus, the null clock and the coverage floor together.
- `info.version` is now `1.8.0`.
-## [1.7.2] — 2026-08-23
+## [1.7.2] - 2026-08-23
### Changed
- **`status` on Match now documents the lifecycle rule.** `completed` is
@@ -877,7 +886,7 @@ Documentation only. No field, endpoint or wire value changed.
changed type — prose only.
- `info.version` is now `1.7.2`.
-## [1.7.1] — 2026-08-19
+## [1.7.1] - 2026-08-19
### Added
- **`event_status_updated_at` on Match.** Nullable ISO-8601 UTC (`Z`)
@@ -894,7 +903,7 @@ Documentation only. No field, endpoint or wire value changed.
### Changed
- `info.version` is now `1.7.1`.
-## [1.7.0] — 2026-08-18
+## [1.7.0] - 2026-08-18
### Added
- **`basis` on `GET /matches/{matchId}/points` responses.** `live` |
@@ -921,7 +930,7 @@ Documentation only. No field, endpoint or wire value changed.
### Changed
- `info.version` is now `1.7.0`.
-## [1.6.0] — 2026-08-18
+## [1.6.0] - 2026-08-18
### Added
- **The three-valued `draw` field on Match.** `singles` | `doubles` | null —
@@ -987,7 +996,7 @@ Documentation only. No field, endpoint or wire value changed.
### Changed
- `info.version` is now `1.6.0`.
-## [1.5.0] — 2026-08-17
+## [1.5.0] - 2026-08-17
### Added
- **The as-of Elo tape.** `GET /rankings?system=elo` (ULTRA, both modes) —
@@ -1053,7 +1062,7 @@ Documentation only. No field, endpoint or wire value changed.
that change.
- `info.version` is now `1.5.0`.
-## [1.4.0] — 2026-08-16
+## [1.4.0] - 2026-08-16
### Added
- **Live per-point events.** `GET /matches/{matchId}/points` (ULTRA) — the
@@ -1107,7 +1116,7 @@ Documentation only. No field, endpoint or wire value changed.
well as `?sequence=clean` (there the served order IS point order).
- `info.version` is now `1.4.0`.
-## [1.3.1] — 2026-08-07
+## [1.3.1] - 2026-08-07
### Added
- **`kind=rally` bulk packages documented.** The `/history/packages` `kind`
@@ -1119,7 +1128,7 @@ Documentation only. No field, endpoint or wire value changed.
`kind` accepts `tape` (default), `rankings` (ULTRA), `rally` (ULTRA)
and `archive`.
-## [1.3.0] — 2026-08-07
+## [1.3.0] - 2026-08-07
### Added
- **Shot-level charting.** `GET /charting/players` (ULTRA) — career
@@ -1168,7 +1177,7 @@ Documentation only. No field, endpoint or wire value changed.
key limit is stated.
- `info.version` is now `1.3.0`.
-## [1.2.0] — 2026-08-03
+## [1.2.0] - 2026-08-03
### Added
- **Results archive (1968–2022).** The history product now runs in two
@@ -1246,7 +1255,7 @@ Documentation only. No field, endpoint or wire value changed.
and `/h2h` on BASIC/History plans, the rankings listing on PRO, rally and
per-player as-of rankings on ULTRA).
-## [1.1.0] — 2026-08-02
+## [1.1.0] - 2026-08-02
### Added
- **Usage endpoint.** `GET /usage` (FREE — any tier): your own durable daily
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 19761ac..b8da239 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -44,6 +44,15 @@ entry. It also fails when an entry names an operation the spec does not
define, or when the samples are stale. `python -m unittest scripts/test_build_code_samples.py`
breaks copies of both files and checks that each of those failures is caught.
+## Changelog feeds
+
+`docs/changelog.atom` (Atom 1.0) and `docs/changelog.json` (JSON Feed 1.1)
+are built from `CHANGELOG.md` together with `docs/changelog.html`, so do not
+edit them by hand. Write each entry heading as `## [x.y.z] - YYYY-MM-DD`.
+The build stops on any other `## ` line. It also fails when a feed breaks its
+format or differs from the page. `python -m unittest scripts/test_changelog_feeds.py`
+runs the same checks and breaks copies of both feeds to show each check works.
+
## Versioning
Changes within `v1` are **additive only**: new endpoints, new optional fields.
diff --git a/docs/auth-quota-and-health.html b/docs/auth-quota-and-health.html
index 5729f61..952eb71 100644
--- a/docs/auth-quota-and-health.html
+++ b/docs/auth-quota-and-health.html
@@ -202,7 +202,7 @@
Live Tennis API DocsGet a free key ↗Explore the documentation
-
Live Tennis API · docs · version 1.13.88
+
Live Tennis API · docs · version 1.13.89
Tennis API authentication, quota and status
How do you authenticate, and how do you see what quota is left? Send your key as X-API-Key on every request. GET /usage returns your own consumption against your plan's quota and works on every tier including FREE, so a client can check its own headroom before a burst. GET /health is the unauthenticated liveness probe.
2 endpoints on this page. Base URL
diff --git a/docs/broadcast-graphics.html b/docs/broadcast-graphics.html
index d28df30..07257bd 100644
--- a/docs/broadcast-graphics.html
+++ b/docs/broadcast-graphics.html
@@ -202,7 +202,7 @@
Live Tennis API DocsGet a free key ↗Explore the documentation
-
Live Tennis API · docs · version 1.13.88
+
Live Tennis API · docs · version 1.13.89
Tennis scores for on-air graphics
How do you put live tennis scores into a broadcast graphics template?GET /broadcast/match/{matchId} returns ONE flat object for one match and GET /broadcast/live the same object for every live match in a single call. Both are ULTRA. A graphics template binds each field to a fixed path, so three things are contractual: the object is flat, every key is present on every read (null, never missing), and the score is already in display form — p1_points reads "40" or "AD" and set_line is one string. Built for 1 Hz: every response carries a strong ETag, so If-None-Match makes an unchanged state a 304 with no body. Detect a new state on sequence, never on served_at, which changes on every read.
2 endpoints on this page. Base URL
diff --git a/docs/changelog.atom b/docs/changelog.atom
new file mode 100644
index 0000000..9a5301d
--- /dev/null
+++ b/docs/changelog.atom
@@ -0,0 +1,1885 @@
+
+
+https://docs.livetennisapi.com/changelog.html
+Live Tennis API changelog
+Every change to the Live Tennis API specification, dated and versioned, newest first.
+2026-11-20T00:00:00Z
+
+
+
+Live Tennis APIhttps://livetennisapi.com
+https://docs.livetennisapi.com/icon-256.png
+MIT licence, https://github.com/livetennisapi/openapi/blob/main/LICENSE
+
+https://docs.livetennisapi.com/changelog.html#v1.13.89
+1.13.89: The changelog as an Atom feed and a JSON Feed, and 2 more changes
+2026-11-20T00:00:00Z
+2026-11-20T00:00:00Z
+
+
+
+Added: The changelog as an Atom feed and a JSON Feed. Fixed: The changelog page shows all 115 entries. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><strong>The changelog as an Atom feed and a JSON Feed.</strong> <code>changelog.atom</code> (Atom 1.0) and <code>changelog.json</code> (JSON Feed 1.1) carry every entry of this changelog. Each one has the version, the date, the items of each section, a summary and the client library versions the entry names. Both are written from the same parse as <code>changelog.html</code>. The build fails when either feed breaks its format or differs from the page. An entry's id is its anchor on the page, so a rebuild never shows an old entry as new. The changelog page and the docs home link both feeds in their head, and llms.txt lists them. Nothing on the wire changed.</li>
+</ul>
+<h3>Fixed</h3>
+<ul>
+<li><strong>The changelog page shows all 115 entries.</strong> Fifteen headings from 1.1.0 to 1.9.3 were written with an em dash. The page folded them into the 1.9.4 entry as plain text. Each is now its own entry, and the build stops on a heading it cannot read.</li>
+<li><strong>Bold and code that wrap onto a second line render in full.</strong> Six older entries showed raw asterisks and backticks there. Two links written in angle brackets showed the brackets. Both now render as on the rest of the page.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.88
+1.13.88: The client libraries page and the code samples name JavaScript 1.10.3
+2026-11-19T00:00:00Z
+2026-11-19T00:00:00Z
+
+
+
+
+Changed: The client libraries page and the code samples name JavaScript 1.10.3.
+<h3>Changed</h3>
+<ul>
+<li><strong>The client libraries page and the code samples name JavaScript 1.10.3.</strong> It fixes requests in a browser, which all failed with "Illegal invocation" in 1.10.2. Its timeout now covers the response body as well as the headers. No JavaScript method changed, so every JavaScript sample keeps its call. The samples now link the source at the <code>v1.10.3</code> tag. The page also links the live scores web starter, a static page built on this client. Nothing on the wire changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.87
+1.13.87: The client libraries page and the code samples name R 0.2.1
+2026-11-17T00:00:00Z
+2026-11-17T00:00:00Z
+
+
+
+Changed: The client libraries page and the code samples name R 0.2.1.
+<h3>Changed</h3>
+<ul>
+<li><strong>The client libraries page and the code samples name R 0.2.1.</strong> It now installs jsonlite, so a clean install can read responses. No R function changed, so every R sample keeps its call. The R samples now link the source at the <code>v0.2.1</code> tag. Nothing on the wire changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.86
+1.13.86: The client libraries page and the code samples name Swift 1.2.1
+2026-11-17T00:00:00Z
+2026-11-17T00:00:00Z
+
+
+
+Changed: The client libraries page and the code samples name Swift 1.2.1.
+<h3>Changed</h3>
+<ul>
+<li><strong>The client libraries page and the code samples name Swift 1.2.1.</strong> It adds <code>Match.hasAnalysis</code>, <code>Match.hasMarket</code> and <code>Match.eventStatusUpdatedAt</code>. It also builds on Linux with Swift 5.10 again. No Swift method changed, so every Swift sample keeps its call. The Swift samples now link the source at the <code>v1.2.1</code> tag. Nothing on the wire changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.85
+1.13.85: Code samples for the official clients
+2026-11-17T00:00:00Z
+2026-11-17T00:00:00Z
+
+
+
+
+
+
+
+
+
+
+Added: Code samples for the official clients.
+<h3>Added</h3>
+<ul>
+<li><strong>Code samples for the official clients.</strong> Every operation now carries an <code>x-codeSamples</code> list. It starts with a curl call, then has one sample for each official client with a method for that operation. The samples cover Python 1.8.0, JavaScript 1.10.2, Go 1.3.0 and PHP 1.2.0. They also cover .NET 1.2.1, Dart 1.2.1, Swift 1.1.1 and R 0.2.0. The interactive reference lists them in the language picker of each operation. Of its own generated snippets it keeps only curl, which follows the values typed into Test Request. <code>code-samples/sdk-methods.yaml</code> records the method and the call of each client for each operation, read from the source of that version. <code>scripts/build_code_samples.py</code> writes the samples into the spec. The build fails when an operation has no entry in that file. It also fails when an entry names an operation the spec does not define, or when the samples are stale. Eleven operations have no client method yet, so they show curl only. Nothing on the wire changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.84
+1.13.84: A SportsML mapping page and test fixtures
+2026-11-15T00:00:00Z
+2026-11-15T00:00:00Z
+
+
+Added: A SportsML mapping page and test fixtures.
+<h3>Added</h3>
+<ul>
+<li><strong>A SportsML mapping page and test fixtures.</strong> <code>sportsml.html</code> maps every field of the <code>Score</code> object to the IPTC SportsML 3.1 element or attribute it becomes. Where nothing fits, it says so. Each SportsML name links to IPTC's schema documentation. A worked example shows a match decided by a match tie-break as score states and as a SportsML document. <code>examples/sportsml/</code> holds six synthetic matches under CC BY 4.0: a deciding-set tie-break, a match tie-break, no-ad games, a retirement, a walkover and a default. <code>check_fixtures.py</code> validates each document against the vendored IPTC schema and replays every point against the JSON states. The build fails when the <code>Score</code> schema gains a field the page does not map. The docs menu, the docs home, the sitemap, llms.txt and the README link the page. Nothing on the wire changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.83
+1.13.83: A client libraries page
+2026-11-12T00:00:00Z
+2026-11-12T00:00:00Z
+
+
+Added: A client libraries page.
+<h3>Added</h3>
+<ul>
+<li><strong>A client libraries page.</strong> <code>client-libraries.html</code> lists the eight official language clients and seven packages built on them. Each row has the install command, the latest published version, the minimum runtime, the licence and the source repository. A second table says whether each package covers REST, the live stream, file export and MCP. A third shows which endpoint groups each language client reaches, read from the source of each published version. The page ends with the smallest working call in each language, copied from that library's README. <code>scripts/check_client_versions.py</code> reads the registries again and names any row that is behind. The build fails when the page names an operation the spec does not define. The docs menu, the docs home, the sitemap, llms.txt and the README link the page. Nothing on the wire changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.82
+1.13.82: The Fixture schema lists every field a fixture row carries
+2026-11-12T00:00:00Z
+2026-11-12T00:00:00Z
+
+
+Fixed: The Fixture schema lists every field a fixture row carries.
+<h3>Fixed</h3>
+<ul>
+<li><strong>The <code>Fixture</code> schema lists every field a fixture row carries.</strong> <code>reason</code>, <code>match_id</code>, <code>gender</code>, <code>is_qualifying</code>, <code>created_at</code>, <code>updated_at</code> and <code>pregame_forecasts</code> have been on <code>/fixtures</code> rows since the dates in their descriptions. The schema named 13 of the 20 fields, so a generated client dropped the other seven. <code>Match</code> gains <code>pregame_forecasts</code> for the same reason. Schema only, the responses are unchanged.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.81
+1.13.81: An export formats page
+2026-11-11T00:00:00Z
+2026-11-11T00:00:00Z
+
+
+
+Added: An export formats page.
+<h3>Added</h3>
+<ul>
+<li><strong>An export formats page.</strong> <code>export-formats.html</code> explains how to export matches, fixtures, players, rankings and tournaments to CSV, JSON Lines or Parquet with <code>livetennisapi-export</code> 0.1.0. It says when each format fits and lists the six datasets with the endpoint and the plan each needs. It also covers paging, how nested fields become dotted columns and when the file is written. The read-back examples for pandas, DuckDB and R were run against files the package wrote. The build fails when the page names an endpoint, a column or a page size the spec does not have. The docs menu, the docs home, the sitemap and llms.txt link the page. Nothing on the wire changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.80
+1.13.80: An errors and rate limits page
+2026-11-09T00:00:00Z
+2026-11-09T00:00:00Z
+
+
+
+
+Added: An errors and rate limits page.
+<h3>Added</h3>
+<ul>
+<li><strong>An errors and rate limits page.</strong> <code>errors-and-rate-limits.html</code> lists every non-2xx status the spec documents and every <code>error</code> code. Each row says what causes it, what to do and which operations name it. It also covers the three 429 bodies and the WebSocket close codes. Statuses, codes and per-plan limits are read from the spec at build time, and the build fails when the written half disagrees with it. The retry section matches the official clients, Python 1.8.0 and JS 1.10.2. They retry a per-minute 429, any 5xx and dropped connections, and wait <code>Retry-After</code> when it is sent. They never retry a daily 429 or <code>abuse_throttled</code>. The docs menu, the docs home, the sitemap and llms.txt link the page. Nothing on the wire changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.79
+1.13.79: The super-tiebreak example in Score.games reads right, and 1 more change
+2026-11-09T00:00:00Z
+2026-11-09T00:00:00Z
+
+
+Fixed: The super-tiebreak example in Score.games reads right. The gender parameter description lost a stray line.
+<h3>Fixed</h3>
+<ul>
+<li><strong>The super-tiebreak example in <code>Score.games</code> reads right.</strong> <code>[[6,4,10],[4,6,5]]</code> is <code>6-4 4-6 [10-5]</code>. The description said <code>6-4 3-6 [10-5]</code>, which does not match the array. Text only, the field is unchanged.</li>
+<li><strong>The <code>gender</code> parameter description lost a stray line.</strong> A run of error codes from another parameter had been pasted at its end. Text only, the parameter is unchanged.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.78
+1.13.78: Shot-level winners and errors: where they are and when they are withheld
+2026-10-29T00:00:00Z
+2026-10-29T00:00:00Z
+
+
+Changed: Shot-level winners and errors: where they are and when they are withheld.
+<h3>Changed</h3>
+<ul>
+<li><strong>Shot-level winners and errors: where they are and when they are withheld.</strong> The family is present where the statistics source counts it: the 2026 Grand Slam main draws (Wimbledon and the US Open near complete, the Australian Open and Roland Garros filling as the completed-match statistics fetch walks back) and the odd regular-tour match. The earlier note that it had not been available since 2026-07-12 described live payloads of the regular tour and is replaced. A family the source filled with zeros is now withheld: a side that won 20 points or more with every by-shot winner and error at 0 is not measured, and the whole family is omitted, totals included. 33 of 33 Roland Garros 2026 qualifying payloads carried such zeros.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.77
+1.13.77: /fixtures and /matches tell one story, and 1 more change
+2026-10-29T00:00:00Z
+2026-10-29T00:00:00Z
+
+
+Fixed: /fixtures and /matches tell one story. A bare digit in the schedule's status field is no longer a match in play.
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>/fixtures</code> and <code>/matches</code> tell one story.</strong> A fixture bound to a match (<code>match_id</code> set) is served with the match's <code>status</code>, in the fixture vocabulary, and the match's <code>start_time</code>, read at request time. The match row moves within minutes of a schedule change; the fixture row followed the 6-hourly refresher, so the two endpoints could disagree on the same match for hours. A customer found a Vienna match <code>live</code> at 09:00Z on one and <code>upcoming</code> at 12:30Z on the other. A vendor-stated <code>cancelled</code> or <code>postponed</code> on the fixture is kept. A null <code>is_qualifying</code> on the fixture takes the match's value when the match states one.</li>
+<li><strong>A bare digit in the schedule's status field is no longer a match in play.</strong> The day-ahead listing carries <code>1</code> on matches that have not started. It was read as a set in play, so 298 fixtures for 29 October were published as <code>live</code>. The digit is <code>live</code> only when the schedule's own in-play flag says so. The stored rows were restated the same day.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.76
+1.13.76: X-LTAPI-Signature-V1 on every webhook delivery, and 2 more changes
+2026-10-28T00:00:00Z
+2026-10-28T00:00:00Z
+
+
+
+
+Added: X-LTAPI-Signature-V1 on every webhook delivery. Fixed: The webhook events enum matches what registration accepts. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><strong><code>X-LTAPI-Signature-V1</code> on every webhook delivery.</strong> The value is <code>t=<unix seconds>,v1=<hex></code>, and the hex is HMAC-SHA256 of <code><t>.<raw request body></code> with your webhook secret. The timestamp is inside the signed string, so a captured delivery re-sent later under a new time no longer verifies. Each body also carries a top-level <code>delivery_id</code>, unique per delivery, to dedupe on. The <code>POST /webhooks</code> description lists the receiver checks: a 5-minute timestamp window, a constant-time compare and a dedupe on <code>delivery_id</code>. It also lists <code>score_withdrawn</code> among the <code>X-LTAPI-Event</code> values, which it had left out.</li>
+</ul>
+<h3>Fixed</h3>
+<ul>
+<li><strong>The webhook <code>events</code> enum matches what registration accepts.</strong> <code>POST /webhooks</code> listed <code>stoppages</code>, which registration refuses with a 400, and the <code>Webhook</code> schema listed only <code>score</code> and <code>break_point</code>. Both now read <code>score</code>, <code>break_point</code>, <code>point</code>, <code>status</code>.</li>
+</ul>
+<h3>Deprecated</h3>
+<ul>
+<li><strong><code>X-LTAPI-Signature</code>, the body-only webhook signature.</strong> It does not cover the timestamp. We keep sending it unchanged until 2027-02-01, and deliveries sent from 00:00 UTC on that date do not carry it. Both headers use the same secret, so moving to the new header needs no re-registration.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.75
+1.13.75: Fields are described by what they hold
+2026-10-28T00:00:00Z
+2026-10-28T00:00:00Z
+
+
+Changed: Fields are described by what they hold.
+<h3>Changed</h3>
+<ul>
+<li><strong>Fields are described by what they hold.</strong> Rounds, stoppages, per-point tags, statistics and surfaces now say what each value means and when it is null. Earlier entries in this changelog were reworded the same way. The 2013 to 2022 archive tape is described as reconstructed point sequences with no capture timestamps. Its <code>point_source_detail</code> value is unchanged. The charting endpoints describe the charted corpus by its coverage. Nothing on the wire changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.74
+1.13.74: The point-by-point page describes the tape by what it holds
+2026-10-26T00:00:00Z
+2026-10-26T00:00:00Z
+
+
+Changed: The point-by-point page describes the tape by what it holds.
+<h3>Changed</h3>
+<ul>
+<li><strong>The point-by-point page describes the tape by what it holds.</strong> The tape is the ordered score states we hold for a completed match, with coverage declared per match. The model's win probability sits only on rows that were observed live.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.73
+1.13.73: Source links point at GitHub again
+2026-10-26T00:00:00Z
+2026-10-26T00:00:00Z
+
+
+Changed: Source links point at GitHub again.
+<h3>Changed</h3>
+<ul>
+<li><strong>Source links point at GitHub again.</strong> The github.com/livetennisapi organisation is public again, so every page, the README and llms.txt link each package's repository on GitHub beside its registry page. The Organization's sameAs names the GitHub org instead of the GitLab mirror.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.72
+1.13.72: duplicates_collapsed on GET /matches/{matchId}/points
+2026-10-26T00:00:00Z
+2026-10-26T00:00:00Z
+
+
+Added: duplicates_collapsed on GET /matches/{matchId}/points.
+<h3>Added</h3>
+<ul>
+<li><strong><code>duplicates_collapsed</code> on <code>GET /matches/{matchId}/points</code>.</strong> How many rows of the page repeated a point the match had already published and were left out. The writer no longer stores such repeats; the count covers rows written before that. <code>last_seq</code> still advances past them.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.71
+1.13.71: Every docs page head carries the X (Twitter) card tag
+2026-10-26T00:00:00Z
+2026-10-26T00:00:00Z
+
+
+Fixed: Every docs page head carries the X (Twitter) card tag.
+<h3>Fixed</h3>
+<ul>
+<li><strong>Every docs page head carries the X (Twitter) card tag.</strong> The 2026-10-25 Ahrefs crawl listed twelve docs pages under "X card missing". The heads already carried the Open Graph tags, so the card tag names the card type and the rest falls back to Open Graph.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.70
+1.13.70: index.html's JSON-LD description is its meta description
+2026-10-23T00:00:00Z
+2026-10-23T00:00:00Z
+
+
+Fixed: index.html's JSON-LD description is its meta description.
+<h3>Fixed</h3>
+<ul>
+<li><strong>index.html's JSON-LD description is its meta description.</strong> The hub page's <code>TechArticle</code> carried a different sentence from its meta tag. The meta text is the one that fits the result-page budget, so the JSON-LD now repeats it. The generator's JSON-LD check covers index.html as well as the generated pages.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.69
+1.13.69: JSON-LD carries raw text, not HTML entities
+2026-10-23T00:00:00Z
+2026-10-23T00:00:00Z
+
+
+Fixed: JSON-LD carries raw text, not HTML entities.
+<h3>Fixed</h3>
+<ul>
+<li><strong>JSON-LD carries raw text, not HTML entities.</strong> The players page's <code>TechArticle</code> description read <code>player&#x27;s</code> inside its JSON-LD while the meta tag decoded to <code>player's</code>: the topic, reference and changelog templates formatted the block with the HTML escaper, and a script element's text is never decoded. Every JSON-LD block is now built as a dict and serialized once with <code>json.dumps</code>, <code></</code> escaped as <code><\/</code>. The reference and changelog descriptions in JSON-LD are the same text as their meta descriptions. The generator refuses a page whose JSON-LD does not parse, contains an entity, or whose <code>TechArticle</code> description differs from the decoded meta description.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.68
+1.13.68: The docs' Organization node matches the apex page's, and 1 more change
+2026-10-23T00:00:00Z
+2026-10-23T00:00:00Z
+
+
+Changed: The docs' Organization node matches the apex page's. Titles fit 60 characters and descriptions 110 to 155.
+<h3>Changed</h3>
+<ul>
+<li><strong>The docs' Organization node matches the apex page's.</strong> Every docs page publishes the same <code>Organization</code> as livetennisapi.com: <code>alternateName</code> and the ten <code>sameAs</code> profiles (RapidAPI, Apify, API.market, Postman, GitLab, PyPI, the two npm packages, the docs site and X). Ahrefs flagged the docs node as incomplete while it carried only the name, url and logo.</li>
+<li><strong>Titles fit 60 characters and descriptions 110 to 155.</strong> Three topic pages carry a shorter head title than their h1, the authentication page's description says what <code>GET /usage</code> is for, and a description the budget cuts no longer ends on a connective. The generator refuses a page outside these budgets, index.html included.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.67
+1.13.67: /players/{playerId}/stats keeps a UTC instant as given, and 3 more changes
+2026-10-22T00:00:00Z
+2026-10-22T00:00:00Z
+
+
+Fixed: /players/{playerId}/stats keeps a UTC instant as given. An omitted to is the request instant. Plus 2 more changes.
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>/players/{playerId}/stats</code> keeps a UTC instant as given.</strong> <code>from</code> and <code>to</code> are compared as instants, and <code>window</code> echoes what was applied. An instant inside a repeated daylight-saving hour was read through a local wall clock. <code>to=2025-10-25T22:30:00Z</code> came back as 23:30Z with <code>from</code> an hour late, <code>from=22:30Z&to=23:15Z</code> was refused and the reversed window was accepted. The same ordering applies to <code>from</code> and <code>to</code> on <code>GET /matches</code> and <code>GET /history/matches</code>.</li>
+<li><strong>An omitted <code>to</code> is the request instant.</strong> It used to be rounded down to the minute, so a <code>from</code> a few seconds earlier answered 400 <code>bad_date</code>. Polls without <code>to</code> inside one minute still share a cached response, and its <code>window.to</code> is the first poll's instant.</li>
+<li><strong><code>freshness.measured.complete</code> and <code>serve_return.matches_with_stats</code> are one judgement.</strong> A deciding match tiebreak written as its points in the stamp and as one game in the served final, or the other way round, is the same finished set. The statistics endpoint compared the totals first and answered <code>null</code> one way and <code>false</code> the other, while the player aggregate counted the match complete. The field text now says what the judgement is.</li>
+<li><strong><code>PlayerRecentMatch.score.games</code> are the games the match object publishes.</strong> The recent scores go through the same projection as <code>GET /matches/{matchId}</code>, so a deciding match tiebreak the row keeps in <code>points</code> is rendered into the deciding-set games slot. Before this, the stored lists were read raw and a match that ended in the breaker showed one set fewer.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.66
+1.13.66: Player.tour describes a doubles team the way the API decides it, and 2 more changes
+2026-10-22T00:00:00Z
+2026-10-22T00:00:00Z
+
+
+
+Fixed: Player.tour describes a doubles team the way the API decides it. Parameter tables show numeric bounds. Plus 1 more change.
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>Player.tour</code> describes a doubles team the way the API decides it.</strong> A team filed under a circuit of its own carries that circuit. Only a team filed under a level word or nothing falls back to the circuit both its members share, and it is <code>null</code> when they disagree or either is unknown. The 1.13.63 text put the fallback first. The field for singles or doubles is the match's <code>draw</code>, not its <code>gender</code>.</li>
+<li><strong>Parameter tables show numeric bounds.</strong> Every parameter with a <code>minimum</code> or <code>maximum</code> now says so next to its default. <code>/players/{playerId}/stats?last=</code> reads as 1 to 50, where it used to show only the default of 20.</li>
+<li><strong>The AI SDK guide says how <code>check_api_status</code> detects the plan.</strong> livetennisapi-ai 1.3.0 reads <code>/usage</code> and reports the key's own tier, so a FREE key with History access reports as FREE. The BASIC estimate applies only to the fallback probe, used when <code>/usage</code> cannot be read.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.65
+1.13.65: The AI SDK guide now pins livetennisapi-ai 1.3.0, which adds player_stats (PRO, with serve_return on ULTRA)
+2026-10-22T00:00:00Z
+2026-10-22T00:00:00Z
+
+
+
+
+Changed: The AI SDK guide now pins livetennisapi-ai 1.3.0, which adds player_stats (PRO, with serve_return on ULTRA).
+<h3>Changed</h3>
+<ul>
+<li>The AI SDK guide now pins livetennisapi-ai 1.3.0, which adds <code>player_stats</code> (PRO, with <code>serve_return</code> on ULTRA). The guide and llms.txt now say 25 tools, the same set as livetennisapi-mcp 1.6.1.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.64
+1.13.64: The AI SDK guide now pins livetennisapi-ai 1.2.1
+2026-10-22T00:00:00Z
+2026-10-22T00:00:00Z
+
+
+
+Changed: The AI SDK guide now pins livetennisapi-ai 1.2.1.
+<h3>Changed</h3>
+<ul>
+<li>The AI SDK guide now pins livetennisapi-ai 1.2.1. That release fixes <code>check_api_status</code> for FREE keys, so the known-issue note and its workaround are gone.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.63
+1.13.63: serve_return.matches_incomplete is documented, and 6 more changes
+2026-10-22T00:00:00Z
+2026-10-22T00:00:00Z
+
+
+
+
+Fixed: serve_return.matches_incomplete is documented. serve_return.matches_with_stats says what it counts. Plus 5 more changes.
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>serve_return.matches_incomplete</code> is documented.</strong> The API already returns it. It counts matches whose measured statistics stop short of the final. Those matches are left out of every sum.</li>
+<li><strong><code>serve_return.matches_with_stats</code> says what it counts.</strong> It counts only matches whose measured statistics reach the final, and only those are summed.</li>
+<li><strong><code>GET /players</code> documents its 400 response.</strong> A <code>restated_since</code> that is not an ISO-8601 instant answers <code>bad_restated_since</code> with the <code>Error</code> body.</li>
+<li><strong><code>Player.tour</code> lists the values the API sends.</strong> They are <code>atp</code>, <code>wta</code> and <code>null</code>. The old text described capitalised and level values that have not been served since 2026-09-24.</li>
+<li><strong>Field tables show nested fields.</strong> Inline objects are spelled out as dotted paths, such as <code>sets.tiebreaks</code> and <code>serve.first_serve_in</code>. A field typed by a named schema links to that schema's table.</li>
+<li><strong>Changelog list items keep their continuation lines.</strong> An unindented line under a bullet rendered as a paragraph inside the list. That is invalid HTML.</li>
+<li><strong>The AI SDK guide states a known issue with FREE keys.</strong> In livetennisapi-ai 1.1.0, <code>check_api_status</code> reports a FREE key as unreachable. The guide names the workaround until 1.2.1 is on npm.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.62
+1.13.62: PlayerRecentMatch.opponent is nullable, and 6 more changes
+2026-10-22T00:00:00Z
+2026-10-22T00:00:00Z
+
+
+Fixed: PlayerRecentMatch.opponent is nullable. PlayerRecentMatch.score.games values are nullable integers. Plus 5 more changes.
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>PlayerRecentMatch.opponent</code> is nullable.</strong> The endpoint returns <code>null</code> when the match row holds no opponent id. The schema said it was always an object.</li>
+<li><strong><code>PlayerRecentMatch.score.games</code> values are nullable integers.</strong> The per-set lists are passed through from the stored row as <code>GET /matches/{matchId}</code> serves them, and nothing on that path coerces the values.</li>
+<li><strong><code>records.last_12_months</code> now says what window it covers.</strong> It is the 365 days ending at <code>to</code>, whatever <code>from</code> says. With the default window it equals <code>overall</code>.</li>
+<li><strong>The <code>/players/{playerId}/stats</code> 400 response references the <code>Error</code> schema.</strong> It names the five codes and says that <code>bad_surface</code> and <code>bad_level</code> list the accepted values in <code>allowed</code>.</li>
+<li><strong>The <code>Player</code> schema gains seven fields the API already returns.</strong> They are <code>ranking_status</code>, <code>ranking_checked_at</code>, <code>itf_ranking</code>, <code>profile_restated_at</code>, <code>profile_version</code>, <code>archive_player_id</code> and <code>itf_id</code> (ULTRA only). <code>GET /players</code> gains the <code>restated_since</code> parameter that <code>profile_restated_at</code> points to. Nothing changed on the wire.</li>
+<li><strong><code>llms.txt</code> and the plans table name the new endpoint.</strong> <code>/players/{playerId}/stats</code> is listed under PRO and its <code>serve_return</code> block under ULTRA.</li>
+<li><strong>The players topic page no longer says <code>/rankings</code> per-player as-of records are PRO.</strong> The rank-ordered listing is PRO, and the as-of records are ULTRA.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.61
+1.13.61: GET /players/{playerId}/stats (PRO; serve_return ULTRA), and 2 more changes
+2026-10-22T00:00:00Z
+2026-10-22T00:00:00Z
+
+
+
+Added: GET /players/{playerId}/stats (PRO; serve_return ULTRA). height_cm and doubles_ranking on GET /players/{playerId}. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><strong><code>GET /players/{playerId}/stats</code> (PRO; <code>serve_return</code> ULTRA).</strong> One player's record over our own completed singles matches, 2023 onward. The default window is the 365 days ending now, and <code>?from=</code>, <code>?to=</code>, <code>?surface=</code>, <code>?level=</code> and <code>?last=</code> narrow it. PRO gets <code>records</code>: W/L overall, over the last 12 months, and by surface, venue, level and season. It also carries the set-based record (tiebreaks, deciding sets, after winning or losing the first set, comebacks, sets won and lost), withdrawals given and received, and the last N matches with the score turned round to the player. Every count carries its sample size, and a percentage on an empty sample is <code>null</code>, not 0. Winners and scores are read the same way the match object publishes them, so this endpoint cannot disagree with <code>GET /matches/{matchId}</code>. ULTRA adds <code>serve_return</code>, the per-match statistics of <code>/matches/{matchId}/statistics</code> summed over the window. Each ratio is published as numerator, denominator and pct, with the number of matches behind it. Below ULTRA the key is absent. A doubles-team id answers 400 <code>doubles_not_supported</code>, and a merged or retired id answers 410 as <code>/players/{playerId}</code> does. For 1968–2022, <code>/history/archive/career</code> remains the endpoint.</li>
+<li><strong><code>height_cm</code> and <code>doubles_ranking</code> on <code>GET /players/{playerId}</code>.</strong> Both are on the single-player read only, on paid tiers, and null on a FREE key. <code>height_cm</code> comes from the results archive's record of the same person and is never inferred. <code>doubles_ranking</code> is <code>{rank, points, observed_on}</code>, the current position on the official ATP/WTA doubles table. It is null when the player is off the table or was last seen there more than 21 days ago. It is never copied into <code>ranking</code>, which stays the singles position. Nothing existing changed type or meaning.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong>The licence is named by its SPDX identifier.</strong> <code>info.license</code> now carries <code>identifier: MIT</code> in place of a URL. The old URL pointed at a GitHub page that no longer loads. No endpoint, field or type changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.60
+1.13.60: The hosted MCP endpoint, which this reference had never mentioned
+2026-10-03T00:00:00Z
+2026-10-03T00:00:00Z
+
+
+Added: The hosted MCP endpoint, which this reference had never mentioned.
+<h3>Added</h3>
+<ul>
+<li><strong>The hosted MCP endpoint, which this reference had never mentioned.</strong> A remote Streamable HTTP server has run at <code>https://mcp.livetennisapi.com/mcp</code> for weeks. The only MCP line here named the self-run package, <code>npx livetennisapi-mcp</code>. A reader who wanted an LLM agent over this API was therefore told to install and host one. The string <code>mcp.livetennisapi.com</code> appeared zero times in the reference, zero times in <code>llms.txt</code> and zero times in the README. An engineer evaluating the API found the endpoint by going looking for it, which is what prompted this. A new <strong>MCP server</strong> section now sets the hosted and self-run routes side by side. The client-libraries table, <code>llms.txt</code> and the README name the endpoint too. Everything in the section was read off the live server on 2026-10-03, not taken from a release note. <code>initialize</code> reports <code>serverInfo</code> version <strong>1.5.0</strong>. <code>tools/list</code> returns <strong>24</strong> tools, set-identical to the 24 the section lists by name. Both documented auth headers were called and both work: <code>Authorization: Bearer twjp_...</code> and <code>X-API-Key: twjp_...</code>. The rate headers read <code>limit=300</code> on a key and <code>limit=60</code> without one. The section states both, and states that the caller's own plan limit still applies underneath. That second sentence is load-bearing. The hosted cap sits above a free key's 30 a minute, so publishing it alone would read as a quota raise. With no key at all only <code>check_api_status</code> answers, and it returns the free-key URL. <code>GET /mcp</code> answers 405 <code>"use POST"</code>, so the section says POST only. <code>GET /health</code> reports the running version. Every plan reaches MCP, free keys included, and each tool reaches exactly what that plan reaches. No price is quoted anywhere, because the endpoint takes the key a customer already has. No field, parameter or type changed, and no endpoint of the REST API changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.59
+1.13.59: ?is_qualifying= on /matches, /history/matches and /fixtures, and ?gender= documented on the same three, and 1 more change
+2026-10-02T00:00:00Z
+2026-10-02T00:00:00Z
+
+
+
+Added: ?is_qualifying= on /matches, /history/matches and /fixtures, and ?gender= documented on the same three. Fixed: ?gender= was accepted and ignored on /matches?status=live and ?status=upcoming.
+<h3>Added</h3>
+<ul>
+<li><strong><code>?is_qualifying=</code> on <code>/matches</code>, <code>/history/matches</code> and <code>/fixtures</code>, and <code>?gender=</code> documented on the same three.</strong> The <code>is_qualifying</code> field has been published on every match and fixture since 2026-08-12 and this reference already called it "the only field that separates a qualifying draw from the main draw" and told callers to "filter accordingly" — while the parameter they would reach for did not exist and was accepted and ignored. Measured on prod before the change: <code>/matches?status=completed&limit=50</code> returned the same fifty ids under <code>is_qualifying=true</code>, <code>is_qualifying=false</code> and no filter at all, every one of the fifty rows carrying <code>is_qualifying: false</code>, while the <code>?draw=</code> control changed the page. The filter is applied in SQL before the page limit on <code>status=live</code>, <code>upcoming</code>, <code>completed</code> and <code>cancelled</code> and on the archive listing, so a deep crawl can page a single draw rather than filtering a full corpus client-side. It is <strong>three-valued like the field</strong>: a row whose flag is <code>null</code> matches neither <code>true</code> nor <code>false</code> and is returned only when the filter is omitted, because returning it under <code>false</code> would assert "main draw" on the strength of nothing. Coverage, measured 2026-10-02 over completed matches: the flag is stated on <strong>98.9%</strong> of the last thirty days and <strong>98.3%</strong> of 2026 outside the UTR circuit, which is where almost all unstated rows are (1,695 of the 1,831 in the last thirty days). Anything other than true/false is a 400 <code>bad_is_qualifying</code>. Asked for by a customer splitting qualifying from main draw for a match-prediction model. <code>?gender=men|women</code> has worked on the match and archive listings since 2026-09-09 and was never documented here; it is now, with the same null contract — a mixed-doubles or team-tie row matches neither value. No field or type changed.</li>
+</ul>
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>?gender=</code> was accepted and ignored on <code>/matches?status=live</code> and <code>?status=upcoming</code>.</strong> Those two statuses are served from a short-lived slate cache keyed only on the status, and the gate that decides whether a request may use it never learned about <code>gender</code>, so a filtered request was answered with the whole slate under a 200. Measured on prod before the fix: <code>status=live</code> returned the same twenty-five rows for <code>gender=men</code>, <code>gender=women</code> and no filter — six of them women's matches and seven with no gender stated — and <code>status=upcoming</code> the same sixty rows of a hundred and forty-three under every value. <code>status=completed</code> and <code>status=cancelled</code> are not cached and answered correctly throughout (sixty of sixty, pure), so the filter itself was never wrong: the request did not reach it. The live listing is polled continuously, so the cache was effectively always warm and the filter was dropped on every call rather than occasionally. Present from 2026-09-09 to 2026-10-02. Callers who filtered the live or upcoming slate by gender and trusted the result were served the unfiltered slate and should re-read anything derived from it; every row carried a correct <code>gender</code> field throughout, so a client that filtered on the field rather than the parameter was unaffected.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.58
+1.13.58: The archived tape's accept clock now begins 2026-08-11, six weeks earlier than 1.13.57 published, because the missing values were recovered rather than written…
+2026-10-01T00:00:00Z
+2026-10-01T00:00:00Z
+
+
+Changed: The archived tape's accept clock now begins 2026-08-11, six weeks earlier than 1.13.57 published, because the missing values were recovered rather than written off.
+<h3>Changed</h3>
+<ul>
+<li><strong>The archived tape's accept clock now begins 2026-08-11, six weeks earlier than 1.13.57 published, because the missing values were recovered rather than written off.</strong> 1.13.57 measured the boundary honestly at 2026-09-23, the day <code>history_archived_scores.content_changed_at</code> landed as a catalogue-only column with no backfill. What that measurement did not say is that the values were still reachable: for every archive row copied before the column existed, the live row it was copied from may still be inside the 90-day retention, and that row holds the value the archiver would itself have written. A backfill carried them across on 2026-10-01 in six bounded tranches, <strong>1,413,471 accept clocks and 1,393,458 accept counters, inside 1,538,176 rows over 14,491 matches</strong>, and left <code>fillable=0</code>. Nothing is derived, rounded or taken from a neighbouring row, and a row whose live row had already been trimmed stays null for good. Measured after the run: the archive carries the clock on <strong>100.0% of every day from 2026-08-11 to 2026-10-01</strong>, 92.5% of 2026-08-10 (the live column's own first instant is <code>2026-08-10T08:07:42Z</code>) and 0.0% of every day before it, which is <strong>1,721,102 of 4,663,181</strong> archived observed rows against 307,631 before. <code>HistoryTapeRow.sequence</code>, the accept counter, has its own later floor because <code>live_scores.match_seq</code> landed after the clock: nothing to 2026-08-13, 68.2% of 08-14, 99.3% of 08-15, then 95.9% or more of every day from <strong>2026-08-16</strong> and 100% from 2026-09-01. Two boundaries, not one. <code>received_at</code> could not move at all and the reference now says so: its live column landed 2026-09-24, a day AFTER the archive column, so every live row carrying an arrival clock was archived with it and the recoverable count is zero. Wire-verified both sides: match 2168 (26 April 2026), before the boundary, still answers 131 observed rows with <code>timestamp</code> on all 131 and <code>accepted_at</code> on none; match 189679's archived copy carried the clock on 0 of its 171 observed rows before the run and on 171 of 171 after it. The practical consequence for a timing study is that the clock no longer disappears when a match ages out of live retention. No field, endpoint or type changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.57
+1.13.57: HistoryTapeRow.accepted_at was documented as covering the whole observed corpus, and on an archived tape it begins 2026-09-23
+2026-10-01T00:00:00Z
+2026-10-01T00:00:00Z
+
+
+Fixed: HistoryTapeRow.accepted_at was documented as covering the whole observed corpus, and on an archived tape it begins 2026-09-23.
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>HistoryTapeRow.accepted_at</code> was documented as covering the whole observed corpus, and on an archived tape it begins 2026-09-23.</strong> The field is the one this reference tells a timing study to use, and its description read "It rides the whole observed corpus, so a study that spans 2026-09-24 must use this one and not <code>received_at</code>", while <code>received_at</code> said "THIS CLOCK HAS A START DATE AND THE OTHER TWO DO NOT". Both sentences were wrong about <code>accepted_at</code>. Measured over the whole archived observed corpus: <strong>307,512 of 4,659,024 rows carry it</strong>, which is 0.0% of every month to August 2026 (286 rows in all of August), 27.3% of September and 100% of October. Wire-verified on match 2168 (26 April 2026), answered with <code>meta.from_archive: true</code>: 131 observed rows, <code>timestamp</code> on all 131 and <code>accepted_at</code> on none. The measurement quoted in 1.13.5x — match 189679, <code>accepted_at</code> on all 171 observed rows — is correct and does not generalise, because that match is still inside live retention and is answered from its live rows (<code>from_archive: false</code>); its own archived copy carries the clock on <strong>0 of 367</strong> rows. So coverage depends on which copy answers the read, <code>meta.from_archive</code> is how you tell, and a null here is a boundary of the record rather than a missing value. <code>timestamp</code> is the only one of the three clocks with no start date. Found while answering a customer who asked for "a timestamp precise enough to line up with exchange quotes" — the exact question the wrong sentence was there to answer. No field, endpoint or type changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.56
+1.13.56: An exact player name that sat inside a longer name could not be asked about on any of the three name endpoints
+2026-10-01T00:00:00Z
+2026-10-01T00:00:00Z
+
+
+Fixed: An exact player name that sat inside a longer name could not be asked about on any of the three name endpoints.
+<h3>Fixed</h3>
+<ul>
+<li><strong>An exact player name that sat inside a longer name could not be asked about on any of the three name endpoints.</strong> Every name lookup is a substring match, so there was no way to say "exactly this one", and a person whose name is a prefix of somebody else's was unreachable: <code>GET /h2h?p1=Carlos%20Alcaraz</code> answered <code>400 ambiguous_name</code> listing <code>Carlos Alcaraz</code> and <code>Carlos Alcaraz Gonzalez</code>, with no string that could pick between them. Measured on the live roster and corpus: <strong>14 of the 1,177 ranked players</strong> on <code>/h2h</code>, <strong>21 of 1,500</strong> distinct archive names on <code>/history/archive/career</code> and <strong>4 of 1,739</strong> charted names on <code>/charting/players</code>, the men's world #2 and #3 among them. Passing a name exactly now selects that person on all three. A name given in part is still a fragment and still refuses with the candidate list, so <code>Sinner</code> continues to list Jannik and Martin rather than guessing, and a single name charted under both genders on <code>/charting/players</code> still needs <code>gender</code>. The refusal also reports the candidates it would have shown before the exact match was applied, so a hint never gets shorter than it was. Verified on the wire: Alcaraz against Sinner returns 18 meetings 11-7 with the surface split and the ULTRA aggregates. No field, endpoint, parameter or error code changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.55
+1.13.55: ITF World Tennis Tour doubles began stating outcome, and the reference still said no doubles draw carries either field
+2026-09-29T00:00:00Z
+2026-09-29T00:00:00Z
+
+
+Fixed: ITF World Tennis Tour doubles began stating outcome, and the reference still said no doubles draw carries either field.
+<h3>Fixed</h3>
+<ul>
+<li><strong>ITF World Tennis Tour doubles began stating <code>outcome</code>, and the reference still said no doubles draw carries either field.</strong> The 1.13.42 note measured 454 completed doubles matches and found no tagged row. That was true when it was written. The first doubles row carried a tag on 2026-09-23, and by 2026-09-29 <code>GET /matches/{id}/points</code> published <code>outcome</code> on <strong>104 of 596</strong> completed ITF doubles matches, 17%, against <strong>0 of 371</strong> for the ATP, WTA and Challenger doubles draws in the same window. The vocabulary is ace and double fault, as on the ITF singles draws. <code>serve</code> is still unstated on every doubles draw, measured 0, and that sentence is unchanged. The cause sat before the tagger: a dedup change let a doubles fixture be adopted and oriented by its member sets instead of being refused, so the ITF per-point join reached rows it had never been able to match. 17% falls between "almost none" and "a clear majority", so the description states it as a measured fraction and sends the reader to <code>enrichment</code> per match instead of to the draw. WTA main-tour qualifying has carried that treatment since 1.13.42. Found by the every-sweep tag-coverage check. No field, endpoint or behaviour changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.54
+1.13.54: The two broadcast surfaces, live since 2026-09-24 and documented nowhere a customer could read, and 1 more change
+2026-09-28T00:00:00Z
+2026-09-28T00:00:00Z
+
+
+Added: The two broadcast surfaces, live since 2026-09-24 and documented nowhere a customer could read. The 1 Hz polling contract, which is the part a live scoreboard gets wrong.
+<h3>Added</h3>
+<ul>
+<li><strong>The two broadcast surfaces, live since 2026-09-24 and documented nowhere a customer could read.</strong> <code>GET /broadcast/match/{matchId}</code> and <code>GET /broadcast/live</code> have served every ULTRA key for four days and appeared in no published reference: the write-up went into the tennis repository, which nothing serves. Both are now specified, with the 54-field <code>BroadcastObject</code> schema in the same order the API returns it, verified field by field against a live response rather than transcribed from the source. The object exists because a graphics template binds each field to a fixed path, so three properties are contractual rather than incidental and are documented as such: it is FLAT, every key is PRESENT on every read (<code>null</code>, never missing — a missing key breaks a binding), and the score is already in DISPLAY form (<code>p1_points</code> reads <code>"40"</code> or <code>"AD"</code>, <code>set_line</code> is one string, each set has its own column). It is a projection of what the API already publishes plus the break-, set- and match-point flags the scoring pipeline derives; nothing in it is a new fact and nothing is guessed.</li>
+<li><strong>The 1 Hz polling contract, which is the part a live scoreboard gets wrong.</strong> Every response carries a strong <code>ETag</code> and <code>Cache-Control: no-cache, max-age=1</code>, so <code>If-None-Match</code> turns an unchanged state into a 304 with no body; ULTRA's 600-a-minute burst window means ten single-match pollers at 1 Hz fit on one key, and <code>/broadcast/live</code> stays one request however many matches are live. A new state is detected on <code>sequence</code>, never on <code>served_at</code>, which changes on every read — stated on both routes because polling on <code>served_at</code> is the mistake that looks like it works.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.53
+1.13.53: The tape now publishes the accept counter the withdrawal ledger points at, and 1 more change
+2026-09-26T00:00:00Z
+2026-09-26T00:00:00Z
+
+
+
+Added: The tape now publishes the accept counter the withdrawal ledger points at. Fixed: Score.sequence said the tape does not carry it, and the stated reason was the gap.
+<h3>Added</h3>
+<ul>
+<li><strong>The tape now publishes the accept counter the withdrawal ledger points at.</strong> <code>GET /matches/{matchId}/withdrawals</code> tells you a <code>sequence</code> was withdrawn and to resume from <code>replaced_by_sequence</code>, and until today no served history-tape row carried that number, so the pointer named a row nothing identified. <code>HistoryTapeRow.sequence</code> is the same per-match accept counter as <code>Score.sequence</code> on a live read. Because a withdrawn state never enters the record at all, a withdrawal now reads as a GAP in the tape's run of counters, which is what makes "resume from the predecessor" resolvable against the served history. Reported by an Ultra customer who verified three withdrawal records against the tape and wrote that "the history rows lack sequence IDs, so these responses alone cannot reconstruct exactly what a live client received". Null, never fabricated, on a row archived before 2026-09-26 (the archive did not keep the counter, so it died with the live row at retention) and on the pre-2023 archive tape, which has no such counter. Unlike <code>accepted_at</code> and <code>received_at</code> it is NOT null on a reconstructed row: a counter is a fact about our own accept order.</li>
+</ul>
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>Score.sequence</code> said the tape does not carry it, and the stated reason was the gap.</strong> The description read "absent on history-tape rows, which are already served in order". Being served in order does not resolve a <code>replaced_by_sequence</code>, which is a pointer at a counter rather than a position in a list. Corrected, and it now points at <code>HistoryTapeRow.sequence</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.52
+1.13.52: The tape's two accept clocks are published, and 3 more changes
+2026-09-25T00:00:00Z
+2026-09-25T00:00:00Z
+
+
+
+Added: The tape's two accept clocks are published. The arrival clock's start date, which is the part that bites. Plus 2 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong>The tape's two accept clocks are published.</strong> <code>HistoryTapeRow.accepted_at</code> and <code>HistoryTapeRow.received_at</code> have ridden every observed row of <code>GET /history/matches/{matchId}</code> since 2026-09-24 and were documented nowhere, so a customer doing timing research on a tape had no published field to read. <code>accepted_at</code> is the instant our arbiter accepted the state, stamped once and never refreshed, which is the clock to difference against: <code>timestamp</code> starts equal to it and is then refreshed while the owning source keeps re-asserting the unchanged state, so on most raw rows <code>timestamp</code> is a later re-assertion instant (on match 189679, 164 of its 171 observed rows carry a <code>timestamp</code> more than a second after <code>accepted_at</code>). <code>received_at</code> is the instant the state reached our edge before arbitration, so <code>accepted_at</code> minus <code>received_at</code> is our own arbitration step. Neither figure is a claim about speed before our edge and neither is the court's clock. Both are null on every reconstructed row, which never had a clock.</li>
+<li><strong>The arrival clock's start date, which is the part that bites.</strong> <code>received_at</code> is stamped only as a state arrives, so it can never be filled in afterwards. The earliest one we hold is <code>2026-09-24T02:00:26Z</code>; on every observed row accepted before that instant it is null while <code>timestamp</code> and <code>accepted_at</code> are populated. A study that spans the boundary has to read <code>accepted_at</code>, which rides the whole observed corpus. Measured on match 189679 (12 September) after a Basic customer reported it: 171 observed rows, <code>accepted_at</code> on all 171, <code>received_at</code> on none.</li>
+<li><strong>What <code>server</code> guarantees on a tiebreak row, and what it does not.</strong> Tiebreak rows are never settled on either sequence. The server genuinely changes during a breaker, so there is no constant to take a majority of, and the field is the source's own assertion carried through. Where a breaker's rows skip points the rotation cannot be reconstructed from what is served: you have the server of a state, not of a numbered point, and <code>meta.points.transitions_legal</code> against <code>transitions_total</code> tells you whether that happened. On the set roll-up row carrying <code>tiebreak_final</code>, <code>server</code> is the value of that committed set-end state and is NOT a preserved copy of the final breaker point's server. Reported by a Basic customer who could not reconcile 21 tiebreak transitions against the standard rotation and was right not to.</li>
+</ul>
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>Score.accepted_at</code> said "absent on history-tape rows".</strong> It has been present on them since 2026-09-24, so the description had been wrong for a day, on the one field a timing study is told to use. Corrected, and the tape row now documents its own copy.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.51
+1.13.51: The score_withdrawn frame is documented on the push feed page, where a streaming consumer looks for it
+2026-09-25T00:00:00Z
+2026-09-25T00:00:00Z
+
+
+Added: The score_withdrawn frame is documented on the push feed page, where a streaming consumer looks for it.
+<h3>Added</h3>
+<ul>
+<li><strong>The <code>score_withdrawn</code> frame is documented on the push feed page, where a streaming consumer looks for it.</strong> 1.13.50 put the record and the frame's fields on the score read and in the schema list, and the frame is a STREAM artifact: it arrives on the same channel as the <code>score</code> frame it retracts. <code>GET /ws-token</code> now carries the frame itself, the reason it exists (the feed publishes every accepted state and runs no trust deference, while the score read runs the legality gate and can serve an older state), what each <code>kind</code> asks you to do, and the three reads that recover a judgement you were offline for. No spec surface changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.50
+1.13.50: A withdrawn score state is now recoverable, and GET /matches/{matchId}/withdrawals serves the record, and 2 more changes
+2026-09-25T00:00:00Z
+2026-09-25T00:00:00Z
+
+
+
+Added: A withdrawn score state is now recoverable, and GET /matches/{matchId}/withdrawals serves the record. The score_withdrawn frame's own fields are published for the first time. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><strong>A withdrawn score state is now recoverable, and <code>GET /matches/{matchId}/withdrawals</code> serves the record.</strong> The <code>verdict</code> object has told a stream consumer since 2026-09-23 that a state it holds was taken back, but it is recomputed on every read over the newest state, so it returned null again as soon as the next state was accepted. In a live match that is seconds. A consumer that disconnected and missed the <code>score_withdrawn</code> frame had no way back to it: the publisher's duplicate ledgers live inside the publishing process, and nothing else recorded the judgement. Every withdrawal is now written down when the judgement is made, before the frame goes out. Three reads use it. <code>GET /matches/{matchId}/withdrawals</code> (ULTRA, the tier that receives the frame) returns every refused <code>sequence</code> on a match, oldest first, each with <code>kind</code>, <code>reason</code>, <code>withdrawn_at</code>, <code>published_at</code>, the <code>sequence</code> served in its place, and the frame exactly as the feed sent it. <code>GET /matches?withdrawn_since=<instant></code> lists the match ids with a withdrawal since then, with a count per match, so a consumer coming back after a gap knows which matches to ask about. And <code>?sequence=N</code> on <code>GET /matches/{matchId}/score</code> makes <code>verdict</code> answer for the one sequence you hold, even after later states have landed. Records are kept 90 days from the withdrawal.</li>
+<li><strong>The <code>score_withdrawn</code> frame's own fields are published for the first time.</strong> <code>type</code>, <code>match_id</code>, <code>kind</code>, <code>superseded_sequence</code>, <code>reason</code>, <code>safe_to_resume</code>, <code>ts</code> and <code>published_at</code>. It carries NO replacement score and no new sequence, which is the question a customer had to ask us: the state to resume from is the one on the next <code>score</code> frame, or on a fresh <code>GET /matches/{matchId}/score</code>, and the sequence served in the refused state's place is on the withdrawal record rather than on the frame. <code>ts</code> is when the judgement was made, <code>published_at</code> when the frame was handed to the feed.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong>Only a discarded withdrawal is re-asserted by <code>?sequence=</code>.</strong> A <code>deferred</code> state may since have been corroborated and a <code>withheld</code> read described that moment's scan, so neither answers from the record; a verdict the read produces itself always wins, because it describes the state actually being served. <code>?sequence=</code> changes <code>verdict</code> and nothing else on the object. A non-integer is <code>400 bad_sequence</code> rather than an accepted-and-ignored filter, because a silently dropped lookup would read as "not withdrawn". <code>?withdrawn_since=</code> likewise takes only <code>limit</code> and <code>offset</code> beside it: any match filter, <code>status</code> or <code>updated_since</code> alongside it is <code>400 bad_withdrawn_since</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.49
+1.13.49: The tournament catalogue takes ?tier=, which it had been accepting and ignoring
+2026-09-24T00:00:00Z
+2026-09-24T00:00:00Z
+
+
+Fixed: The tournament catalogue takes ?tier=, which it had been accepting and ignoring.
+<h3>Fixed</h3>
+<ul>
+<li><strong>The tournament catalogue takes <code>?tier=</code>, which it had been accepting and ignoring.</strong> <code>GET /tournaments</code> publishes each row's <code>tier</code> and <code>tier_source</code>, and the two match listings have filtered on the same vocabulary since 1.13.28. The catalogue did not: <code>?tier=atp_500</code>, <code>?tier=laver_cup</code> and an outright invalid value each returned the entire 10,302-row catalogue under a 200, byte-identical to no filter at all, because the level was attached to a page after it had already been selected. It now selects. The one difference from the match listings is the season, and it is stated on the parameter: a catalogue row has no season of its own, so <code>?tier=</code> here reads the CURRENT season, the same season the row's own <code>tier</code> field reports, while on <code>/matches</code> and <code>/history/matches</code> it reads each match's own season. A tournament with no row for the current season matches no value. An invalid value is now a 400 <code>bad_tier</code> carrying the vocabulary, as it always was elsewhere.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.48
+1.13.48: tiebreak_final on the history tape row is documented for the first time, and the page-first rule the reference stated is no longer true, and 1 more change
+2026-09-24T00:00:00Z
+2026-09-24T00:00:00Z
+
+
+Fixed: tiebreak_final on the history tape row is documented for the first time, and the page-first rule the reference stated is no longer true. The row that closes a tiebreak set carries the score whichever row the set counter moves on.
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>tiebreak_final</code> on the history tape row is documented for the first time, and the page-first rule the reference stated is no longer true.</strong> A PRO customer wired a bot to the closing score and reported that match 195134 served a <code>tiebreaks</code> array of <code>{p1: 2, p2: 7}</code> while no tape row carried <code>tiebreak_final</code>, and that it stayed that way on every later read. Two things were wrong here. The field had never appeared in this reference on <code>HistoryTapeRow</code> at all, though it is served on <code>GET /history/matches/{matchId}</code> and a customer had been pointed at it in writing; it is now described beside <code>is_tiebreak</code>. And the <code>MatchPoints</code> text said a page that *starts* on a roll-up row "has no previous row to read and uses the recorded finals alone", which was true when written and is not true now: since 2026-09-24 such a page reads the one row before it, so the value does not depend on where the page was cut. Both endpoints resolve the score the same way and agree with the response's own <code>tiebreaks</code> array.</li>
+<li><strong>The row that closes a tiebreak set carries the score whichever row the set counter moves on.</strong> The tape's set counter lags the games array by a row: the row that took 195134's first set to 6-7 still read <code>sets</code> <code>[0, 0]</code>, and the counter reached <code>[0, 1]</code> on the row after it. The stamp only fired when the counter grew on that same row, so a set whose counter lagged was never stamped, and nothing revisits a tape. The games on the row now decide it, judged by the tournament's own set test so the Next Gen first-to-4 format is read by its own rule. Measured on the live endpoint before the change: of 9 live matches whose <code>tiebreaks</code> array stated a final, 2 carried it on no row, both first-set breakers.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.47
+1.13.47: A set completion that does not end the match is assigned a sequence only once confirmed
+2026-09-24T00:00:00Z
+2026-09-24T00:00:00Z
+
+
+Changed: A set completion that does not end the match is assigned a sequence only once confirmed.
+<h3>Changed</h3>
+<ul>
+<li><strong>A set completion that does not end the match is assigned a <code>sequence</code> only once confirmed.</strong> Live since 2026-09-24 and stated here for the first time. A live scoring display can render the set-ending frame ON set point, before the point is decided, and step back when the point is not won; a customer measured exactly that twice. A rise in <code>sets</code> that does not end the match, arriving while the state we hold is set point for the side that would gain the set, is now held until the same source keeps presenting it for 10 s or more, a source at least as trusted presents it, or the same source shows play continuing inside the new set. The unconfirmed frame is never stored, so no <code>sequence</code> ever carries it and nothing is published as provisional — a <code>sequence</code> that carries a set completion is one that met the rule. Until then <code>GET /matches/{matchId}/score</code>, the push feed and the WebSocket all keep serving the set-point state; a frame nobody confirms within 120 s is discarded. A set completion that <strong>ends</strong> the match is exempt and follows the completion rules instead. Measured over the 30 days to 2026-09-24: a median 10 s added between the point and the published set-end, and on a feed that does not re-present states the set-end lands with the first point of the next set (90th percentile 133 s). It does not cover a set genuinely won and then corrected on court, which can still be published and rewound — read <code>corroborated</code> beside it, and <code>verdict</code> says so when a state is taken back.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.46
+1.13.46: corroborated and corroborated_age_seconds on the Score: has an independent source vouched for THIS EXACT state?, and 1 more change
+2026-09-24T00:00:00Z
+2026-09-24T00:00:00Z
+
+
+Added: corroborated and corroborated_age_seconds on the Score: has an independent source vouched for THIS EXACT state? changing_sources_count on the Score is documented.
+<h3>Added</h3>
+<ul>
+<li><strong><code>corroborated</code> and <code>corroborated_age_seconds</code> on the Score: has an independent source vouched for THIS EXACT state?</strong> Raised by a customer who acted on a completed set that was taken back seconds later. <code>corroborated</code> is <code>true</code> when a source other than the one that wrote the state has since presented us the same state, <code>false</code> when none has — including a brand-new state nobody has had a chance to agree with yet — and <code>null</code> on the archive fallback of a retention-trimmed completed match, where <code>age_seconds</code> and <code>sources_count</code> are null too. <code>corroborated_age_seconds</code> is the seconds since that vouch, null when there is none. <strong>This is the field to gate a set-end on.</strong> <code>sources_count</code> and <code>changing_sources_count</code> are window measures over the match and say nothing about the state in front of you, so a set completion that only one source has announced can sit beside a perfectly healthy count — and a set-end is the one state whose correction changes a result rather than a number. It is evidence, not proof: a vouched state can still be corrected, so compare <code>corroborated_age_seconds</code> with <code>age_seconds</code> and treat a vouch younger than the state it vouches for as the weaker signal it is. When a state is taken back the <code>sequence</code> goes down and <code>verdict</code> says so. Present wherever <code>sources_count</code> is; a yes/no, never a name.</li>
+<li><strong><code>changing_sources_count</code> on the Score is documented.</strong> Live since 2026-09-24 and absent from this reference. How many distinct sources CHANGED the score in the last 300 seconds, counted on the content clock alone, so a source re-asserting an unchanged state is not counted — the difference from <code>sources_count</code>, which a feed stuck re-sending its last state holds at 1 for as long as the echo lasts while this reads 0. The window is wider on purpose: a point lands every 30-60 s and a changeover is about 90 s. A cross-confirmation is agreement, not a change, so this can read below <code>sources_count</code> and never above it.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.45
+1.13.45: GET /history/incidents and GET /history/incidents/{incidentId}/matches are documented for the first time, and 1 more change
+2026-09-24T00:00:00Z
+2026-09-24T00:00:00Z
+
+
+Added: GET /history/incidents and GET /history/incidents/{incidentId}/matches are documented for the first time. The verdict object on the Score is documented.
+<h3>Added</h3>
+<ul>
+<li><strong><code>GET /history/incidents</code> and <code>GET /history/incidents/{incidentId}/matches</code> are documented for the first time.</strong> Both have been live and neither appeared anywhere in this reference, so the register a customer is meant to reconcile against could not be found from the documentation. <code>/history/incidents</code> returns the published data-quality incidents, oldest window first, each with its <code>window_utc</code> (start, end and the <code>basis</code> it was read from), <code>rule_before</code>, <code>rule_after</code>, <code>defect</code>, <code>affected_surfaces</code>, <code>how_to_tell</code>, <code>corrections</code> and <code>matches</code>. The register is part of a release and is not inferred at read time, so a record does not change between releases. <code>/history/incidents/{incidentId}/matches</code> streams one row per affected match as JSONL, or CSV with <code>?format=csv</code>, with the columns <code>incident_id</code>, <code>match_id</code>, <code>tournament</code>, <code>tour</code>, <code>round</code>, <code>draw_stage</code>, <code>scheduled_time</code>, <code>affected_field</code>, <code>served</code> and <code>expected</code>. A match whose derivation would be a guess is left out and the incident's <code>matches.note</code> says which, so the export is exact rather than complete. Measured on the live API 2026-09-24: five incidents across the kinds <code>pricing_rule</code>, <code>version_stamp</code>, <code>field_orientation</code> and <code>field_repair</code>.</li>
+<li><strong>The <code>verdict</code> object on the Score is documented.</strong> Live since 2026-09-23 and absent from this reference. It is null on an ordinary read. When the legality gate refuses a state a stream has already published, it carries <code>kind</code> (<code>withdrawn</code>, <code>deferred</code> or <code>withheld</code>), <code>superseded_sequence</code> (the <code>sequence</code> a stream consumer is holding), <code>reason</code> and <code>safe_to_resume</code>, which is true exactly when <code>kind</code> is <code>withdrawn</code>. The push feed, the WebSocket and webhooks announce the same judgement as a <code>score_withdrawn</code> frame naming the same sequence, so a stream consumer learns a state was taken back without polling. <code>GET /matches/{matchId}/score</code> now names the field and points at the incident register.</li>
+</ul>
+<p>No field, endpoint or behaviour changed by this release; both were already served.</p>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.44
+1.13.44: GET /matches/{matchId}/points: the row that closes a tiebreak set now carries tiebreak_final, and the reference did not say so
+2026-09-23T00:00:00Z
+2026-09-23T00:00:00Z
+
+GET /matches/{matchId}/points: the row that closes a tiebreak set now carries tiebreak_final, and the reference did not say so.
+<ul>
+<li><strong><code>GET /matches/{matchId}/points</code>: the row that closes a tiebreak set now carries <code>tiebreak_final</code>, and the reference did not say so.</strong> Raised by a Pro customer running trading bots, who reported that the tape stops one point short of every tiebreak and that a field filling once a day was no use to a decision taken while the match is live. A tiebreak's last point is carried, like every set-winning point, by the set roll-up row (<code>tiebreak</code> false, the set banked 7-6, <code>score</code> 0-0), and that row never stated the breaker's score at closure. It now carries <code>tiebreak_final: [p1, p2]</code> in our player order, e.g. <code>[7, 5]</code>. Exact when the previous row is the decided score or one point from it, so it is derived from the stream itself and is present as soon as the set closes; otherwise taken from the match's recorded tiebreak finals, the same finals <code>GET /history/matches/{matchId}</code> publishes as <code>tiebreaks</code>; <code>null</code> when neither can state it, never guessed. Absent on every other row, including the breaker rows. A page that *starts* on a roll-up row has no previous row to read and uses the recorded finals alone. Measured at the endpoint over the 30 hours to 2026-09-23 20:30Z: 60 completed matches held 67 tiebreak sets and 36 of them served a closing score, only 7 of those matches having a recorded final at the time. Documentation of an already-served field; no behaviour changed by this release.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.43
+1.13.43: The reference said a FREE key is refused the history endpoints outright. It is not, and has not been since 2026-08-07: a FREE key is served 20 history calls…
+2026-09-23T00:00:00Z
+2026-09-23T00:00:00Z
+
+
+Fixed: The reference said a FREE key is refused the history endpoints outright. It is not, and has not been since 2026-08-07: a FREE key is served 20 history calls per calendar month.
+<h3>Fixed</h3>
+<ul>
+<li><strong>The reference said a FREE key is refused the history endpoints outright. It is not, and has not been since 2026-08-07: a FREE key is served 20 history calls per calendar month.</strong> Three separate statements carried the wrong rule — the FREE tier summary ("No historical results"), the <code>/matches</code> description ("requires BASIC ... and returns <code>403 upgrade_required</code> on a FREE key"), and the <code>status</code> parameter. Measured on the live API 2026-09-23 with a fresh FREE key: <code>GET /history/matches</code> answered <code>200</code> and kept answering until the monthly allowance was spent, after which it answered <code>403 upgrade_required</code> carrying <code>free_history_taste: "used"</code> and the detail "your 20 free history calls this month are used". A reader planning against the old text would have concluded the endpoints were closed to them and either bought a tier they did not yet need or abandoned the evaluation; a reader who tried anyway got a <code>200</code> the reference said was impossible and could not tell an allowance from a private grant. <code>/history/matches</code>, <code>/history/coverage</code>, <code>status=completed</code> on <code>/matches</code> and the FREE tier summary now all state the allowance and the refusal that follows it.</li>
+</ul>
+<p>No field, endpoint or behaviour changed; every correction is to a description.</p>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.42
+1.13.42: Every per-point coverage count in this reference was measured in our own tables and published as though it described the API. It did not, and the gap is large, and 1 more change
+2026-09-23T00:00:00Z
+2026-09-23T00:00:00Z
+
+
+Fixed: Every per-point coverage count in this reference was measured in our own tables and published as though it described the API. It did not, and the gap is large. enrichment per match was correct throughout, on every one of these matches, and basis has always named which sequence you were given.
+<h3>Fixed</h3>
+<ul>
+<li><strong>Every per-point coverage count in this reference was measured in our own tables and published as though it described the API. It did not, and the gap is large.</strong> <code>GET /matches/{id}/points</code> serves a COMPLETED match's stored per-point stream only while that stream passes our quality bar. Where it does not, the endpoint serves a measured-complete RECONSTRUCTION instead, and a reconstruction carries no clock and no <code>serve</code> / <code>outcome</code> tags at all. So the share of matches whose tags a reader can fetch is lower, sometimes far lower, than the share of matches a source tagged. Re-measured 2026-09-23 on what the endpoint publishes, beside what the previous text claimed: <strong>Davis Cup World Group <code>serve</code> 2 of 22 (published as 22 of 22), World Group I 1 of 42 (published as 41 of 42)</strong>; WTA qualifying <code>serve</code> <strong>18 of 48</strong> (published as 48 of 48); ATP Challenger qualifying <code>serve</code> <strong>137 of 197</strong> (195 of 197) and <code>outcome</code> <strong>134 of 197</strong> (191 of 197); WTA 125 qualifying <code>serve</code> <strong>47 of 63</strong> (61 of 63) and <code>outcome</code> <strong>30 of 63</strong> (24 of 63); ITF qualifying <code>outcome</code> <strong>181 of 215</strong> men's and <strong>142 of 176</strong> women's (189 of 209, 143 of 168); WTA main-tour qualifying <code>outcome</code> <strong>5 of 48</strong>, against 47 of 88 in the same events' main draws (published as 3 of 48 against 82 of 88). Both descriptions now state which layer they count and name <code>basis</code> as the per-response answer.</li>
+<li><code>enrichment</code> per match was correct throughout, on every one of these matches, and <code>basis</code> has always named which sequence you were given. Only the prose was wrong.</li>
+</ul>
+<p>No field, endpoint or behaviour changed; every correction is to a description.</p>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.41
+1.13.41: The outcome description named the wrong tour as the qualifying gap, and it was wrong in both directions, and 2 more changes
+2026-09-22T00:00:00Z
+2026-09-22T00:00:00Z
+
+
+Fixed: The outcome description named the wrong tour as the qualifying gap, and it was wrong in both directions. tour: challenger pools two populations with different outcome coverage, and the reference described only one of them. Plus 1 more change.
+<h3>Fixed</h3>
+<ul>
+<li><strong>The <code>outcome</code> description named the wrong tour as the qualifying gap, and it was wrong in both directions.</strong> It read "Qualifying draws follow their own tour, with one gap: ... Challenger (men) qualifying states no outcome at all." Measured 2026-09-22 over completed matches carrying a per-point stream since 2026-09-12, <strong>ATP Challenger qualifying states <code>outcome</code> on 191 of 197 matches, with the full five-value vocabulary</strong> (6,881 unforced errors, 4,275 winners, 4,267 forced errors, 1,026 aces, 774 double faults) — a reader who segregated a corpus on that sentence discarded the single richest qualifying population in the product. The gap is on the <strong>WTA main tour</strong>, which the old sentence implicitly cleared: WTA qualifying states <code>outcome</code> on <strong>3 of 48</strong>, against 82 of 88 in the same events' main draws. ITF qualifying is unchanged in direction (189 of 209 men's, 143 of 168 women's) and WTA 125 qualifying is partial (24 of 63). <code>enrichment</code> per match was correct on every one of these matches throughout — only the prose was wrong, and the prose is what a corpus is planned from.</li>
+<li><strong><code>tour: challenger</code> pools two populations with different <code>outcome</code> coverage, and the reference described only one of them.</strong> It read "Challenger: all five". That is true of ATP Challenger (<code>tier: challenger_50</code> … <code>challenger_175</code>): 252 of 255 completed main-draw matches with a stream carry the full five. It is not true of the WTA 125 events, which are also served as <code>tour: challenger</code> (<code>tier: wta_125</code>) and state <strong>ace and double fault only</strong> — 137 of 141 main-draw matches carry <code>outcome</code>, none of them a winner, forced error or unforced error. Read <code>tier</code>, not <code>tour</code>, to tell the two apart. Both descriptions now say so.</li>
+<li><strong>The <code>serve</code> qualifying denominators are re-measured</strong> (WTA qualifying 48 of 48, ATP Challenger qualifying 195 of 197, WTA 125 qualifying 61 of 63) and name the populations by <code>tier</code> rather than by "Challenger (men)" / "Challenger (women)", which the served fields do not distinguish.</li>
+</ul>
+<p>No field, endpoint or behaviour changed; every correction is to a description.</p>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.40
+1.13.40: tier and tier_source on every match and tournament object, with ?tier= on GET /matches and GET /history/matches, and 5 more changes
+2026-09-22T00:00:00Z
+2026-09-22T00:00:00Z
+
+
+
+Added: tier and tier_source on every match and tournament object, with ?tier= on GET /matches and GET /history/matches. A correction to a published result is signalled, never a silent edit: basis: restatement on the status history, result_restated_at / result_version on every match object, and ?restated_since= on GET /history/matches. Plus 4 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong><code>tier</code> and <code>tier_source</code> on every match and tournament object, with <code>?tier=</code> on <code>GET /matches</code> and <code>GET /history/matches</code>.</strong> <code>tier</code> is the level the tournament was played at in the season of the match — the official category as the tour publishes it — and answers the question <code>category</code> cannot: is this a WTA 125, an ITF W35 or a Challenger 75? The vocabulary is closed: <code>grand_slam</code>; the ATP levels (<code>atp_finals</code>, <code>atp_1000</code>, <code>atp_500</code>, <code>atp_250</code>, <code>next_gen_finals</code>); the WTA levels (<code>wta_finals</code>, <code>wta_elite_trophy</code>, <code>wta_1000</code>, <code>wta_500</code>, <code>wta_250</code>, <code>wta_125</code>); <code>challenger_175</code> … <code>challenger_50</code>; the ITF World Tennis Tour categories (<code>itf_m15</code>, <code>itf_m25</code>, <code>itf_w15</code> … <code>itf_w100</code>, with the 2023 women's categories kept as printed that season); the team events (<code>united_cup</code>, <code>davis_cup</code>, <code>bjk_cup</code>, <code>laver_cup</code>, <code>olympics</code>); <code>juniors</code>; <code>utr</code>; <code>exhibition</code>; or <code>null</code> — nothing we hold names the level, never guessed. The tier is per season and it moves under one <code>tournament_id</code>: Dallas was <code>atp_250</code> in 2023 and 2024 and <code>atp_500</code> from 2025, so a 2024 Dallas match reads <code>atp_250</code> and a 2025 one <code>atp_500</code> under the same id. A match's tier is resolved by the calendar year of its <code>scheduled_time</code>, never copied from the tournament row; <code>/tournaments</code> rows carry the current season's, null when this season's calendar does not list the event. <code>tier_source</code> says how the level was established — <code>calendar</code> (the official per-season tour calendar), <code>name</code> (an unambiguous name rule: the ITF category is in the event's official name, team and UTR events are named as such), <code>wikipedia</code> (the season's schedule page, only where the official calendar could not state the level for that season), <code>resolver</code> (resolved after the seed dataset by the same rules) — and is null exactly when <code>tier</code> is null. The 2023–2026 seasons were verified event by event against the official calendars (99.1% of tournament-seasons resolved; the rest deliberately null). <code>?tier=</code> takes comma-separated exact values, read from the same per-season table so filter and field cannot disagree; a null tier matches no value; an unknown value is a <code>400 bad_tier</code> with the offending values in <code>bad</code> and the full vocabulary in <code>allowed</code>. <code>category</code> is unchanged and remains the coarse class.</li>
+<li><strong>A correction to a published result is signalled, never a silent edit: <code>basis: restatement</code> on the status history, <code>result_restated_at</code> / <code>result_version</code> on every match object, and <code>?restated_since=</code> on <code>GET /history/matches</code>.</strong> Raised by a licensed results customer settling off our results. Once a match has been published as <code>completed</code> (or as a cancelled walkover with a winner), any later change to <code>status</code>, <code>event_status</code>, <code>winner</code> or the final score is appended to <code>GET /matches/{matchId}/status-history</code> as a new row with <code>basis: restatement</code>, at the instant we made the change: <code>status</code> / <code>event_status</code> carry before and after as usual (the same value twice when only the winner or the score moved) and <code>score</code> is the result after the correction. This holds for every path that can change a result — a source's late final, a second authority's correction, an operator's repair, a completion reopened and re-closed — because it is enforced where the result is written, not by each path remembering to say so; a re-assertion of the same result, or a change to points alone, writes nothing. <code>result_restated_at</code> is the instant of the newest such row (null while the result stands as first published); <code>result_version</code> is <code>1</code> plus the number of corrections — record it with the result you settle on, and a higher number on re-read means the one you hold was superseded. <code>?restated_since=<ISO instant></code> keeps only the matches corrected after that instant — the poll to run after each settlement pass; a <code>Z</code> or an offset, a naive value read as UTC, and a bare date refused as <code>400 bad_restated_since</code>, because a day is not an instant. Rows with this basis exist from 2026-09-22; earlier corrections were not signalled and nothing is reconstructed for them.</li>
+<li><strong><code>winner</code> on every row of every history tape: who won the point that produced the row.</strong> Raised by a modelling customer initialising serve states from the tape — <code>server</code> and <code>serve</code> were stated per row, and the third fact a model needs was only implicit in the score step. <code>winner</code> (<code>1</code> | <code>2</code> | null) is judged from the score step between the row and the previous served row (the previous raw row on the raw sequence, the previous clean row under <code>?sequence=clean</code>), by the same rule the completeness ledger counts a legal transition with — and from nothing else: never the serve, the server, the outcome tag or the pattern of play. Null on the first row and wherever the step is not one attributable point: a re-sent row, a set opener, a backward correction, a multi-game jump. A step whose games total rises by exactly one attributes to the side whose count rose — the game point was theirs — whatever in-game points the step skipped, a missed poll or a game withheld whole (<code>meta.points.games_withheld</code>). Present on the raw and clean sequences, on <code>?points=complete</code> reads and on the pre-2023 archive tape; equal to <code>point_winner</code> wherever that older key is present, which is kept unchanged for existing readers. Derived once per response, never stored.</li>
+<li><strong><code>schema_version</code> on the packages listing, and bulk tape rows carry the enriched tape.</strong> Under <code>schema_version</code> 2 a <code>kind=tape</code> JSONL line is exactly what <code>GET /history/matches/{matchId}</code> returns for that match on its default basis, produced by the same code — the same rows with <code>origin</code>, <code>serve</code>, <code>outcome</code> and <code>winner</code> beside the score columns, and the full <code>meta</code> (<code>enrichment</code>, <code>reconstructed_at</code>, <code>observed_span</code>, <code>meta.points</code> where served) — so a model reads its serve states from the bulk file just as it would from the endpoint; the CSV appends <code>origin</code>, <code>serve</code>, <code>outcome</code>, <code>winner</code> after <code>danger</code>, in that order, and its column order grows only at the end. <code>1</code> is the shape every month built before 22 September 2026 carries: score columns only, no <code>meta.enrichment</code>, and a CSV that ends at <code>danger</code>. Months built before that date keep shape <code>1</code> until they are rebuilt — newest month first, then backwards, <code>built_at</code> and <code>sha256</code> moving as each flips — so check the manifest rather than assuming; a consumer that needs the enriched fields should require <code>schema_version >= 2</code> and, for a month still at <code>1</code>, read the per-match endpoint for the matches it needs. <code>null</code> on the non-tape kinds.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong>A completed match on the tours with a complete per-point record is re-joined after the match against that complete sequence.</strong> ATP 250/500/1000, ATP Challenger and WTA 1000/500/250/125: the live join paired <code>serve</code> / <code>outcome</code> onto points as they were captured, so a point tagged after our capture went untagged. A completed match is now re-joined against the complete sequence by the same exact-state rule, and its serve/outcome coverage exceeds what was captured live. The late tags arrive as <code>point_update</code> frames and <code>tagged_at</code> revisions like any other — collect them with <code>?changed_since=</code>. Nothing is inferred; a point that is never stated stays null.</li>
+<li><strong>The pre-2023 archive tape no longer claims <code>point_winner</code> is null throughout a per-game tape.</strong> On a per-game tape (556 matches, 555 of them 2013) consecutive rows differ by a whole game, and the reference said no point was attributable there. Each row in fact reads the side whose game count rose — the game point was theirs — and says nothing about the game's other points; <code>winner</code> reads the same. No field changed shape or value.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.39
+1.13.39: is_qualifying states the one case where a round label DOES decide the draw
+2026-09-22T00:00:00Z
+2026-09-22T00:00:00Z
+
+
+Changed: is_qualifying states the one case where a round label DOES decide the draw.
+<h3>Changed</h3>
+<ul>
+<li><strong><code>is_qualifying</code> states the one case where a round label DOES decide the draw.</strong> The field said it is "never inferred from the round label", which is the right instinct and slightly too strong: main-draw vocabulary (<code>Semi-finals</code>, <code>Final</code>) is indeed never read that way, but a round label that itself reads <code>Qualification Round 1</code> is STATING the draw, and it reads <code>true</code> — a <code>false</code> beside such a round is a payload contradicting itself rather than asserting main draw. Three matches (Chengdu, 22 Sep) were briefly served <code>is_qualifying: false</code> with <code>round_code: "Q1"</code> beside them and were repaired the same morning. Readers segregating a corpus by draw need to know which way that conflict resolves. No field changed shape or type.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.38
+1.13.38: GET /markets states that the scope is MATCH-WINNER ONLY, and returns at most one market
+2026-09-22T00:00:00Z
+2026-09-22T00:00:00Z
+
+
+Changed: GET /markets states that the scope is MATCH-WINNER ONLY, and returns at most one market.
+<h3>Changed</h3>
+<ul>
+<li><strong><code>GET /markets</code> states that the scope is MATCH-WINNER ONLY, and returns at most one market.</strong> A prospect asked whether the odds endpoints carry game spreads / handicaps (<code>A -3.5 @ 1.90</code>) in addition to match-winner prices. Every description in this reference already said "match-winner market", but nowhere said what that EXCLUDES, and the endpoint's own summary said "market(s)". Both are now explicit: no handicaps or game spreads, no totals, no set-winner books, no per-game or per-set derivative; <code>data</code> holds at most one object and <code>meta.count</code> is 0 when nothing is mapped. Venues do list tennis derivatives beside the match-winner book and this API publishes none of them — a derivative is refused at the mapping step. No field changed shape or value.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.37
+1.13.37: is_qualifying is documented on the match object, and 1 more change
+2026-09-21T00:00:00Z
+2026-09-21T00:00:00Z
+
+
+
+Added: is_qualifying is documented on the match object. Changed: round and round_code now state that they are DRAW-RELATIVE.
+<h3>Added</h3>
+<ul>
+<li><strong><code>is_qualifying</code> is documented on the match object.</strong> The field has been served on <code>GET /matches</code>, <code>GET /matches/{matchId}</code> and <code>GET /history/matches</code> for some time and was absent from this reference entirely, which left the one field that separates a qualifying draw from the main draw invisible to anyone reading the docs. Three-valued: <code>true</code>/<code>false</code> are the source's own assertion, <code>null</code> means no source has ever stated it, and <code>null</code> is not <code>false</code>.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong><code>round</code> and <code>round_code</code> now state that they are DRAW-RELATIVE.</strong> Both describe the round *within* the draw the match belongs to, and the draw is named by <code>is_qualifying</code> — not by the round. A qualifying semi-final carries <code>round: "... - Semi-finals"</code> and <code>round_code: SF</code>, exactly as a main-draw semi-final does; <code>Q</code>/<code>Q1</code>..<code>Q4</code> appear only where the label itself names the round as qualifying, which most labels do not. <code>round_code</code> previously said "this is the field to branch on" without that limit, and two customers independently read a qualifying match as a main-draw one and reported it as a data fault. The data was correct in every case; the documentation was not. No field changed shape or value.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.36
+1.13.36: GET /charting/players states a sample of 11,803 charted matches, not 11,646
+2026-09-21T00:00:00Z
+2026-09-21T00:00:00Z
+
+
+Changed: GET /charting/players states a sample of 11,803 charted matches, not 11,646.
+<h3>Changed</h3>
+<ul>
+<li><strong><code>GET /charting/players</code> states a sample of 11,803 charted matches, not 11,646.</strong> The charted corpus was refreshed on 2026-09-19 and 184 matches — 133 of them played after 2026-05-24, including the US Open women's draw to the quarter-finals — were loaded into the product on 2026-09-21. <code>matches_charted</code> on every response has always been the true per-player denominator and was never affected; the curated-coverage note beside it was quoting an older total and understating the sample. <code>GET /rally/matches</code> grew in the same load: 11,822 charted matches and 1,875,132 shot-by-shot points, with the newest women's chart now 2026-09-09. The men's corpus is unchanged because no men's match has been charted since 2026-05-21. No field changed shape.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.35
+1.13.35: meta.points.games_short on GET /history/matches/{matchId} counts the completed games the tape holds too few points for, and 5 more changes
+2026-09-21T00:00:00Z
+2026-09-21T00:00:00Z
+
+
+
+
+Added: meta.points.games_short on GET /history/matches/{matchId} counts the completed games the tape holds too few points for. basis_reason on GET /matches/{matchId}/points says why a reconstruction was served instead of the stream. Plus 4 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong><code>meta.points.games_short</code> on <code>GET /history/matches/{matchId}</code> counts the completed games the tape holds too few points for.</strong> Raised by a prospect evaluating per-point histories. A game the tape skips whole is a legal <code>0-0 → 0-0</code> boundary with the games counter up by one, and the transition test alone cannot see it — so a tape could read <code>complete: true</code> with a whole game missing. The measurement now walks the sequence by game boundary and compares what is held for each completed game with the fewest points that game can have contained given the last state it shows (from 0-0 at least 4; from 15-30 at least 5; from 40-40 at least 8; a tiebreak at least 7) — a lower bound, never an estimate. A game the tape skips whole counts; so does a game that ended with no further rows after its last stored point. <code>> 0</code> forces <code>complete</code> to <code>false</code>; nothing is inferred or repaired, and the game's boundary rows are still served exactly as stored. <code>null</code> = nothing measured (an empty sequence).</li>
+<li><strong><code>basis_reason</code> on <code>GET /matches/{matchId}/points</code> says why a reconstruction was served instead of the stream.</strong> Present on the <code>reconstruction</code> basis only: <code>stream_absent</code> (no stored stream rows), <code>stream_incomplete</code> (the stream is legal but does not measure complete — it joined mid-match or stopped short — and carries no tags) or <code>stream_illegal</code> (at least one transition is not attributable to one point: a gap or a torn row). A stream that is complete, or tagged and legal end to end, is always served on the <code>live</code> basis and the key is absent. Additive; nothing else in the response changes shape.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong><code>meta.points.complete</code> now means the served sequence holds at least every point the score requires — not merely that consecutive rows are consistent with each other.</strong> True only when the sequence opens at 0-0, every transition is a legal single-point step, it carries at least as many point transitions as the final scoreline implies the match contained, every completed game holds at least as many point transitions as it must have contained (<code>games_short</code> is <code>0</code>), it reaches a finished final scoreline at love (or the match ended early — retirement/walkover), it is not known-truncated, and no game was withheld (<code>games_withheld</code>). <code>ends_at_final</code> on the same object now also requires that the last row shows no game in progress — points at love or null, not a tiebreak in progress: a row carrying a finished spine with a game still underway (<code>[[6,6],[3,4]]</code> at 15-0) reads <code>false</code>, where before the spine alone decided. Some tapes that read <code>complete: true</code> until now will read <code>false</code>; no row is changed.</li>
+<li><strong>A recurring score state on the history tape pairs its <code>serve</code>/<code>outcome</code> by nearest clock, instead of reading null.</strong> Tags on <code>GET /history/matches/{matchId}</code> rows are joined from the match's per-point stream by exact score state, and a state the stream asserts more than once (a deuce cycle revisiting deuce) cannot be told apart by the score, so until now it read <code>null</code> outright. Where a state recurs the join now pairs by nearest clock within 90 seconds — the tape row's observed clock against the stream point's, each stream point used at most once — and reads <code>null</code> otherwise (no candidate within the window, or two at the same distance). Every tag is an observed pairing, state and where needed clock, never an inference from the pattern of play. Reconstructed rows and rows no stream point matches still read <code>null</code>.</li>
+</ul>
+<h3>Fixed</h3>
+<ul>
+<li><strong>A stored live stream that is complete, or tagged and legal end to end, is served on a completed match; a reconstruction is projected only when the stream falls short.</strong> Until now a measured-complete recorded sequence displaced the stream unconditionally, so a completed match could lose its <code>serve</code>/<code>outcome</code> tags and every per-point clock the moment a reconstruction landed — a projection carries neither, so a complete tagged stream is strictly more information than any reconstruction of the same match. <code>GET /matches/{matchId}/points</code> now keeps the <code>live</code> basis whenever the stream is itself measured complete (match-closing point included) or carries tags and is legal end to end (every transition one attributable point, judged in playing order), and serves the projection only when the stream has no rows, is incomplete and untagged, or holds a transition nobody can attribute — <code>basis_reason</code> says which. When the reconstruction serves it still serves wholesale; the two sequences are never interleaved. If a completed match reads <code>reconstruction</code>, re-read from <code>after_seq=0</code> rather than resuming a live cursor into it.</li>
+<li><strong>A double fault stated on a first serve is a contradiction; the point is served untagged.</strong> A source point whose serve number and outcome contradict each other — a double fault on serve 1 — was carried through as stated. <code>serve</code> and <code>outcome</code> on such a point now both read <code>null</code> on <code>GET /matches/{matchId}/points</code> and on the WebSocket frames rather than tagged wrongly; nothing is corrected to a second serve. Stored tags that carried the contradiction have been cleared.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.34
+1.13.34: The reference said Davis Cup and the Grand Slams "read null throughout" for the per-point tags; serve is in fact stated on the top Davis Cup tiers
+2026-09-21T00:00:00Z
+2026-09-21T00:00:00Z
+
+
+Changed: The reference said Davis Cup and the Grand Slams "read null throughout" for the per-point tags; serve is in fact stated on the top Davis Cup tiers.
+<h3>Changed</h3>
+<ul>
+<li><strong>The reference said Davis Cup and the Grand Slams "read null throughout" for the per-point tags; <code>serve</code> is in fact stated on the top Davis Cup tiers.</strong> <code>serve</code> and <code>outcome</code> are stated separately, and the old sentence excluded both on the strength of one of them: team ties and Slams carry no per-point outcome, so <code>outcome</code> is genuinely <code>null</code> there — but the serve number is stated on them. Measured over completed matches with a per-point stream since 2026-09-12: Davis Cup World Group states <code>serve</code> on 22 of 22 matches and World Group I on 41 of 42, while World Group II states it on 0 of 38; <code>outcome</code> is null on all of them, on every one of the 14,132 rows. The machine-readable authority was right throughout and is unchanged — a World Group I match answers <code>enrichment: {"serve": "stated", "outcome": "none"}</code> and a World Group II match answers <code>{"serve": "none", "outcome": "none"}</code> — so this corrects the prose to match what the API already returns, in the direction of more coverage, not less. The <code>serve</code> description now names the Davis Cup tiers in its coverage list with the measurement; the <code>outcome</code> description now treats the two fields separately instead of excluding a competition from both. No field, endpoint or behaviour change.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.33
+1.13.33: serve and outcome said nothing about qualifying or doubles draws, and the internal reference called both blanket-null
+2026-09-21T00:00:00Z
+2026-09-21T00:00:00Z
+
+
+Changed: serve and outcome said nothing about qualifying or doubles draws, and the internal reference called both blanket-null.
+<h3>Changed</h3>
+<ul>
+<li><strong><code>serve</code> and <code>outcome</code> said nothing about qualifying or doubles draws, and the internal reference called both blanket-<code>null</code>.</strong> The per-point tags (added 2026-09-12) were documented tour by tour for main draws only. Measured over completed matches carrying a per-point stream since the source went live on 2026-09-12: WTA qualifying states <code>serve</code> on 48 of 48, Challenger (men) qualifying on 153 of 155, Challenger (women) on 49 of 51; ITF qualifying states <code>outcome</code> (aces and double faults) on 95 of 107 men's and 62 of 74 women's. Qualifying draws are covered like their own main draw, with exactly one real gap — <code>outcome</code> on Challenger (men) qualifying, which no source states at all. The doubles half of the old claim is confirmed and now carries its denominator: across 454 completed doubles matches in the same window, not one row carries either field. Both descriptions now state where each field is stated, including the qualifying and doubles positions; <code>enrichment</code> per match remains the authority and none of it is a promise about a match not yet played. Found by the every-sweep tag-coverage check, which reads 85 of 86 Challenger qualifying matches carrying <code>serve</code> against a documented <code>null</code>. No field, endpoint or behaviour change.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.32
+1.13.32: A match that ends in a tiebreak closes with the set roll-up row after the decisive tiebreak score
+2026-09-21T00:00:00Z
+2026-09-21T00:00:00Z
+
+
+Changed: A match that ends in a tiebreak closes with the set roll-up row after the decisive tiebreak score.
+<h3>Changed</h3>
+<ul>
+<li><strong>A match that ends in a tiebreak closes with the set roll-up row after the decisive tiebreak score.</strong> The match-closing row on <code>GET /matches/{matchId}/points</code> was described as if the last point were always a game point one point short of the final, and a match decided in a tiebreak did not fit that shape. The docs now say what the stream actually holds: when a completed match ends in a tiebreak, the stream's last stored row is the decisive tiebreak score itself (7-3, or 8-6) and the closing row is the set roll-up after it — the next game number, <code>number</code> 0, <code>tiebreak</code> false, <code>sets</code> incremented for the tiebreak winner, the set banked 7-6 in <code>games</code>, <code>score</code> 0-0, <code>server</code> null, <code>winner</code> the tiebreak winner — the same row the stream stores after every other set-ending tiebreak, so <code>ends_at_final</code> reads <code>true</code> there. A 10-point match tiebreak still cannot be stated as one point and is refused as before (<code>ends_at_final: false</code>), as is a tape that never reached the final.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.31
+1.13.31: GET /matches/{matchId}/points now closes a completed match with the match-closing point, and ends_at_final says whether it did, and 1 more change
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+Added: GET /matches/{matchId}/points now closes a completed match with the match-closing point, and ends_at_final says whether it did. tagged_at on every point row, and ?changed_since= on the same endpoint, so late serve/outcome tags are collectable without a socket.
+<h3>Added</h3>
+<ul>
+<li><strong><code>GET /matches/{matchId}/points</code> now closes a completed match with the match-closing point, and <code>ends_at_final</code> says whether it did.</strong> Raised by a prospect evaluating per-point tapes, who asked where the point that wins the match is. Every row is the state *after* a point, so a game-winning point is carried by the next game's <code>number: 0</code> opener — and the match-winning point had no next row to be carried by: the live stream never held it, and a serve statistic built off the stream was missing every match's last point. On a completed match served on the <code>live</code> basis the page now ends with one terminal row: <code>seq</code> = last + 1, <code>number</code> 0, <code>sets</code> / <code>games</code> the final score, <code>score</code> 0-0, <code>tiebreak</code> false, <code>server</code> null (nobody serves next), <code>winner</code> the match winner, <code>ts</code> the instant the final score was observed. It is built at read time from the stream's last row and the observed final score, and only when the two are one point apart; nothing is fabricated otherwise. <code>ends_at_final</code> (boolean) on the response says whether the sequence served ends on that row: <code>false</code> on a completed match whose stream stops short of it (a retirement or walkover, a capture that stopped two or more points short, a closer from deuce or from a match tiebreak that cannot be stated as one point), always <code>false</code> on a live match, and on the <code>reconstruction</code> basis judged from the projected sequence's last frame. <code>serve</code> / <code>outcome</code> on the terminal row are null: no source's tag for a match's last point is stored yet. WebSocket and push frames are unchanged.</li>
+<li><strong><code>tagged_at</code> on every point row, and <code>?changed_since=</code> on the same endpoint, so late serve/outcome tags are collectable without a socket.</strong> The same prospect asked how to pick up tags that land after a row was read. <code>after_seq</code> is a cursor by <code>seq</code>, so it can never return a row already held — and a <code>serve</code> / <code>outcome</code> tag that lands late lands on exactly such a row; until now the only way to see it was the <code>point_update</code> frame on a WebSocket. Every row now carries <code>tagged_at</code>, the UTC instant its tags landed (null while none has, and on every <code>reconstruction</code>-basis frame), and <code>?changed_since=<ISO-8601 instant></code> (<code>2026-09-20T00:35:18Z</code>; <code>Z</code> or an offset, a naive value is read as UTC, a date alone is refused) returns only the rows whose <code>ts</code> <strong>or</strong> <code>tagged_at</code> is later than that instant, in <code>seq</code> order, paged as usual and composable with <code>after_seq</code>. The post-match recipe: read the match, keep the instant, re-read with <code>changed_since=<that instant></code> and replace held rows by <code>seq</code>. Anything that is not an ISO-8601 timestamp is a <code>400 bad_changed_since</code>. On the <code>reconstruction</code> basis no row carries a clock or a tag, so a <code>changed_since</code> read of it is an empty page: that sequence is final at first read.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.30
+1.13.30: The ATP per-point outcome row said "full" as a projection, not a measurement
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+Changed: The ATP per-point outcome row said "full" as a projection, not a measurement.
+<h3>Changed</h3>
+<ul>
+<li><strong>The ATP per-point outcome row said "full" as a projection, not a measurement.</strong> In the 30 days since ATP per-point outcomes went live (2026-09-12), no ATP main-tour event was actually played — the only tour tennis in that window was a Grand Slam and Davis Cup, neither of which carries per-point outcomes — so every ATP main-tour match carried no outcome tags at all, while the docs asserted <code>full</code> (ace, double fault, winner, forced error, unforced error) unconditionally. The row now names the tours covered (ATP 250 / 500 / 1000), states plainly that Grand Slams and Davis Cup are not on it and read null throughout (check <code>enrichment</code> per match), and gives the first date the claim was actually measured against a played tour event: the week of 2026-09-22.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.29
+1.13.29: The GET /history/matches listing's point-completeness ledger now judges the same raw sequence the per-match read serves, re-sent rows dropped — not the clean…, and 1 more change
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+
+Changed: The GET /history/matches listing's point-completeness ledger now judges the same raw sequence the per-match read serves, re-sent rows dropped — not the clean collapse. Fixed: tape.computed_at on the GET /history/matches listing carried the serving host's UTC offset (+03:00 on prod) instead of a trailing Z.
+<h3>Changed</h3>
+<ul>
+<li><strong>The <code>GET /history/matches</code> listing's point-completeness ledger now judges the same raw sequence the per-match read serves, re-sent rows dropped — not the <code>clean</code> collapse.</strong> The ledger's previous default basis, <code>clean</code>, collapses the tape to one row per distinct score state; that collapse can delete a real deuce point — two rows sitting at 40-40 with no advantage row between them is a legitimate rally, not a duplicate — and call the resulting, shorter tape point-complete when the tape actually served is not. The ledger now judges the raw sequence with provable re-sends dropped, the exact judgement the per-match read makes and publishes as <code>meta.points.complete</code>, so <code>tape.points_complete_default</code> on the listing and <code>meta.points.complete</code> on a fetched match agree by construction instead of disagreeing on a sequence the read never served. Measured on 7,717 completed matches scored since 1 Sep: the <code>clean</code> collapse called 70 of them point-complete whose served (raw) tape was not, and missed 810 whose served tape was. <code>clean</code> remains available only as a request option (<code>?sequence=clean</code>); rows scored before 2026-09-20 keep <code>points_basis: clean</code> until the nightly re-score reaches them. <code>tape.points_basis</code>, <code>tape.points_complete_default</code> and <code>tape.points_complete</code> descriptions updated to match.</li>
+</ul>
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>tape.computed_at</code> on the <code>GET /history/matches</code> listing carried the serving host's UTC offset (<code>+03:00</code> on prod) instead of a trailing <code>Z</code>.</strong> Same instant, valid ISO-8601, but every other timestamp in the payload — <code>meta.generated_at</code>, each row's <code>timestamp</code> — ends in <code>Z</code>, so a consumer diffing <code>computed_at</code> against either had to normalise it by hand first. Now projected to UTC with a trailing <code>Z</code> like the rest of the surface.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.28
+1.13.28: meta.points.games_withheld on GET /history/matches/{matchId}, and 3 more changes
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+
+Added: meta.points.games_withheld on GET /history/matches/{matchId}. tape.points_complete_default, tape.points_complete_recon and tape.points_basis on the GET /history/matches listing. Plus 2 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong><code>meta.points.games_withheld</code> on <code>GET /history/matches/{matchId}</code>.</strong> Present whenever reconstructed rows are in the response: the count of games whose recorded points were internally inconsistent — a recorded score state that repeats or moves backwards partway through the game — and whose in-game points are therefore withheld rather than published. The game's opening <code>0-0</code> row is still served, so the games spine stays intact; nothing is repaired, reordered or inferred — a game is served exactly as recorded, or not at all. <code>> 0</code> forces <code>meta.points.complete</code> to <code>false</code>. <code>null</code> means not measured: either the reconstruction predates this check, or it was built by merging more than one point record rather than from a single one.</li>
+<li><strong><code>tape.points_complete_default</code>, <code>tape.points_complete_recon</code> and <code>tape.points_basis</code> on the <code>GET /history/matches</code> listing.</strong> <code>points_complete</code> on that listing is a best-basis OR: the ledger's default basis is the <code>clean</code> collapse of an observed tape, while the per-match read serves and measures the <code>raw</code> sequence, so <code>points_complete: true</code> can sit next to <code>meta.points.complete: false</code> on the same match with both honest about their own sequence. These three fields say which measurement produced the <code>true</code> and on which sequence — <code>points_complete_recon: true</code> means "fetch it with <code>?points=complete</code>". All null when not yet measured.</li>
+<li><strong><code>meta.points.resent_rows</code> on <code>GET /history/matches/{matchId}</code>.</strong> A row identical to the row before it is the same state re-sent by a source on its timer, which the raw tape carries by design — no point separates the two rows, so they no longer count as a non-point transition. <code>transitions_total</code> is now <code>rows − 1 − resent_rows</code>. <code>resent_rows</code> is <code>0</code> on a <code>clean</code> or <code>recon</code> basis: the collapse already removes re-sends, and in a reconstruction one row is one point, so a repeat there is a tear in the recorded sequence and is never dropped.</li>
+</ul>
+<h3>Fixed</h3>
+<ul>
+<li><strong>A finished-match point record that moved backwards inside a game was being reconstructed as if it were legal.</strong> Since July 2026 some recorded point sequences have not been monotone within a game (a real example: 40-30, 15-15, 15-30, 30-30, 15-30, 30-30, 30-30, 40-30 — 547 of 586 August ATP/WTA matches carried at least one such game). The tape builder kept the longest forward-moving chain through the inconsistency and published the result as an ordinary reconstructed row, so a state like 30-30 repeating three times in a row was served as fact. Those games are now withheld whole instead — counted in the new <code>games_withheld</code> field — rather than fitted into a plausible-looking sequence.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.27
+1.13.27: GET /history/matches/{matchId} tape rows now carry origin, and observed rows carry serve / outcome, and 2 more changes
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+
+Added: GET /history/matches/{matchId} tape rows now carry origin, and observed rows carry serve / outcome. meta.reconstructed_at, meta.observed_span and meta.enrichment on the same endpoint. reconstructed_at is the write time (UTC) of the newest reconstructed row actually served, null when none is; observed_span is {"first", "last"} — the timestamp of the first and last observed rows served, in served order, null when none is; enrichment is {"serve": "stated" | "none", "outcome": "full" | "ace_double_fault" | "none"}, the vocabulary that actually landed on this response's rows (none/none on a wholly reconstructed tape). Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><strong><code>GET /history/matches/{matchId}</code> tape rows now carry <code>origin</code>, and observed rows carry <code>serve</code> / <code>outcome</code>.</strong> Raised by a prospect evaluating reconstructed tapes as evidence for an independent model, who asked how to tell observed from reconstructed rows explicitly, when a reconstruction landed, and whether the per-point serve/outcome tags reach the history tape at all. <code>origin</code> (<code>observed</code> | <code>reconstructed</code>) is read from the row's stored provenance — the same fact that nulls the clock on a reconstructed row, never the null clock itself — so a caller no longer has to treat <code>timestamp: null</code> as an inferred marker. <code>serve</code> (<code>1</code> | <code>2</code> | null) and <code>outcome</code> (<code>ace</code> | <code>double_fault</code> | <code>winner</code> | <code>forced_error</code> | <code>unforced_error</code> | null) are joined onto <strong>observed rows only</strong> from the per-point stream by EXACT score state: set count, every set's games, in-game points and the tiebreak flag all equal. A tape row stores no point ordinal, so a score state the stream asserts more than once (a deuce cycle revisiting deuce) is ambiguous and reads null, as does a row no stream point matches and every <code>reconstructed</code> row — nothing is inferred from the score. All additive on <code>GET /history/matches/{matchId}</code>; the pre-2023 <code>ArchiveTape</code> (every row reconstructed) does not carry <code>origin</code>, <code>serve</code> or <code>outcome</code>.</li>
+<li><strong><code>meta.reconstructed_at</code>, <code>meta.observed_span</code> and <code>meta.enrichment</code></strong> on the same endpoint. <code>reconstructed_at</code> is the write time (UTC) of the newest reconstructed row actually served, null when none is; <code>observed_span</code> is <code>{"first", "last"}</code> — the <code>timestamp</code> of the first and last observed rows served, in served order, null when none is; <code>enrichment</code> is <code>{"serve": "stated" | "none", "outcome": "full" | "ace_double_fault" | "none"}</code>, the vocabulary that actually landed on this response's rows (<code>none</code>/<code>none</code> on a wholly reconstructed tape). All three are measured on the rows returned, after any <code>sequence=clean</code> collapse.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong><code>meta.points.available_complete</code> is now read-through instead of nightly-only.</strong> It still answers from the nightly ledger when the ledger has an entry. When the ledger has none yet — a reconstruction that landed after the last nightly run — and this read served a reconstruction, it now answers from the live verdict just measured on that reconstruction instead of <code>null</code>. It reads <code>null</code> only when there is no ledger entry and nothing was reconstructed on this read to measure. No field or endpoint added; behaviour change on an existing field.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.26
+1.13.26: Team-competition matches now carry the surface of their tie, and 1 more change
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+
+Fixed: Team-competition matches now carry the surface of their tie. Changed: Match.surface and Tournament.surface descriptions now state the per-tie rule for team competitions instead of describing the null as permanent.
+<h3>Fixed</h3>
+<ul>
+<li><strong>Team-competition matches now carry the surface of their tie.</strong> Davis Cup, Billie Jean King Cup and Laver Cup matches published <code>surface: null</code> on the match and on <code>tournament.surface</code> — 268 of 268 team-event matches in the 60 days to 2026-09-20. 1.13.25 documented that null as structural; it was only structural for the data we held. Every surface we held was per tournament, and a team "tournament" (<code>ATP Davis Cup - World Group I</code>) is a tier with a draw stage, not a surface; a tie is played on whatever its host nation chose. Each tie's published record states the surface, venue and indoor/outdoor flag, so the API now holds one row per tie and resolves each match to its tie by competition, date window and the nations on court — answering only when a single tie remains, never from a nation, a name or a previous tie. Existing team-event matches (last 60 days and upcoming) were backfilled with one audit row per write; new ties are picked up as they are drawn. Ties whose venue the ITF has not announced stay null. <code>tournament.surface</code> for a team competition remains null by design: the surface belongs to the tie, and it is published on the match. Raised by a public report that Davis Cup match 191867 returned <code>surface: null</code>; that match now serves <code>hard</code>. Payload shape unchanged.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong><code>Match.surface</code> and <code>Tournament.surface</code> descriptions</strong> now state the per-tie rule for team competitions instead of describing the null as permanent.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.25
+1.13.25: A match that had been given a tournament_id was never given the surface that id serves, and 1 more change
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+
+Fixed: A match that had been given a tournament_id was never given the surface that id serves. Changed: Match.surface and Tournament.surface now say when they are null and why.
+<h3>Fixed</h3>
+<ul>
+<li><strong>A match that had been given a <code>tournament_id</code> was never given the surface that id serves.</strong> <code>matches.tournament_key</code> was backfilled on 2026-09-04 for matches that had been publishing <code>tournament_id: null</code>; the pass gave those rows an id and stopped there. The tournament-surface fallback runs only in the ingest upsert, and every repaired row was already terminal, so it never ran again: 2,533 matches (2026-06-21..2026-09-04, all ITF) published <code>surface: null</code> while <code>GET /tournaments/{id}</code> — for the id that same row hands you — answered <code>"surface": "hard"</code>. Two surfaces for one question, each written by a different writer. Measured and wire-confirmed before the repair on match <code>186637</code> (<code>surface: null</code>, <code>tournament_id: "12158"</code>) against <code>/tournaments/12158</code> (<code>"surface": "hard"</code>), and after it on the same row. All 2,533 now carry the surface their own tournament states; <code>indoor</code> was filled from the same tournament row where the match held none. Nothing after 2026-09-04 was affected — the same commit's ingest hook sets the key before the fallback reads it, so the forward path has been correct throughout and the weekly count is flat zero since. The repair resolves surface through the same chain the ingest uses (tournament-name map first, tournament row second), so a later upsert cannot answer one of these rows differently. Match-level <code>surface</code> null over the previous 90 days: 19.8% before, 11.8% after. No field or endpoint added.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong><code>Match.surface</code> and <code>Tournament.surface</code> now say when they are null and why.</strong> Both were documented as nullable with no account of the population, while the <code>tournament_id</code> field beside them carries a measured one. The remaining match-level nulls are two structural groups and an 8-row residue: UTR events, which are not in our tournament list at all (and so publish <code>tournament_id: null</code> too), and TEAM COMPETITIONS — Davis Cup, Billie Jean King Cup, Laver Cup — where the surface is the host nation's choice per tie, so neither the season-long tournament row nor the match has one to state. Raised by a public report that Davis Cup matches return <code>surface: null</code>. For those 35 tournament rows a value outside hard/clay/grass is rejected rather than published. No behaviour change.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.24
+1.13.24: GET /matches/{matchId}/score no longer answers 404 for a settled match whose live-score rows were retired by the 90-day retention sweep, or that only ever had…
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+Changed: GET /matches/{matchId}/score no longer answers 404 for a settled match whose live-score rows were retired by the 90-day retention sweep, or that only ever had an archived final.
+<h3>Changed</h3>
+<ul>
+<li><code>GET /matches/{matchId}/score</code> no longer answers 404 for a settled match whose live-score rows were retired by the 90-day retention sweep, or that only ever had an archived final. When there is no live row at all, a match with a non-null <code>outcome</code> (completed, retired, walkover, default, abandoned, unresolved) is answered from its archived final — the same read <code>GET /matches?status=completed</code> already embeds, through the same serializer — so the listing and the single-match read can no longer disagree about whether a score exists. Unchanged: a live match always reads its live tape (the archive never outranks it); an upcoming match with no row, a cancelled match that was never played, and a settled match with nothing recorded anywhere still answer 404. An archived final carries <code>age_seconds</code>, <code>observed_age_seconds</code>, <code>sources_count</code> and <code>accepted_at</code> as null — no clock is claimed for a state nobody watched — and <code>timestamp</code> null where the archived row has none; the Score object still carries no data-source label, <code>GET /history/matches/{matchId}</code> <code>meta.point_source</code> says whether the final was observed or reconstructed. Behaviour change on the API side 2026-09-20; measured at the change: 13,437 completed matches from the previous 180 days, 157,170 all time, now answer <code>/score</code> instead of 404. No field or endpoint added.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.23
+1.13.23: GET /rankings accepted ?surface=, ?min_matches=, ?activity_weeks= and a contradicting ?tour= on the official systems, and silently ignored all four
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+Fixed: GET /rankings accepted ?surface=, ?min_matches=, ?activity_weeks= and a contradicting ?tour= on the official systems, and silently ignored all four.
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>GET /rankings</code> accepted <code>?surface=</code>, <code>?min_matches=</code>, <code>?activity_weeks=</code> and a contradicting <code>?tour=</code> on the official systems, and silently ignored all four.</strong> Measured on production before the fix: <code>?system=atp&surface=clay</code>, <code>?system=atp&min_matches=200</code>, <code>?system=atp&activity_weeks=4</code> and <code>?system=atp&tour=wta</code> each returned <code>200</code> and the plain overall ATP table, byte-identical to the same call with no parameter at all; <code>?system=atp&tour=zzz</code> was accepted too. All four are documented "Elo only" / "Elo listing" and are read nowhere else, so a caller asking for a clay ATP leaderboard received the overall one with nothing in the response to say the filter had not run. The <code>elo</code> branch was correct throughout and is the control that makes this a defect rather than an empty population - <code>?system=elo&tour=atp&surface=clay</code> returns a genuinely different top 10 from <code>surface=hard</code>, and <code>surface=zzz</code> is refused with the valid list. Unlike the <code>has_market</code> hole fixed in 1.13.21, this one cannot be closed by making the filter act: there is no clay ATP ranking and no WTA row in an ATP table. <code>surface</code>, <code>min_matches</code> and <code>activity_weeks</code> are now a <code>400</code> when no requested system is <code>elo</code> (a mixed <code>?system=atp&system=elo&surface=clay</code> still passes); <code>tour</code> is a <code>400</code> only when it contradicts a system the caller NAMED, or when that system is not an ATP/WTA walk. <code>?system=atp&tour=atp</code> is redundant rather than wrong and still answers, and the implicit default system set is unchanged.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.22
+1.13.22: GET /history/archive/players now takes ?id= (alias ?player_id=), so the corpus person id we publish can finally be looked up, and 2 more changes
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+
+Added: GET /history/archive/players now takes ?id= (alias ?player_id=), so the corpus person id we publish can finally be looked up. Fixed: A corpus person id sent to GET /players/{playerId} returned a bare 404 that explained nothing. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><strong><code>GET /history/archive/players</code> now takes <code>?id=</code> (alias <code>?player_id=</code>), so the corpus person id we publish can finally be looked up.</strong> Archive match rows have published that id as <code>winner.player_id</code> / <code>loser.player_id</code> since 2026-08-03, and this endpoint has published it as <code>id</code> and described it in those words — but the only surface accepting the value was <code>/rankings?archive_player=</code> — ULTRA, <code>system=elo</code> only, and it answers with RATINGS, not the person. So the PERSON could not be reached by id at all: a reader holding a corpus id could only go back to the NAME. It is a filter on the list rather than a <code>/history/archive/players/{id}</code> detail route on purpose: the natural key is <code>(tour, source_pid)</code>, the two tours number independently, and 14,627 corpus person ids are live in BOTH (measured 2026-09-20) — a detail route would have to pick a tour and would return the wrong human for those ids without saying so. The list returns every person wearing the id; add <code>tour</code> to narrow it to exactly one. Sending <code>id</code> and <code>player_id</code> with different values is a <code>400</code>.</li>
+</ul>
+<h3>Fixed</h3>
+<ul>
+<li><strong>A corpus person id sent to <code>GET /players/{playerId}</code> returned a bare <code>404</code> that explained nothing.</strong> The roster (<code>/players</code>, ids 1-37033) and the results-archive person registry (100001-270580) are separate, deliberately disjoint id spaces - 0 of 137,483 corpus ids resolve on <code>/players/{playerId}</code> - and the only id-shaped route with <code>player</code> in its name was the roster one. The <code>404</code> now carries <code>detail</code> and a <code>see</code> pointer at <code>/history/archive/players?id=...</code> when the id resolves in the archive registry; an id from neither space keeps the plain body. The same signpost is on <code>/players/{playerId}/stoppages</code> and <code>/players/{playerId}/injuries</code>. The status is deliberately still <code>404</code> and not <code>410</code>: <code>410</code> asserts the id once existed HERE, and it never did. The person is never named in the <code>404</code> - the archive is History-gated and <code>/players/{playerId}</code> is not.</li>
+<li><strong><code>GET /players/{playerId}</code> was missing its <code>410</code> in this specification.</strong> The route has answered <code>410 PlayerMerged</code> with a forwarding address for a merged or retired player id for as long as its two stoppage aliases have, and both aliases documented it while the main detail endpoint did not. Now documented, along with a <code>PlayerNotFound</code> schema that states exactly when <code>detail</code> and <code>see</code> are present.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.21
+1.13.21: GET /matches accepted ?has_market= and silently ignored it, and status=cancelled ignored ?has_analysis= as well, and 1 more change
+2026-09-20T00:00:00Z
+2026-09-20T00:00:00Z
+
+
+
+Fixed: GET /matches accepted ?has_market= and silently ignored it, and status=cancelled ignored ?has_analysis= as well. Added: has_analysis and has_market are now DOCUMENTED as query parameters on GET /matches. has_analysis has worked on live, upcoming and completed since 2026-09-14 and has_market works everywhere from today, but neither appeared in this specification, so the only filter a reader could find was has_market on /history/matches.
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>GET /matches</code> accepted <code>?has_market=</code> and silently ignored it, and <code>status=cancelled</code> ignored <code>?has_analysis=</code> as well.</strong> The filter was implemented in the shared query builder on 2026-09-15 and exposed on <code>/history/matches</code> the same day, but it was never wired to <code>/matches</code>: every row published <code>has_market</code>, the query string was accepted, and the unfiltered page came back under a <code>200</code>. Measured on production before the fix — <code>status=live</code> returned the same 14 rows (2 with a market) for <code>has_market=true</code>, <code>has_market=false</code> and no filter at all; <code>status=completed</code> the same 100 rows (38 with a market); <code>/history/matches</code> was correct throughout (40/40 and 0/40). Five of the eight status x filter combinations returned a wrong answer under a success code. Both filters now apply on every status, and an unparseable value is a <code>400 bad_has_market</code> / <code>bad_has_analysis</code> as documented.</li>
+</ul>
+<h3>Added</h3>
+<ul>
+<li><code>has_analysis</code> and <code>has_market</code> are now DOCUMENTED as query parameters on <code>GET /matches</code>. <code>has_analysis</code> has worked on <code>live</code>, <code>upcoming</code> and <code>completed</code> since 2026-09-14 and <code>has_market</code> works everywhere from today, but neither appeared in this specification, so the only filter a reader could find was <code>has_market</code> on <code>/history/matches</code>. The prose telling callers to "filter the slate first" was therefore advice with no documented instrument behind it.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.20
+1.13.20: The two win-probability regime boundaries published in 1.13.19 were three hours late, and are corrected here: 2026-08-18T08:29:28Z and 2026-09-16T07:59:54Z, and 2 more changes
+2026-09-19T00:00:00Z
+2026-09-19T00:00:00Z
+
+
+Fixed: The two win-probability regime boundaries published in 1.13.19 were three hours late, and are corrected here: 2026-08-18T08:29:28Z and 2026-09-16T07:59:54Z. With it, the measured lag of the version string: the +itfdraw-2026-09-16 suffix first appears on rows generated 2026-09-17T05:04:25Z, which is 21 h 04 m after the rule landed, not 18 h 02 m. Plus 1 more change.
+<h3>Fixed</h3>
+<ul>
+<li><strong>The two win-probability regime boundaries published in 1.13.19 were three hours late, and are corrected here: 2026-08-18T08:29:28Z and 2026-09-16T07:59:54Z.</strong> Both switch instants were read off <code>config.updated_at</code>, a column this application stores in <strong>naive local wall time (UTC+3)</strong>, and were published as if they were UTC. Verified six independent ways on 2026-09-19: six config rows whose own value is an epoch each read exactly +3 h against their <code>updated_at</code> (e.g. <code>monitoring.worker_heartbeat_ts</code> = 2026-09-19T20:22:46.465Z stored as <code>23:22:46.467</code>). The corrected instants are the config switches themselves — <code>win_probability.match_tiebreak_detection_enabled</code> at 2026-08-18 08:29:28.93Z and <code>win_probability.itf_pricing_draw_rule_enabled</code> at 2026-09-16 07:59:54.76Z — which is the earliest instant from which a state can carry the new rule, and therefore the safe place to cut.</li>
+<li>With it, the measured lag of the version string: the <code>+itfdraw-2026-09-16</code> suffix first appears on rows generated 2026-09-17T05:04:25Z, which is <strong>21 h 04 m</strong> after the rule landed, not 18 h 02 m. That figure had been derived from the same three-hour-late instant. <code>2026-09-17T05:04:25Z</code> and <code>2026-09-12T12:08:17Z</code> were both taken from <code>win_probability_meta.generated_at</code>, a timezone-aware column, and are unchanged and correct.</li>
+<li>Anyone who segregated a corpus on 2026-08-18T11:29:28Z or 2026-09-16T11:02:00Z should re-cut: the three hours before each corrected instant are on the wrong side of the split.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.19
+1.13.19: Score.win_probability_meta: the two win-probability regime boundaries are published here for the first time, so a customer can segregate a recorded corpus…, and 3 more changes
+2026-09-19T00:00:00Z
+2026-09-19T00:00:00Z
+
+
+Changed: Score.win_probability_meta: the two win-probability regime boundaries are published here for the first time, so a customer can segregate a recorded corpus without asking us. The same description states, newly measured, that model_version is not a safe discriminator across the second boundary: the +itfdraw-2026-09-16 suffix first appears on rows generated 2026-09-17T05:04:25Z, 18 h 02 m after the rule landed (corrected to 21 h 04 m in 1.13.20). Plus 2 more changes.
+<h3>Changed</h3>
+<ul>
+<li><code>Score.win_probability_meta</code>: the two win-probability <strong>regime boundaries</strong> are published here for the first time, so a customer can segregate a recorded corpus without asking us. Two changes moved published live probabilities before <code>model_version</code> reflected them — 2026-08-18T11:29:28Z, when the over-inclusive deciding-set rule for lower-tier singles began, and 2026-09-16T11:02:00Z, when the draw-based rule replaced it (BOTH INSTANTS WERE THREE HOURS LATE AND ARE CORRECTED IN 1.13.20 — use 08:29:28Z and 07:59:54Z) (up to 0.20 on affected deciding-set states) — so the cut is on <code>generated_at</code>/<code>timestamp</code>, never on the version string. Published in the application's internal reference on 2026-09-17 and not ported here until now; docs.livetennisapi.com, which is the artifact a customer opens, carried none of it.</li>
+<li>The same description states, newly measured, that <code>model_version</code> is <strong>not</strong> a safe discriminator across the second boundary: the <code>+itfdraw-2026-09-16</code> suffix first appears on rows generated 2026-09-17T05:04:25Z, 18 h 02 m after the rule landed (corrected to 21 h 04 m in 1.13.20). Measured against production on 2026-09-19: in that gap 33,510 states across 351 matches — 9,719 states over 113 matches at the ITF M15/W15/W35 levels the rule governs — were computed under the corrected rule while still carrying <code>markov-population-2026-08-23</code>. A cut on a whole-day boundary mislabels exactly those rows; a cut at the 11:02:00Z instant does not. The previous internal wording ("timestamp only up to 2026-09-17") was right to the day and vague by eighteen hours.</li>
+<li><code>Score.win_probability_meta.model_version</code> now carries a description of its own, and the pre-stamp caveat is stated with its instant: no row generated before 2026-09-12T12:08:17Z carries <code>model_version</code> or <code>generated_at</code> at all.</li>
+<li><code>HistoryTapeRow.win_probability_p1</code>: documented that the tape carries no model-regime stamp — no <code>model_version</code> or <code>generated_at</code> column here or in the bulk <code>tape</code> packages — so a recorded tape corpus is segregated by the row's own <code>timestamp</code> against those two instants.</li>
+</ul>
+<p>Documentation only. No field, endpoint or wire value changed.</p>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.18
+1.13.18: LivePoint.number (GET /matches/{matchId}/points, the point frame, the webhook point event): the field is documented as null when we joined the game already in…, and 1 more change
+2026-09-19T00:00:00Z
+2026-09-19T00:00:00Z
+
+
+
+Fixed: LivePoint.number (GET /matches/{matchId}/points, the point frame, the webhook point event): the field is documented as null when we joined the game already in progress, and the API now writes null there instead of 0. Changed: Documented, in all three places the replay guidance appears, that (set, game, number) orders points and does not identify them: the tuple may repeat, and number may be null.
+<h3>Fixed</h3>
+<ul>
+<li><code>LivePoint.number</code> (<code>GET /matches/{matchId}/points</code>, the <code>point</code> frame, the webhook <code>point</code> event): the field is documented as <strong>null when we joined the game already in progress</strong>, and the API now writes null there instead of <code>0</code>. <code>0</code> asserts a game's opening state, and on a game we picked up mid-way that is false — how many points it already holds is not derivable from the score, because the deuce zone maps many ordinals onto one score, and it is never guessed. Measured on 2026-09-19, 1,341 rows over three days carried <code>number: 0</code> at a non-love score. A second defect in the same computation is fixed with it: a late-arriving row for a *different* game committed between two points of the game in progress reset the ordinal, so on match 191801 <code>seq</code> 65 and 67 were both <code>(set 2, game 1, number 0)</code> — at 15-15 and 15-40 — with <code>seq</code> 66, a point of set 1 game 9, between them. The ordinal now continues within its own game regardless of what arrived in between. The schema already permitted null; only the description and the written values change.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li>Documented, in all three places the replay guidance appears, that <code>(set, game, number)</code> <strong>orders</strong> points and does not <strong>identify</strong> them: the tuple may repeat, and <code>number</code> may be null. A repeat on the live basis is a re-statement of a game by a second source, not a correction — there is no revision id, superseded-<code>seq</code>, version or correction flag, because rows are append-only and never rewritten. <code>after_seq</code> therefore never needs a full refetch, and the page-level <code>quality</code> field already reads <code>revised</code> when a page contains such a re-statement. <code>seq</code> remains the only unique, stable per-row key.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.17
+1.13.17: Native WebSocket close codes, documented for the first time, and 1 more change
+2026-09-18T00:00:00Z
+2026-09-18T00:00:00Z
+
+
+
+Added: Native WebSocket close codes, documented for the first time. Changed: GET /matches/{matchId}/points, the point frame and the points signal: seq is arrival order, not match order, on the live basis.
+<h3>Added</h3>
+<ul>
+<li>Native WebSocket <strong>close codes</strong>, documented for the first time. Every refusal sends its <code>error</code> frame and then closes with a code that says what to do next — <code>1013</code> Try Again Later for transient refusals (<code>connection_limit:per_key</code>, <code>connection_limit:server</code>, <code>service_unavailable</code>), <code>1008</code> Policy Violation for anything a retry cannot fix (<code>unauthorized</code>, <code>upgrade_required</code>, <code>email_unverified</code>, <code>client_blocked</code>, <code>bad_json</code>, <code>no_topics</code>, and any mid-stream loss of access), <code>1012</code> on a restart, <code>1000</code> on a normal close. The close reason repeats the frame's <code>error</code> string, so <code>(code, reason)</code> is a complete diagnosis even if the frame was missed. Behaviour change the same day: refusals raised during the *handshake* previously closed <code>1000</code> with an empty reason, indistinguishable from an orderly shutdown, so a client awaiting its <code>subscribed</code> ack saw only a normal close. The error frame was, and still is, delivered before the close; only the close code and reason changed.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><code>GET /matches/{matchId}/points</code>, the <code>point</code> frame and the <code>points</code> signal: <strong><code>seq</code> is arrival order, not match order, on the live basis.</strong> It is assigned in commit order, and points of a live match arrive by more than one path at different speeds, so a point from a set that has just ended can be committed after points from the set that follows it and carry the higher <code>seq</code>. Each row is self-consistent; reading the tape in <code>seq</code> order can show the set or game counter step backwards. Measured over a recent seven-day window this affected a minority of live matches, and never the <code>reconstruction</code> basis. <code>seq</code> remains the right key for paging, dedup and resume, and is the wrong key for chronology — sort by <code>(set, game, number)</code> to replay in playing order. The spec previously said "ordered per match by <code>seq</code>", which reads as chronological. Documentation only; nothing on the wire changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.16
+1.13.16: GET /usage → today.remaining_day: documented as the day allowance less SERVED calls (calls - errors), and the API now computes it that way
+2026-09-18T00:00:00Z
+2026-09-18T00:00:00Z
+
+
+Changed: GET /usage → today.remaining_day: documented as the day allowance less SERVED calls (calls - errors), and the API now computes it that way.
+<h3>Changed</h3>
+<ul>
+<li><code>GET /usage</code> → <code>today.remaining_day</code>: documented as the day allowance less <strong>SERVED</strong> calls (<code>calls - errors</code>), and the API now computes it that way. Refused requests have never spent the daily allowance — the <code>429</code> guidance in this spec already said so — but the endpoint subtracted gross calls, so it under-reported what a key could still send by exactly its error count. Measured on 2026-09-18, one ULTRA key was shown 14,360 fewer requests remaining than the API would have served it. <code>today.calls</code> and <code>today.errors</code> are unchanged: raw counters of everything the key sent.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.15
+1.13.15: event_status (match object): the enum omitted Finished and Unresolved, and 2 more changes
+2026-09-18T00:00:00Z
+2026-09-18T00:00:00Z
+
+
+
+Fixed: event_status (match object): the enum omitted Finished and Unresolved. Changed: event_status description: "NULL means the match completed normally" was wrong in the direction that matters — null means nothing was ever stated for the match, which covers matches that ran their course and matches nothing was ever said about alike. Plus 1 more change.
+<h3>Fixed</h3>
+<ul>
+<li><code>event_status</code> (match object): the enum omitted <strong><code>Finished</code></strong> and <strong><code>Unresolved</code></strong>. <code>Finished</code> is the value the field carries on a normally-completed match and is by far its most common — 144,266 of the 150,678 rows carrying one, measured on 2026-09-18 — so a client generated from this spec with strict enum validation rejected the majority of completed matches. Both values are listed now. The API has always published them; only the spec was wrong, so nothing on the wire changed.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><code>event_status</code> description: "NULL means the match completed normally" was wrong in the direction that matters — null means nothing was ever stated for the match, which covers matches that ran their course and matches nothing was ever said about alike. Branch on <code>outcome</code>, not on the absence of a badge.</li>
+<li><code>event_status</code> description: <code>event_status: Finished</code> while <code>status</code> is still <code>live</code> is documented as the pending-final state (one source has called the match over, the final is not confirmed). The score stands still through it — <code>stale</code> true, <code>age_seconds</code> climbing. Measured over the seven days to 2026-09-18, across 649 matches, that gap closed in 111 s at the median and 911 s at the ninetieth percentile; 85 ran past ten minutes.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.14
+1.13.14: Plan end: when a paid plan's period ends (or a renewal goes unpaid) the same key now drops to the FREE tier automatically instead of being switched off —…
+2026-09-17T00:00:00Z
+2026-09-17T00:00:00Z
+
+
+Changed: Plan end: when a paid plan's period ends (or a renewal goes unpaid) the same key now drops to the FREE tier automatically instead of being switched off — nothing is re-issued, free endpoints keep answering at the free limits, paid endpoints answer 403 upgrade_required from that moment, and subscribing again lifts the same key.
+<h3>Changed</h3>
+<ul>
+<li>Plan end: when a paid plan's period ends (or a renewal goes unpaid) the same key now drops to the FREE tier automatically instead of being switched off — nothing is re-issued, free endpoints keep answering at the free limits, paid endpoints answer <code>403 upgrade_required</code> from that moment, and subscribing again lifts the same key. Behaviour change on the billing side effective 2026-09-17 19:59 UTC; keys of plans that ended in the previous 30 days were moved to FREE the same evening. No field or endpoint changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.13
+1.13.13: GET /history/matches/{matchId} tiebreaks / meta.tiebreaks_source: the timing was stated wrongly
+2026-09-17T00:00:00Z
+2026-09-17T00:00:00Z
+
+
+Changed: GET /history/matches/{matchId} tiebreaks / meta.tiebreaks_source: the timing was stated wrongly.
+<h3>Changed</h3>
+<ul>
+<li><code>GET /history/matches/{matchId}</code> <code>tiebreaks</code> / <code>meta.tiebreaks_source</code>: the timing was stated wrongly. The <code>summary</code> and <code>reconstruction</code> kinds were described as "recorded at completion"; they are written by a finals pass that runs once a day, so a match that finished earlier the same day commonly reads null for its 7-6 sets and carries them from the next pass onward. Only the <code>tape</code> kind is available the instant a match ends. Measured 2026-09-17: 72-100% of 7-6 sets populated on matches completed over the five previous days, 8% on matches completed the same day. Documentation only — no behaviour or field changed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.12
+1.13.12: GET /history/matches/{matchId} tiebreaks: per-set breaker finals are now filled from the point-by-point reconstruction and from the sources' set summaries when…
+2026-09-16T00:00:00Z
+2026-09-16T00:00:00Z
+
+
+Changed: GET /history/matches/{matchId} tiebreaks: per-set breaker finals are now filled from the point-by-point reconstruction and from the sources' set summaries when the live tape stopped one point short (measured before this change: null on about 99% of 7-6 sets — 297 measured, 3 populated).
+<h3>Changed</h3>
+<ul>
+<li><code>GET /history/matches/{matchId}</code> <code>tiebreaks</code>: per-set breaker finals are now filled from the point-by-point reconstruction and from the sources' set summaries when the live tape stopped one point short (measured before this change: null on about 99% of 7-6 sets — 297 measured, 3 populated). New <code>meta.tiebreaks_source</code> says, per set, whether the final came from <code>tape</code>, <code>summary</code> or <code>reconstruction</code>. The response shape is unchanged.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.11
+1.13.11: GET /players/{playerId}/stoppages (PRO) and its alias GET /players/{playerId}/injuries: one player's in-match stoppages (medical timeouts, trainer calls…
+2026-09-16T00:00:00Z
+2026-09-16T00:00:00Z
+
+
+Added: GET /players/{playerId}/stoppages (PRO) and its alias GET /players/{playerId}/injuries: one player's in-match stoppages (medical timeouts, trainer calls; toilet breaks, pauses and other stoppages on request) plus the matches the player retired from or gave a walkover, newest first, over a window that defaults to the last 180 days. meta.latest_medical_timeout / previous_medical_timeout answer "the latest and the one before"; meta.record_starts states where each record begins (stoppage events from 2026-09-12; outcomes from 2026-08-18).
+<h3>Added</h3>
+<ul>
+<li><code>GET /players/{playerId}/stoppages</code> (PRO) and its alias <code>GET /players/{playerId}/injuries</code>: one player's in-match stoppages (medical timeouts, trainer calls; toilet breaks, pauses and other stoppages on request) plus the matches the player retired from or gave a walkover, newest first, over a window that defaults to the last 180 days. <code>meta.latest_medical_timeout</code> / <code>previous_medical_timeout</code> answer "the latest and the one before"; <code>meta.record_starts</code> states where each record begins (stoppage events from 2026-09-12; outcomes from 2026-08-18). In-match stoppages and match outcomes only; there is no off-court injury record.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.10
+1.13.10: Score.sequence: a backwards move on GET /matches/{matchId}/score has two documented causes, not one — a withdrawn state, or the read deferring to a strictly…
+2026-09-15T00:00:00Z
+2026-09-15T00:00:00Z
+
+
+Changed: Score.sequence: a backwards move on GET /matches/{matchId}/score has two documented causes, not one — a withdrawn state, or the read deferring to a strictly higher-trust source's fresh state (about 1% of live reads).
+<h3>Changed</h3>
+<ul>
+<li><code>Score.sequence</code>: a backwards move on <code>GET /matches/{matchId}/score</code> has two documented causes, not one — a withdrawn state, or the read deferring to a strictly higher-trust source's fresh state (about 1% of live reads). On the push feed and the native WebSocket the sequence only ever rises.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.9
+1.13.9: Score.accepted_at (string|null, live score reads and push frames): the instant we accepted this state, stamped once and never refreshed — the clock to…, and 2 more changes
+2026-09-15T00:00:00Z
+2026-09-15T00:00:00Z
+
+
+
+Added: Score.accepted_at (string|null, live score reads and push frames): the instant we accepted this state, stamped once and never refreshed — the clock to difference a latency study against. Changed: Score.timestamp documented for what it is: our clock, stamped at acceptance and then refreshed (at most every 8 s) while the owning source keeps re-asserting an unchanged state, so on a live read it is usually the last-assertion instant. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><code>Score.accepted_at</code> (string|null, live score reads and push frames): the instant we accepted this state, stamped once and never refreshed — the clock to difference a latency study against.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><code>Score.timestamp</code> documented for what it is: our clock, stamped at acceptance and then refreshed (at most every 8 s) while the owning source keeps re-asserting an unchanged state, so on a live read it is usually the last-assertion instant. It was described as "when this state was true"; it never was. Never an observation time from outside our process.</li>
+<li>The <code>Score</code> schema on this site now carries the full field set and descriptions: <code>sequence</code>, <code>age_seconds</code>, <code>stale</code>, <code>observed_age_seconds</code>, <code>sources_count</code> and <code>detail</code> were missing from the published table, and <code>sets</code>/<code>server</code>/<code>is_tiebreak</code>/<code>win_probability_p1</code>/<code>danger</code>/<code>timestamp</code> had no description.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.8
+1.13.8: Wording correction to 1.13.7: the pre-match hold rates enter the engine snapped to a 0.01 grid, not "rounded to three decimals"
+2026-09-15T00:00:00Z
+2026-09-15T00:00:00Z
+
+
+Changed: Wording correction to 1.13.7: the pre-match hold rates enter the engine snapped to a 0.01 grid, not "rounded to three decimals".
+<h3>Changed</h3>
+<ul>
+<li>Wording correction to 1.13.7: the pre-match hold rates enter the engine <strong>snapped to a 0.01 grid</strong>, not "rounded to three decimals". The mechanism and every field are unchanged; only the stated granularity was wrong.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.7
+1.13.7: win_probability_meta.market_anchored is documented for what it always was: the market-prior anchor applies to this match (a two-sided pre-match price and a…, and 1 more change
+2026-09-15T00:00:00Z
+2026-09-15T00:00:00Z
+
+
+
+Changed: win_probability_meta.market_anchored is documented for what it always was: the market-prior anchor applies to this match (a two-sided pre-match price and a pre-play first score). Added: win_probability_meta.anchor_effective (boolean|null): this row's win_probability_p1 differs from win_probability_p1_model.
+<h3>Changed</h3>
+<ul>
+<li><code>win_probability_meta.market_anchored</code> is documented for what it always was: the market-prior anchor <strong>applies to this match</strong> (a two-sided pre-match price and a pre-play first score). It was described as "whether the anchor moved <code>win_probability_p1</code>", which is a per-row claim the flag never made — a small anchor shift can solve to a probability identical to <code>win_probability_p1_model</code>. Recorded values keep their meaning.</li>
+</ul>
+<h3>Added</h3>
+<ul>
+<li><code>win_probability_meta.anchor_effective</code> (boolean|null): this row's <code>win_probability_p1</code> differs from <code>win_probability_p1_model</code>. The per-row test <code>win_probability_p1 != win_probability_p1_model</code> works on frames recorded before this field existed.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.6
+1.13.6: GET /history/matches/{id}/prices (PRO): the per-point price history — one row per played point with the score state the BASIC tape publishes and, per neutral…, and 3 more changes
+2026-09-15T00:00:00Z
+2026-09-15T00:00:00Z
+
+
+
+Added: GET /history/matches/{id}/prices (PRO): the per-point price history — one row per played point with the score state the BASIC tape publishes and, per neutral side, the match-winner quote in force when the point was captured (bid, ask, mid, spread), with lag_seconds and an honest resolution label (tick, minute, coarse). GET /history/matches?has_market=true|false and a has_market flag on every history row, so a match with a price tape can be enumerated before it is pulled. Plus 2 more changes.
+<h3>Added</h3>
+<ul>
+<li><code>GET /history/matches/{id}/prices</code> (PRO): the per-point price history — one row per played point with the score state the BASIC tape publishes and, per neutral side, the match-winner quote in force when the point was captured (<code>bid</code>, <code>ask</code>, <code>mid</code>, <code>spread</code>), with <code>lag_seconds</code> and an honest <code>resolution</code> label (<code>tick</code>, <code>minute</code>, <code>coarse</code>). Works on a live match too; 404 <code>no_market</code> when no market is mapped.</li>
+<li><code>GET /history/matches?has_market=true|false</code> and a <code>has_market</code> flag on every history row, so a match with a price tape can be enumerated before it is pulled.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li>Price retention: from 2026-09-15 the in-play ticks of a match with a mapped market are kept at full resolution and are not deleted; pre-match and idle ticks keep the existing tiers. Earlier ticks were already thinned.</li>
+<li><code>GET /matches/{id}/prices</code> answers 404 <code>no_market</code> for a known match without a market (it said <code>not_found</code>).</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.5
+1.13.5: pbp_coverage: "point" now means a per-point stream has delivered for the match — at least one played point past the seq 1 opener
+2026-09-14T00:00:00Z
+2026-09-14T00:00:00Z
+
+
+Changed: pbp_coverage: "point" now means a per-point stream has delivered for the match — at least one played point past the seq 1 opener.
+<h3>Changed</h3>
+<ul>
+<li><code>pbp_coverage: "point"</code> now means a per-point stream has <strong>delivered</strong> for the match — at least one played point past the <code>seq</code> 1 opener. Until then it reads <code>"game"</code>, including for a listed match whose stream holds only its opener because play has not started. Previously any point row, including the opener written before the first ball, was enough for <code>"point"</code>, so a client told to gate on the field could subscribe to a match that never advanced. The admission gate we recommend is <code>sequence > 1</code> together with <code>stale: false</code>. On the server, a match flagged live whose entire tape is still opener rows is demoted back to upcoming after twenty minutes, so such matches no longer linger in the live list.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.4
+1.13.4: ITF World Tennis Tour qualifying rounds (singles) are now listed and live-scored: matches appear in GET /matches with round, tournament and scheduled_time…
+2026-09-14T00:00:00Z
+2026-09-14T00:00:00Z
+
+
+Added: ITF World Tennis Tour qualifying rounds (singles) are now listed and live-scored: matches appear in GET /matches with round, tournament and scheduled_time, receive live scores while in play and complete on the official result.
+<h3>Added</h3>
+<ul>
+<li>ITF World Tennis Tour <strong>qualifying rounds</strong> (singles) are now listed and live-scored: matches appear in <code>GET /matches</code> with <code>round</code>, <code>tournament</code> and <code>scheduled_time</code>, receive live scores while in play and complete on the official result. Sundays and Mondays, the qualifying days, were previously almost empty because the published schedule carries main draws only. <code>GET /fixtures</code> still follows that schedule and does not list qualifying.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.3
+1.13.3: WTA / WTA 125 stoppages now arrive in two layers
+2026-09-13T00:00:00Z
+2026-09-13T00:00:00Z
+
+
+Changed: WTA / WTA 125 stoppages now arrive in two layers.
+<h3>Changed</h3>
+<ul>
+<li>WTA / WTA 125 stoppages now arrive in two layers. The match state is read within about 20 seconds and yields <code>trainer_called</code>, <code>toilet_break_*</code> and <code>stoppage_*</code> rows (player null). A later event layer with a variable delay — measured on 2026-09-13 from about one minute to an hour behind play — adds <code>medical_timeout_start/end</code> with the player and the exact instants. The 1.13.1 note implied the event layer was live; it is not.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.2
+1.13.2: ATP main-tour stoppages: the live match stage is now watched for its medical-timeout and interruption stages, and 1 more change
+2026-09-13T00:00:00Z
+2026-09-13T00:00:00Z
+
+
+
+Added: ATP main-tour stoppages: the live match stage is now watched for its medical-timeout and interruption stages. Changed: ITF coverage note: the first medical timeout was observed and published on 2026-09-13.
+<h3>Added</h3>
+<ul>
+<li>ATP main-tour stoppages: the live match stage is now watched for its medical-timeout and interruption stages. Those stages exist in the vocabulary but have not yet been observed on a tennis match, so ATP rows are documented as mapped, not yet proven; they carry <code>player: null</code> and an <code>at</code> equal to the instant the stage was observed.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li>ITF coverage note: the first medical timeout was observed and published on 2026-09-13.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.1
+1.13.1: Scorer-stated stoppages for WTA and WTA 125 singles: medical_timeout_*, trainer_called*, toilet_break_* and stoppage_* rows are now published for those tours…, and 2 more changes
+2026-09-13T00:00:00Z
+2026-09-13T00:00:00Z
+
+
+
+Added: Scorer-stated stoppages for WTA and WTA 125 singles: medical_timeout_*, trainer_called*, toilet_break_* and stoppage_* rows are now published for those tours (measured over 138 finished matches: 27 physio calls, 65 treatment records, 22 suspensions). Scorer-stated stoppages for ITF World Tennis Tour singles from the court's live-scoring state: toilet_break_*, medical_timeout_*, trainer_called* and stoppage_* rows on entering and leaving the state, with duration_seconds; player is null on these rows because the live-scoring state names the state, not the player. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li>Scorer-stated stoppages for <strong>WTA and WTA 125</strong> singles: <code>medical_timeout_*</code>, <code>trainer_called*</code>, <code>toilet_break_*</code> and <code>stoppage_*</code> rows are now published for those tours (measured over 138 finished matches: 27 physio calls, 65 treatment records, 22 suspensions). Same fields, same endpoints, same WebSocket signal. <code>stoppage_start.reason</code> gains <code>heat</code>, <code>darkness</code> and <code>injury</code>; every <code>stoppage_end</code> carries <code>reason: "resumed"</code>.</li>
+</ul>
+<ul>
+<li>Scorer-stated stoppages for <strong>ITF World Tennis Tour</strong> singles from the court's live-scoring state: <code>toilet_break_*</code>, <code>medical_timeout_*</code>, <code>trainer_called*</code> and <code>stoppage_*</code> rows on entering and leaving the state, with <code>duration_seconds</code>; <code>player</code> is null on these rows because the live-scoring state names the state, not the player.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li>Coverage note: WTA, WTA 125, Challenger, UTR and ITF singles are covered; ATP main tour is not covered yet and is stated as not covered.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.13.0
+1.13.0: GET /events (PRO): the slate-wide events feed — the rows of GET /matches/{id}/events for every match in one call, oldest first, cursor by after_id…, and 2 more changes
+2026-09-13T00:00:00Z
+2026-09-13T00:00:00Z
+
+
+
+Added: GET /events (PRO): the slate-wide events feed — the rows of GET /matches/{id}/events for every match in one call, oldest first, cursor by after_id (meta.next_cursor, null = caught up), since as the first-call UTC lower bound, type as a comma-separated list of event types or the family name stoppages. SlateEvent schema; Event.type documents the full vocabulary. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><code>GET /events</code> (PRO): the <strong>slate-wide events feed</strong> — the rows of <code>GET /matches/{id}/events</code> for every match in one call, oldest first, cursor by <code>after_id</code> (<code>meta.next_cursor</code>, null = caught up), <code>since</code> as the first-call UTC lower bound, <code>type</code> as a comma-separated list of event types or the family name <code>stoppages</code>. Rows carry <code>id</code> and <code>match_id</code>. One request per tick covers the whole live slate: medical timeouts across every live match on PRO by polling.</li>
+<li><code>SlateEvent</code> schema; <code>Event.type</code> documents the full vocabulary.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li>Stoppage coverage, corrected: scorer-stated <code>medical_timeout_*</code>, <code>trainer_called*</code> and <code>toilet_break_*</code> rows exist for <strong>Challenger and UTR singles</strong>. The 1.11.0/1.12.0 text said ATP, WTA and WTA 125 as well; measured over 11 days those tours' scorer data carries none (0 of 70 finished matches each, against 22 of 70 for Challenger). Main-tour and ITF medical timeouts are not covered yet; the note changes the day they are live.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.12.0
+1.12.0: serve and outcome on live point rows (GET /matches/{matchId}/points, the point frames): the serve the point was played on (1 | 2) and how it ended (ace |…, and 2 more changes
+2026-09-12T00:00:00Z
+2026-09-12T00:00:00Z
+
+
+Added: serve and outcome on live point rows (GET /matches/{matchId}/points, the point frames): the serve the point was played on (1 | 2) and how it ended (ace | double_fault | winner | forced_error | unforced_error), as stated for the point, joined onto our rows by exact state; null when not stated. point_update frame on the point opt-in — a point's serve/outcome landing after its point frame; apply by seq. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><strong><code>serve</code> and <code>outcome</code> on live point rows</strong> (<code>GET /matches/{matchId}/points</code>, the <code>point</code> frames): the serve the point was played on (1 | 2) and how it ended (<code>ace</code> | <code>double_fault</code> | <code>winner</code> | <code>forced_error</code> | <code>unforced_error</code>), as stated for the point, joined onto our rows by exact state; <code>null</code> when not stated. Coverage per match in the new <code>enrichment</code> object; per tour in the reference.</li>
+<li><strong><code>point_update</code> frame</strong> on the point opt-in — a point's <code>serve</code>/<code>outcome</code> landing after its <code>point</code> frame; apply by <code>seq</code>. Asked by an ULTRA customer trading on serve outcome.</li>
+<li><strong><code>published_at</code></strong> on every data frame of the native WebSocket and the push feed: the UTC instant (ms) the frame left our process — the third clock next to the state's <code>timestamp</code> and a point's <code>ts</code>, so processing and transport latency can be separated. Asked by a prospect running a latency benchmark.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.11.0
+1.11.0: Medical timeouts, trainer calls and toilet breaks as events on GET /matches/{matchId}/events: medical_timeout_start/end, trainer_called/_end…, and 2 more changes
+2026-09-12T00:00:00Z
+2026-09-12T00:00:00Z
+
+
+Added: Medical timeouts, trainer calls and toilet breaks as events on GET /matches/{matchId}/events: medical_timeout_start/end, trainer_called/_end, toilet_break_start/end, each with player (the player concerned), at (UTC), the score at that instant, reason and duration_seconds on the end row — as the match scorer states them (ATP, WTA, Challenger, WTA 125, UTR singles). stoppage_start/end now carry a stated reason (weather | other) when one exists. pause_start / pause_end (basis: inferred): interruptions of play measured from our own point clocks, on every live match, never labelled medical. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><strong>Medical timeouts, trainer calls and toilet breaks as events</strong> on <code>GET /matches/{matchId}/events</code>: <code>medical_timeout_start/end</code>, <code>trainer_called/_end</code>, <code>toilet_break_start/end</code>, each with <code>player</code> (the player concerned), <code>at</code> (UTC), the <code>score</code> at that instant, <code>reason</code> and <code>duration_seconds</code> on the end row — as the match scorer states them (ATP, WTA, Challenger, WTA 125, UTR singles). <code>stoppage_start/end</code> now carry a stated <code>reason</code> (<code>weather</code> | <code>other</code>) when one exists. Asked by a prospect building on medical timeouts.</li>
+<li><strong><code>pause_start</code> / <code>pause_end</code></strong> (<code>basis: inferred</code>): interruptions of play measured from our own point clocks, on every live match, never labelled medical.</li>
+<li><strong><code>signals:["stoppages"]</code></strong> on the native WebSocket and the stoppage family on the <code>signal:*</code> push channels — every row above pushed the moment it is recorded.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.10.1
+1.10.1: A ranking tie is two rows with the same rank (doubles partners always tie)
+2026-09-12T00:00:00Z
+2026-09-12T00:00:00Z
+
+
+Clarified: A ranking tie is two rows with the same rank (doubles partners always tie).
+<h3>Clarified</h3>
+<ul>
+<li>A ranking tie is two rows with the same <code>rank</code> (doubles partners always tie).</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.10.0
+1.10.0: win_probability_p1_model and win_probability_meta on ULTRA live score objects: the same model read computed without the market-prior anchor (a probability no…, and 3 more changes
+2026-09-12T00:00:00Z
+2026-09-12T00:00:00Z
+
+
+
+Added: win_probability_p1_model and win_probability_meta on ULTRA live score objects: the same model read computed without the market-prior anchor (a probability no market price touched), plus model_version, generated_at and market_anchored for the pair. win_probability_p1 is unchanged. basis on status-ledger rows (observed | backfill); the ledger now reaches back to the per-match stamps held before it existed (completions from 2026-08-21, promotions to live from 2026-09-05), one reconstructed row per stamp, labelled. Plus 2 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong><code>win_probability_p1_model</code></strong> and <strong><code>win_probability_meta</code></strong> on ULTRA live score objects: the same model read computed without the market-prior anchor (a probability no market price touched), plus <code>model_version</code>, <code>generated_at</code> and <code>market_anchored</code> for the pair. <code>win_probability_p1</code> is unchanged. Asked by an ULTRA customer who needed to know whether the live probability is independent of market prices — on anchored matches it is not, and now the row says so.</li>
+<li><strong><code>basis</code></strong> on status-ledger rows (<code>observed</code> | <code>backfill</code>); the ledger now reaches back to the per-match stamps held before it existed (completions from 2026-08-21, promotions to live from 2026-09-05), one reconstructed row per stamp, labelled.</li>
+</ul>
+<ul>
+<li><strong><code>system=atp_doubles</code> / <code>system=wta_doubles</code></strong> on <code>/rankings</code>: the official weekly doubles tables (individual players), in both modes at the same gates as <code>atp</code>/<code>wta</code>, never included implicitly. Loaded from 2023 forward.</li>
+</ul>
+<h3>Clarified</h3>
+<ul>
+<li>Status-ledger rows are ordered by their instant (<code>at</code>), not insertion order.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.9.9
+1.9.9: The status ledger records from 2026-09-11T22:45:48Z (its first row), not the calendar date of the deploy
+2026-09-12T00:00:00Z
+2026-09-12T00:00:00Z
+
+
+Clarified: The status ledger records from 2026-09-11T22:45:48Z (its first row), not the calendar date of the deploy.
+<h3>Clarified</h3>
+<ul>
+<li>The status ledger records from <strong>2026-09-11T22:45:48Z</strong> (its first row), not the calendar date of the deploy. Reported by an ULTRA customer reading match 188711.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.9.8
+1.9.8: GET /matches/{matchId}/status-history — the per-match status ledger: every status / event_status transition with the UTC instant we published it, the value…, and 1 more change
+2026-09-12T00:00:00Z
+2026-09-12T00:00:00Z
+
+
+Added: GET /matches/{matchId}/status-history — the per-match status ledger: every status / event_status transition with the UTC instant we published it, the value before and after, the derived outcome and the score at that instant. stoppage_start / stoppage_end events on GET /matches/{matchId}/events: an in-play suspension with the score at the moment, reason (unknown until a source states one — never inferred) and duration_seconds on the end row.
+<h3>Added</h3>
+<ul>
+<li><strong><code>GET /matches/{matchId}/status-history</code></strong> — the per-match status ledger: every <code>status</code> / <code>event_status</code> transition with the UTC instant we published it, the value before and after, the derived <code>outcome</code> and the score at that instant. Append-only, oldest first; rows exist from 2026-09-12. History capability (BASIC and the Historical Data plans). Asked for by a researcher reconciling corrections against the moment they were published.</li>
+<li><strong><code>stoppage_start</code> / <code>stoppage_end</code> events</strong> on <code>GET /matches/{matchId}/events</code>: an in-play suspension with the score at the moment, <code>reason</code> (<code>unknown</code> until a source states one — never inferred) and <code>duration_seconds</code> on the end row. Asked for by a product builder wanting medical timeouts as an event.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.9.7
+1.9.7: Match.outcome is now documented on the Match schema (it has been on every match since 2026-08-18): completed | retired | walkover | default | abandoned |…, and 1 more change
+2026-09-10T00:00:00Z
+2026-09-10T00:00:00Z
+
+
+Added: Match.outcome is now documented on the Match schema (it has been on every match since 2026-08-18): completed | retired | walkover | default | abandoned | unresolved | null, derived from status + event_status. outcome: unresolved — a match every source lost before a result is closed unfinished and says so: score is the last state observed, winner is null, no result is asserted; it flips to completed with the proven final when an authority confirms the result.
+<h3>Added</h3>
+<ul>
+<li><strong><code>Match.outcome</code></strong> is now documented on the Match schema (it has been on every match since 2026-08-18): <code>completed | retired | walkover | default | abandoned | unresolved | null</code>, derived from <code>status</code> + <code>event_status</code>.</li>
+<li><strong><code>outcome: unresolved</code></strong> — a match every source lost before a result is closed unfinished and says so: <code>score</code> is the last state observed, <code>winner</code> is null, no result is asserted; it flips to <code>completed</code> with the proven final when an authority confirms the result. Until 2026-09-10 such closes were published as <code>completed</code> with a mid-match score, which an ULTRA customer read as a wrong result.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.9.6
+1.9.6: stats.ratings_as_of on GET /players/{id} — the date (UTC) the ratings block was last refreshed, and 2 more changes
+2026-09-10T00:00:00Z
+2026-09-10T00:00:00Z
+
+
+
+Added: stats.ratings_as_of on GET /players/{id} — the date (UTC) the ratings block was last refreshed. GET /history/packages/{period}?format=corrections — a CSV keyed by match_id (match_id, field, before, after, tournament_key, source, corrected_at) on every package whose stored data was repaired after publication, listed in the manifest as format: corrections; 404 when a package has none. Plus 1 more change.
+<h3>Added</h3>
+<ul>
+<li><strong><code>stats.ratings_as_of</code></strong> on <code>GET /players/{id}</code> — the date (UTC) the <code>ratings</code> block was last refreshed. The current Elo is now refreshed every week for every rated player from the published rating tables (Tuesdays); before this the rating carried no date and a May capture read as this week's. Reported by a PRO customer building pre-match research.</li>
+<li><strong><code>GET /history/packages/{period}?format=corrections</code></strong> — a CSV keyed by <code>match_id</code> (<code>match_id, field, before, after, tournament_key, source, corrected_at</code>) on every package whose stored data was repaired after publication, listed in the manifest as <code>format: corrections</code>; 404 when a package has none. First use: <code>surface</code> corrections on the 2023-02 → 2024-12 tape packages (19,439 Challenger matches whose court surface a tournament-name rule had set to grass; repaired and rebuilt 2026-09-10). Reported by a History Pro customer.</li>
+</ul>
+<h3>Fixed</h3>
+<ul>
+<li>Merged player ids: a merge now repoints the weekly ranking tables too, so <code>/rankings</code> rows carry the same <code>player_id</code> the match records carry, and the retired id answers 410 with <code>merged_into</code> (see Merged and retired player ids).</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.9.5
+1.9.5: GET /matches/{matchId}/prices?cursor= — keyset paging past 500 ticks: while meta.has_more is true, pass meta.next_cursor back as ?cursor= for the next (older)…, and 4 more changes
+2026-09-09T00:00:00Z
+2026-09-09T00:00:00Z
+
+
+
+Added: GET /matches/{matchId}/prices?cursor= — keyset paging past 500 ticks: while meta.has_more is true, pass meta.next_cursor back as ?cursor= for the next (older) page; pages never overlap or skip a tick; anything else is 400 bad_cursor. Match.live_at — the instant our feed last reported the match in play (UTC), the closest thing to an actual start time. Plus 3 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong><code>GET /matches/{matchId}/prices?cursor=</code></strong> — keyset paging past 500 ticks: while <code>meta.has_more</code> is true, pass <code>meta.next_cursor</code> back as <code>?cursor=</code> for the next (older) page; pages never overlap or skip a tick; anything else is <code>400 bad_cursor</code>. Mirrors tennis 3123b27f (live 2026-09-09 06:01Z).</li>
+<li><strong><code>Match.live_at</code></strong> — the instant our feed last reported the match in play (UTC), the closest thing to an actual start time. Null for matches that went live before 2026-09-05 or were never observed live. Same commit.</li>
+<li><strong><code>410 Gone</code> with a <code>MatchMerged</code> body</strong> on the ten per-match routes when a match id was merged into another record: <code>error: merged</code>, <code>merged_into</code> (the end of the chain, or null), <code>merged_at</code>, <code>detail</code>. Live since 2026-09-05 (tennis d131a12f).</li>
+<li><strong>Tape point <code>is_unreturned</code> / <code>unreturned_kind</code></strong> (rally tapes, tennis 6e7b82f6, 2026-09-05).</li>
+</ul>
+<h3>Clarified</h3>
+<ul>
+<li>Price ticks are kept for <strong>30 days</strong> and then deleted: an older match answers an empty <code>data</code> / <code>prices</code> array while its market stays mapped — retention, not a fault. Stated on both prices endpoints.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.9.4
+1.9.4: Price.side (per-match and per-market prices endpoints, and the match embed) now follows players.p1 / players.p2 through our name-verified match-market mapping, and 1 more change
+2026-09-09T00:00:00Z
+2026-09-09T00:00:00Z
+
+
+
+Fixed: Price.side (per-match and per-market prices endpoints, and the match embed) now follows players.p1 / players.p2 through our name-verified match-market mapping. Clarified: Price.mid is the venue's observed midpoint, never a model estimate; synthetic describes only bid/ask (mid +/- 0.005 when true); pre-match ticks are synthetic: true by design because the live book attaches at match start; timestamp is our observation clock (UTC), not a venue publication time.
+<h3>Fixed</h3>
+<ul>
+<li><strong><code>Price.side</code></strong> (per-match and per-market prices endpoints, and the match embed) now follows <code>players.p1</code> / <code>players.p2</code> through our name-verified match-market mapping. Until 2026-09-09 it followed the venue's display order, which lists roughly half of all pairings the other way round, so on those markets <code>side: 1</code> was in fact <code>players.p2</code>. The mapping is stored, so re-reading any tick - historical included - returns the correct side. Reported by a PRO customer cross-checking pre-match prices against rankings; mirrors tennis cbd3f647 (live 2026-09-09 04:59Z).</li>
+</ul>
+<h3>Clarified</h3>
+<ul>
+<li><code>Price.mid</code> is the venue's observed midpoint, never a model estimate; <code>synthetic</code> describes only <code>bid</code>/<code>ask</code> (<code>mid</code> +/- 0.005 when true); pre-match ticks are <code>synthetic: true</code> by design because the live book attaches at match start; <code>timestamp</code> is our observation clock (UTC), not a venue publication time.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.9.3
+1.9.3: Match.tournament_id null cases restated with the real causes and a measured rate: not in our tournament list (UTR), or a match whose name matched no single…
+2026-09-04T00:00:00Z
+2026-09-04T00:00:00Z
+
+
+Clarified: Match.tournament_id null cases restated with the real causes and a measured rate: not in our tournament list (UTR), or a match whose name matched no single edition in that list.
+<h3>Clarified</h3>
+<ul>
+<li><strong><code>Match.tournament_id</code> null cases</strong> restated with the real causes and a measured rate: not in our tournament list (UTR), or a match whose name matched no single edition in that list. The API now resolves the latter at ingest by name + event type + date against the published fixtures, and the 2026 backlog was backfilled the same day (2,547 rows); ITF null rate went from about 25% to about 2.5%. No wire change.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.9.2
+1.9.2: GET /matches?status=cancelled is now in the status enum, with what it covers (feed-cancelled, walkover with no stated winner, postponed and never played), its…, and 3 more changes
+2026-09-04T00:00:00Z
+2026-09-04T00:00:00Z
+
+
+
+Added: GET /matches?status=cancelled is now in the status enum, with what it covers (feed-cancelled, walkover with no stated winner, postponed and never played), its tier (BASIC or any History plan, like completed), and the one restriction: it pages with limit/offset and refuses updated_since. tournament_id= filter on /matches (every status) and /history/matches — the stable id each match row carries and /tournaments publishes. Plus 2 more changes.
+<h3>Added (spec catch-up — every one of these has been live on the API; the document lagged)</h3>
+<ul>
+<li><strong><code>GET /matches?status=cancelled</code></strong> is now in the <code>status</code> enum, with what it covers (feed-cancelled, walkover with no stated winner, postponed and never played), its tier (BASIC or any History plan, like <code>completed</code>), and the one restriction: it pages with <code>limit</code>/<code>offset</code> and refuses <code>updated_since</code>.</li>
+<li><strong><code>tournament_id=</code></strong> filter on <code>/matches</code> (every status) and <code>/history/matches</code> — the stable id each match row carries and <code>/tournaments</code> publishes. The description states there is no separate edition/occurrence id: one edition is <code>tournament_id</code> plus a <code>from</code>/<code>to</code> window, and matches with a null <code>tournament_id</code> (tournament not yet catalogued) fall outside the filter.</li>
+<li><strong>Change feed on <code>/matches</code></strong>: <code>updated_since=</code> / <code>cursor=</code> parameters and the <code>meta.next_cursor</code> / <code>meta.watermark</code> fields, with the at-least-once and no-<code>from</code>/<code>to</code> rules.</li>
+</ul>
+<h3>Clarified</h3>
+<ul>
+<li><code>meta.has_more</code> says how to enumerate a filtered set completely (page <code>offset</code> by <code>limit</code> until false) and that <code>total</code> is null on the terminal listings, so <code>has_more</code> is the only end-of-data signal there. Asked by a Basic customer on Discord; no wire change.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.9.1
+1.9.1: Player.ranking / Player.ranking_points now say what they are: the official singles ranking POSITION (the ordinal), ATP table for men and WTA table for women…
+2026-09-02T00:00:00Z
+2026-09-02T00:00:00Z
+
+
+Clarified: Player.ranking / Player.ranking_points now say what they are: the official singles ranking POSITION (the ordinal), ATP table for men and WTA table for women chosen by the player, refreshed ahead of each match the player has with us, null when no ranking is held, and always the CURRENT record even on historical matches (/rankings?as_of= is the as-of surface).
+<h3>Clarified</h3>
+<ul>
+<li><strong><code>Player.ranking</code> / <code>Player.ranking_points</code></strong> now say what they are: the official singles ranking POSITION (the ordinal), ATP table for men and WTA table for women chosen by the player, refreshed ahead of each match the player has with us, <code>null</code> when no ranking is held, and always the CURRENT record even on historical matches (<code>/rankings?as_of=</code> is the as-of surface). Asked by a Basic customer; no wire change.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.9.0
+1.9.0: has_analysis and has_market on Match — two booleans on every row of GET /matches and on the detail, every tier, and 1 more change
+2026-09-02T00:00:00Z
+2026-09-02T00:00:00Z
+
+
+Added: has_analysis and has_market on Match — two booleans on every row of GET /matches and on the detail, every tier. Distinguishable absence on GET /matches/{matchId}/analysis and GET /markets/{matchId}/prices.
+<h3>Added</h3>
+<ul>
+<li><strong><code>has_analysis</code> and <code>has_market</code> on <code>Match</code></strong> — two booleans on every row of <code>GET /matches</code> and on the detail, every tier. They carry the same fact the per-match endpoints answer 404 about, so a slate is filtered in one call instead of one 404 per match. Shipped to production 2026-09-02.</li>
+<li><strong>Distinguishable absence on <code>GET /matches/{matchId}/analysis</code> and <code>GET /markets/{matchId}/prices</code>.</strong> The status stays <code>404</code> (unchanged, and shipped clients branch on it), but the body now says which absence it is: <code>{"error":"not_found"}</code> for an id that does not exist, and <code>{"error":"no_analysis"|"no_market","match_id":…,"coverage":"none","detail":…}</code> for a real match we hold nothing for — the same <code>coverage: "none"</code> vocabulary <code>/matches/{matchId}/statistics</code> already uses in its <code>200</code>. New <code>error</code> codes <code>no_analysis</code>, <code>no_market</code> documented on the <code>Error</code> schema.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.8.0
+1.8.0: GET /history/archive/matches/{archiveId}/tape (getArchiveTape) — the RECONSTRUCTED 2013–2022 point-by-point tape for one archive result: the score sequence…, and 8 more changes
+2026-09-01T00:00:00Z
+2026-09-01T00:00:00Z
+
+
+
+Added: GET /history/archive/matches/{archiveId}/tape (getArchiveTape) — the RECONSTRUCTED 2013–2022 point-by-point tape for one archive result: the score sequence behind the published result: reconstructed point sequences, no capture timestamps. ArchiveTape schema. Plus 7 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong><code>GET /history/archive/matches/{archiveId}/tape</code></strong> (<code>getArchiveTape</code>) — the RECONSTRUCTED 2013–2022 point-by-point tape for one archive result: the score sequence behind the published result: reconstructed point sequences, no capture timestamps. Shipped to production 2026-09-01; the spec was the last place it was missing. Tier: core ULTRA, <strong>or any active History plan including Starter</strong> (which opens it on a FREE core key). The archive RESULT stays on BASIC, so core BASIC and core PRO read the result and are refused the tape — <code>403 upgrade_required</code> carrying <code>capability: archive_tape</code>.</li>
+<li><strong><code>ArchiveTape</code> schema.</strong> Same envelope as <code>HistoryTape</code> so one parser reads both halves of the tape product, with the differences that are true: <code>match</code> is the winner/loser-shaped archive row (rows are WINNER-FIRST, not p1/p2), <code>profiles</code> is always <code>[]</code>, and <code>meta</code> carries <code>archive_match_id</code> rather than <code>match_id</code> — an archive id is not a match id, and passing one to the other's routes resolves a different, real record without erroring. Rows are <code>HistoryTapeRow</code>, reused unchanged.</li>
+<li><strong><code>kind=archive_tape</code> on <code>/history/packages</code> and <code>/history/packages/{period}</code>, and on <code>HistoryPackage.kind</code>.</strong> Ten per-year bulk files, <code>period</code> 2013 through 2022, JSONL + CSV, all <code>ready</code>. The JSONL record is byte-for-byte what the per-match endpoint returns; the CSV is the flat per-row view keyed on <code>archive_match_id</code> and deliberately carries no <code>timestamp</code>, <code>win_probability_p1</code> or <code>danger</code> column. A SEPARATE gate from the per-match tape: ULTRA, a History Pro/Business subscription, or an active one-off package window — a Starter grant reads tapes one at a time and does not download years of them. <strong>Core PRO carries neither gate.</strong></li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong><code>info.description</code> states the provenance and the coverage, thin spots included.</strong> Nobody watched these matches: <code>timestamp</code>, <code>win_probability_p1</code> and <code>danger</code> are null on EVERY row and cannot be filled in later — the production table has no timestamp column at all, so this is a structural fact and not a convention. That is the opposite of the 2023→now tape, which is our own recording, where the rows we actually watched carry a real clock and most of them a model probability. The corpus is 97,901 matches / 14,340,663 rows, seasons <strong>2013–2022 only</strong>: 977,903 archive results from 1968–2012 have no tape and never will. Coverage of the era is 19.3% overall and 44.9% of tour-level play — main-draw tour buckets 91.6–98.7%, ATP Challenger main 55.3% and Challenger qualifying 33.6%, slam QUALIFYING 16.0% (ATP) / 18.1% (WTA), ITF and futures effectively zero (25 of 116,575 ATP futures). Stated together, because a strong number published without its thin counterpart sells a corpus nobody has.</li>
+<li><strong><code>ArchiveTape.meta.coverage</code> names the measured split rather than glossing the label.</strong> <code>reconstructed_partial</code> (3,594 matches) has TWO causes and does not say which: 3,038 are point-granular tapes of matches that genuinely stopped early (3,027 retirements, 11 defaults) — the larger cause — and the other 556 carry the label only because their tape is per-GAME. Read <code>granularity</code> and the match's own score, not the label, when the question is whether the whole match is there. <code>granularity</code> is <code>point</code> on 99.4% of the corpus; the 556 <code>game</code> tapes are 555 in 2013 and one in 2014.</li>
+<li><strong><code>Coverage</code> says where its vocabulary stops.</strong> It describes the 2023+ tape; the archive tape reuses two of its five values and derives <code>reconstructed_partial</code> differently.</li>
+<li><strong><code>ArchiveMatch</code> points at the tape</strong>, and <code>GET /history/archive/matches/{archiveId}</code> says the RESULT stays on BASIC either way.</li>
+<li>The plain-HTML reference gains an FAQ entry — "Is there point-by-point data before 2023?" — and the history FAQ, <code>llms.txt</code> digest and README plan summary carry the same numbers, so an answer engine reading any one of them gets the corpus, the null clock and the coverage floor together.</li>
+<li><code>info.version</code> is now <code>1.8.0</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.7.2
+1.7.2: status on Match now documents the lifecycle rule, and 1 more change
+2026-08-23T00:00:00Z
+2026-08-23T00:00:00Z
+
+
+Changed: status on Match now documents the lifecycle rule. info.version is now 1.7.2.
+<h3>Changed</h3>
+<ul>
+<li><strong><code>status</code> on Match now documents the lifecycle rule.</strong> <code>completed</code> is asserted only for a match we observed being played or whose match-winner market settled decisively; a closed market alone never finishes a match. <code>cancelled</code> with <code>event_status: null</code> means positive evidence the match was not played as scheduled (the market settled void) and no stated reason for why — <code>outcome</code> / fixture <code>reason</code> stay null rather than guessed, and the row upgrades to a completed walkover if a <code>Walk Over</code> with a stated winner lands later. Before 2026-08-23 a void market closure completed the match and stamped <code>Finished</code>; those rows were corrected. No field added or changed type — prose only.</li>
+<li><code>info.version</code> is now <code>1.7.2</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.7.1
+1.7.1: event_status_updated_at on Match, and 1 more change
+2026-08-19T00:00:00Z
+2026-08-19T00:00:00Z
+
+
+
+Added: event_status_updated_at on Match. Changed: info.version is now 1.7.1.
+<h3>Added</h3>
+<ul>
+<li><strong><code>event_status_updated_at</code> on Match.</strong> Nullable ISO-8601 UTC (<code>Z</code>) timestamp, right after <code>event_status</code>: when <code>event_status</code> last CHANGED — the instant WE recorded the walkover / retirement / cancellation / postponement / suspension (or its clearing), not when the tournament desk did. It is the field to measure our admin-status latency with. Bumps only on a change of value (a re-read of the same status never moves it; a clear back to null does). Null while <code>event_status</code> has never changed since the field was introduced (2026-08-19) — never backfilled, never guessed. Inherited by every schema built on Match (<code>MatchDetail</code>, <code>HistoryMatch</code>). Additive only — no existing field moved or changed type.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><code>info.version</code> is now <code>1.7.1</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.7.0
+1.7.0: basis on GET /matches/{matchId}/points responses, and 1 more change
+2026-08-18T00:00:00Z
+2026-08-18T00:00:00Z
+
+
+
+Added: basis on GET /matches/{matchId}/points responses. Changed: info.version is now 1.7.0.
+<h3>Added</h3>
+<ul>
+<li><strong><code>basis</code> on <code>GET /matches/{matchId}/points</code> responses.</strong> <code>live</code> | <code>reconstruction</code> — which base served the page. Live capture is inherently partial: the stream serves what arrived while the match ran, and the match-closing point never streams live. For a COMPLETED match where a measured-complete recorded point sequence exists, the endpoint now serves that complete sequence instead, projected into the same point-frame shape — love-love opener through the match-closing point, <code>seq</code> contiguous <code>1..N</code>, <code>quality</code> <code>clean</code> — and says so with <code>basis: "reconstruction"</code>. Every other case (every live match, and any completed match without a measured-complete recorded sequence) stays <code>basis: "live"</code> — the persisted stream rows, byte-for-byte what was served before. Completeness beats the partial live capture WHOLESALE — the two sequences are never interleaved (they share no key, so any merge would fabricate an order), the same rule <code>?points=complete</code> follows on the history read. On projected frames <code>ts</code> is null on every row: the recorded sequence carries no per-point clock and none is fabricated (<code>LivePoint.ts</code> now documents this). <code>after_seq</code> pagination and <code>seq</code> dedup work identically on either basis, but the two bases are different sequences: after a match completes and flips to <code>reconstruction</code>, re-read from <code>after_seq=0</code> rather than resuming a live cursor into it. Additive only — no existing field moved or changed type.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><code>info.version</code> is now <code>1.7.0</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.6.0
+1.6.0: The three-valued draw field on Match, and 5 more changes
+2026-08-18T00:00:00Z
+2026-08-18T00:00:00Z
+
+
+
+Added: The three-valued draw field on Match. ?draw=singles|doubles on four listings — GET /matches, /history/matches, /tournaments and /fixtures: the axis the tour filter deliberately collapses; the two compose (?tour=itf&draw=doubles is the ITF doubles slice). Plus 4 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong>The three-valued <code>draw</code> field on Match.</strong> <code>singles</code> | <code>doubles</code> | null — same vocabulary as the new <code>?draw=</code> filter, decided by one shared definition, so filter and field cannot disagree. Evidence order: a doubles-team participant proves doubles over any event type; otherwise the stated event type decides. Null means neither says anything — no stated event type, or a team tie (Davis Cup / BJK Cup / United Cup class), where one event type covers both singles and doubles rubbers and we will not guess which this is. Null is NOT singles. <code>is_doubles</code> stays for compatibility and is now documented as LOSSY, with its evidence order: false also covers "unknown", which is not a claim of singles — prefer <code>draw</code>, whose null says so honestly.</li>
+<li><strong><code>?draw=singles|doubles</code> on four listings</strong> — <code>GET /matches</code>, <code>/history/matches</code>, <code>/tournaments</code> and <code>/fixtures</code>: the axis the <code>tour</code> filter deliberately collapses; the two compose (<code>?tour=itf&draw=doubles</code> is the ITF doubles slice). A row whose draw is null matches NEITHER value — null is an answer, not a wildcard. An unknown value is a 400 <code>bad_draw</code> with the allowed values. Two honesty notes carried into the spec: on <code>/tournaments</code> the answer comes from the event type alone (a tournament row has no participants to supply the doubles-team evidence matches have), and <code>?draw=doubles</code> alone also returns mixed and exhibition doubles that no <code>?tour=</code> value reaches.</li>
+<li><strong><code>GET /history/coverage</code></strong> (BASIC, or any Historical Data API plan) — the measured completeness rollup per tour × draw bucket: the numbers to read BEFORE choosing what to backtest, in one call instead of paging the archive. A prebuilt snapshot rebuilt nightly right after the completeness ledger reconverges — never computed at read time — so <code>as_of</code> (= <code>built_at</code>) dates every number and <code>ledger_max_computed_at</code> is the newest underlying per-match measurement; <code>503 coverage_unavailable</code> before the first nightly build. Buckets are atp/wta/challenger/itf/juniors × singles/doubles plus <code>other</code> (team ties, mixed, exhibitions and matches with no stated event type — counted, never dropped, so the totals cannot lie); a bucket with zero completed matches is OMITTED, never emitted as zeros. Each bucket carries the five verifiable numbers (<code>completed</code>, <code>any_tape</code>, <code>point_complete</code>, <code>complete_on_default_read</code>, <code>share</code>), and <code>method</code> states the full measurement rule so every number carries its own definition. As of 2026-08-18: 174,393 completed matches; 91,318 point-complete on the best basis (52.4%) against 81,196 on the default read alone (46.6%); ITF singles 51.1% against ITF doubles 3.5% — do not extrapolate a completeness rate across a tour group.</li>
+<li><strong><code>tape.starts_at_love</code> and <code>tape.computed_at</code> on <code>/history/matches</code> items</strong> (both nullable, present where enabled). <code>starts_at_love</code> follows the SAME best-basis rule as <code>points_complete</code> — true when either measured basis opens at the 0-0 state, so if any on-disk sequence opens at love you can obtain one that does. <code>computed_at</code> is when the ledger last measured the match: every field in the tape's measured block is a nightly-reconverged cache, and this is the as-of to quote with any of them. Null on either means the match has not been measured — never a guess.</li>
+<li><strong>The complete-basis addendum package files.</strong> An affected month's manifest may also list <code>tennis_history_points_complete_<period>.jsonl.gz</code> / <code>.csv.gz</code> (marked <code>compression: gzip</code>; <code>bytes</code>/<code>sha256</code> cover the compressed bytes — exactly the download): for exactly the matches whose complete point sequence exists only as the on-disk reconstruction, the same tape <code>?points=complete</code> serves (reconstruction contract: null timestamps and null model fields). The base files keep carrying every match's DEFAULT read — already the complete tape for most point-complete matches — and are never rewritten by the addendum; their <code>bytes</code> and <code>sha256</code> values do not move.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><code>info.version</code> is now <code>1.6.0</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.5.0
+1.5.0: The as-of Elo tape, and 7 more changes
+2026-08-17T00:00:00Z
+2026-08-17T00:00:00Z
+
+
+
+Added: The as-of Elo tape. covers_from_start on GET /matches/{matchId}/points (bool|null): whether the persisted stream opens at the match's 0-0 opener (seq 1 is the love-love state) — strictly affirmative; null only when the match has no rows at all. Plus 6 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong>The as-of Elo tape.</strong> <code>GET /rankings?system=elo</code> (ULTRA, both modes) — our own computed Elo as point-in-time records: per-player as-of (also via <code>archive_player</code> for the ~62,000 rated people outside the roster, <code>tour</code> required there) and a leaderboard (<code>tour</code> required — the ATP and WTA walks are disjoint; exactly one <code>surface</code>, default <code>overall</code>; <code>min_matches</code> default 20 and <code>activity_weeks</code> default 52 / max 104, both echoed in <code>meta.coverage.qualified</code>). Four independent ladders (overall/hard/clay/grass), ATP from 1877, WTA from 1968, main tours plus challengers plus the futures tier. <code>rating</code> always; <code>rank</code> leaderboard-only; <code>points</code> always null; <code>matches</code> is the ladder-scoped count. <code>meta.coverage</code> gains <code>newest_available</code>, <code>players_rated</code> / <code>players_linked</code>, <code>qualified</code> and the <code>model</code> block (<code>publication_lag_days: 14</code> — a week's results become effective 14 days after the week begins, so the failure direction is staleness, never look-ahead). The corpus head is frozen at 2026-05-25 and does not advance; <code>elo</code> is never included implicitly. The free current Elo on <code>GET /players/{id}</code> is a DIFFERENT scale (~150 Elo of per-player standard deviation apart) — never present a rating from one scale against a rating from the other. <code>kind=elo</code> yearly bulk packages join <code>/history/packages</code>.</li>
+<li><strong><code>covers_from_start</code> on <code>GET /matches/{matchId}/points</code></strong> (bool|null): whether the persisted stream opens at the match's 0-0 opener (seq 1 is the love-love state) — strictly affirmative; null only when the match has no rows at all.</li>
+<li><strong>The point channel family enters the <code>/ws-token</code> vocabulary.</strong> <code>point:match:{match_id}</code> / <code>point:slate</code> are documented as <code>point_match</code> / <code>point_slate</code> in the channel vocabulary, listed only when the point feed is enabled server-side and the key's plan carries the point surface. A channel named in that response is a promise; a missing one will not deliver.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong>Measured statistics stop rounding the truth.</strong> <code>MatchStatisticsMeasured</code> now states: match totals only (no per-set measured statistics); absent = not measured while a present 0 is a real measured zero; the three coverage tiers with their hard zeros — the winners/unforced/forced-errors family historically on ~43% of ATP singles, ~24% of WTA singles and ~47% of tour doubles, and NONE of Challenger, ITF or juniors (24,552 payloads, 2026-07-31) — and that the family has not been available since 2026-07-12 (0 of 4,513 August payloads carry it, measured 2026-08-17).</li>
+<li><strong><code>errors_total</code> is documented as the total of FORCED errors</strong>, with the evidence (3,766 payload sides, June–July 2026: equals the per-stroke error sum in 96.2%; smaller than <code>unforced_errors_total</code> in 11.7% of sides — impossible for a superset; per-match points accounting closes only under the forced reading, median residual 0 over 367 matches). Total errors = <code>errors_total</code> + <code>unforced_errors_total</code>; the <code>*_errors</code> shot family is the forced-error breakdown, and <code>groundstroke_errors</code> = <code>forehand_errors</code> + <code>backhand_errors</code>, a rollup rather than an addition.</li>
+<li><strong>UTR honesty.</strong> <code>system=utr</code> is documented as observed from UTR's public search: withheld ratings are ABSENT, never 0; per-player as-of only — no listing by design (a table of only the players we happen to track would be a fake leaderboard); history since 2026-07-29; and the sweep's deliberate rankless / no-Elo (ITF-skewed) bias with its measured coverage (2026-08-17, players active in the last 60 days): ITF 931 of 5,606 (16.6%), Challenger 197 of 1,903 (10.4%), WTA 43 of 573 (7.5%), ATP 15 of 525 (2.9%).</li>
+<li>*(recorded retroactively)* On 2026-08-16 the <code>/ws-token</code> description was expanded in place to teach the push-feed protocol (Centrifugo v2 connect/subscribe/heartbeat, token-per-reconnect, the SDK <code>PushStream</code> pointer) without a version bump; this entry is the changelog record of that change.</li>
+<li><code>info.version</code> is now <code>1.5.0</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.4.0
+1.4.0: Live per-point events, and 6 more changes
+2026-08-16T00:00:00Z
+2026-08-16T00:00:00Z
+
+
+
+Added: Live per-point events. WebSocket points signal. Plus 5 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong>Live per-point events.</strong> <code>GET /matches/{matchId}/points</code> (ULTRA) — the live per-point event stream of one match in <code>seq</code> order, paged with <code>?after_seq=</code> (the resume cursor; up to 500 rows per page, continue on <code>last_seq</code> while <code>has_more</code>). It is the REST catch-up for the WebSocket <code>point</code> frames, which are best-effort with no replay. Per-match <code>pbp_coverage</code> states honestly whether the match has a true per-point stream (<code>point</code>) or only the snapshot score path (<code>game</code> — <code>points</code> empty, an answer rather than an error); per-point coverage is never promised slate-wide. New schemas <code>MatchPoints</code>, <code>LivePoint</code> and <code>PointFrame</code>; new error codes <code>bad_after_seq</code> and <code>points_disabled</code> (the surface switched off server-side).</li>
+<li><strong>WebSocket <code>points</code> signal.</strong> The native <code>/ws</code> feed's <code>signals</code> array may now name <code>points</code> to opt into one <code>point</code> frame (schema <code>PointFrame</code>) per persisted point of the subscribed matches. Config-gated, off by default — the <code>subscribed</code> ack echoes the signals actually active. The push feed carries the same frames on their own channel family (<code>point:match:{match_id}</code>, <code>point:slate</code>), deliberately separate from the score channels; webhooks gain the matching <code>point</code> event (one POST per live point, <code>X-LTAPI-Event: point</code>). Point frames are events, not states — a missed one does not self-correct; recovery is the REST catch-up read.</li>
+<li><strong>Point-complete tape reads.</strong> <code>?points=default|complete</code> on <code>GET /history/matches/{matchId}</code>: <code>complete</code> opts out of observed-rows-first precedence and serves a whole-match reconstruction WHOLE, in point order, where one exists (<code>point_winner</code> on every row, null timestamps/model fields per the reconstruction contract); where none exists the response is the default read plus <code>meta.points</code> — no error. Cannot combine with <code>sequence=clean</code> (400 <code>bad_combination</code>); an unknown value is a 400 <code>bad_points</code>; where not yet enabled, <code>complete</code> answers 400 <code>points_read_disabled</code> rather than silently serving the default. The tape's <code>meta.points</code> block (schema <code>PointsMeta</code>) reports the measured point-completeness of exactly the sequence returned — computed at read time, per match, never a stored blanket claim.</li>
+<li><strong>Point-completeness on the listing.</strong> <code>HistoryMatch.tape</code> gains <code>points_complete</code> (bool|null — the nightly ledger's best-basis verdict; null means not yet measured, never a guess) and <code>completeness</code> (0..1|null), and <code>GET /history/matches</code> gains the <code>?points_complete=true|false</code> filter (applied after the page is cut, exactly like <code>?coverage=</code>; anything but true/false is a 400 <code>bad_points_complete</code>).</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong>Model win-probability copy tells the truth about density.</strong> The plan and FAQ copy no longer claims the tape carries the model win-probability "at every point" / "per point": the model stamp is best-effort and rides the rows where the model ran — <code>meta.model_rows</code> is the count, and null is the honest value elsewhere. The current-score snapshot is described as overwritten on every score commit, not "on every point".</li>
+<li><code>HistoryTapeRow.point_winner</code> is documented on <code>?points=complete</code> rows as well as <code>?sequence=clean</code> (there the served order IS point order).</li>
+<li><code>info.version</code> is now <code>1.4.0</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.3.1
+1.3.1: kind=rally bulk packages documented
+2026-08-07T00:00:00Z
+2026-08-07T00:00:00Z
+
+
+Added: kind=rally bulk packages documented.
+<h3>Added</h3>
+<ul>
+<li><strong><code>kind=rally</code> bulk packages documented.</strong> The <code>/history/packages</code> <code>kind</code> enum gains <code>rally</code> — the charted rally corpus (shot-by-shot) as YEARLY exports, ULTRA, <code>period</code> = <code>YYYY</code> like <code>archive</code>. The kind has been live in the API (it shipped alongside the per-match rally endpoints); the spec simply did not list it. Additive only: the enum, the <code>HistoryPackage.kind</code> field, and the package-shape prose now match the served surface — <code>kind</code> accepts <code>tape</code> (default), <code>rankings</code> (ULTRA), <code>rally</code> (ULTRA) and <code>archive</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.3.0
+1.3.0: Shot-level charting, and 7 more changes
+2026-08-07T00:00:00Z
+2026-08-07T00:00:00Z
+
+
+
+Added: Shot-level charting. Push-feed token. Plus 6 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong>Shot-level charting.</strong> <code>GET /charting/players</code> (ULTRA) — career serve/return profile: serve placement (deuce/ad × wide/body/T), return depth and outcomes, net and serve-and-volley conversion, clutch break/game/set-point serving and returning, winners and unforced errors by wing, rally-length and shot-direction tendencies, summed over the player's charted matches (<code>name</code> keyed, <code>gender=men|women</code> disambiguates, ambiguous fragments refused with candidates). <code>GET /charting/matches/{chartingMatchId}</code> (ULTRA) — every stat family for one charted match, both players, per-set split. Coverage is curated: 11,646 charted matches back to the 1960s, concentrated on the majors, not full-slate.</li>
+<li><strong>Push-feed token.</strong> <code>GET /ws-token</code> (ULTRA): mints a short-lived signed token plus the push WebSocket URL and channel vocabulary — <code>match:{match_id}</code> per-match streams and <code>slate:all</code> for every live score frame. A separate high-fan-out surface from the native <code>/ws</code> feed.</li>
+<li><strong>H2H stat splits.</strong> On ULTRA, <code>GET /h2h</code> adds a per-player <code>stats</code> block: serve/return/break-point aggregates over the pairing — <code>archive_serve</code> (serve-side, from 1991) and <code>current</code> (2023+, adding return and break-point conversion, aces and winners), each with <code>meetings_with_stats</code>.</li>
+<li><strong>Abuse throttle documented.</strong> The 429 family now documents all three body shapes: the per-minute limit (<code>rate_limited</code> with <code>upgrade_url</code>/<code>tier</code>/ <code>price</code>), the per-day quota (<code>rate_limited</code> with <code>scope: "day"</code>, <code>limit_per_day</code> and <code>resets_at</code> — an absolute ISO instant), and <code>abuse_throttled</code> with <code>retry_at_epoch</code> — a 24-hour block for clients hammering far past their cap, which a well-behaved retry loop never sees.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong>2026-08-06 quota grid re-set (recorded here retroactively).</strong> On 2026-08-06 the daily quotas were cut, with no grandfathering: FREE 100/day (was 1,000), BASIC 1,000/day (was 10,000), PRO 10,000/day (was 100,000); ULTRA unchanged at 500,000/day; per-minute limits unchanged (30/60/300/600). The shipped 1.2.0 spec text was edited in place on that date without a version bump — this entry is the changelog record of that change. Older entries below quote the pre-cut grid as it stood then.</li>
+<li><strong>Tours.</strong> Coverage phrasing is now the five tours everywhere — ATP, WTA, Challenger, ITF and juniors — matching the <code>tour</code> filter enum (<code>atp, wta, challenger, itf, juniors</code>).</li>
+<li><strong>WebSocket copy.</strong> The subscribe frame is documented as <code>{"topics":["live-scores"]}</code> (+ optional <code>signals</code>) — the previously shown <code>action</code> key is not read by the server. Score frames are documented as carrying the ULTRA model fields (<code>win_probability_p1</code>, <code>danger</code>) live — a null means the model had no output for that point. The 2-connections-per- key limit is stated.</li>
+<li><code>info.version</code> is now <code>1.3.0</code>.</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.2.0
+1.2.0: Results archive (1968–2022), and 11 more changes
+2026-08-03T00:00:00Z
+2026-08-03T00:00:00Z
+
+
+
+Added: Results archive (1968–2022). Head-to-head. Plus 10 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong>Results archive (1968–2022).</strong> The history product now runs in two continuous, non-overlapping halves: the point-by-point tape (2023→now) and the results archive (1968–2022). Four new endpoints, all BASIC (or any Historical Data API plan): <code>GET /history/archive/matches</code> (winner/loser- shaped results — ATP and WTA, main draws, qualifying and the ITF/futures tiers, 1968 through 2022, with final score, seeds, ranks at the time; filters <code>tour</code>, <code>name</code>, <code>from</code>/<code>to</code>, <code>round</code>, <code>level</code>; its own id space; <code>event_date</code> is the TOURNAMENT START date; ends 2022-12-31, exactly where the tape begins), <code>GET /history/archive/matches/{archiveId}</code> (one result, with per-match serve statistics where the era recorded them — null before 1991 mostly, never synthesised), <code>GET /history/archive/players</code> (bios + career-high rank and the week it was first reached), and <code>GET /history/archive/career</code> (career aggregates — sums and ratios of sums only; <code>serve.matches_with_stats</code> states the serve-stat coverage). New schemas <code>ArchiveMatch</code>, <code>ArchivePlayer</code>, <code>ArchivePlayerBio</code>, <code>ArchiveCareer</code>.</li>
+<li><strong>Head-to-head.</strong> <code>GET /h2h?p1=&p2=</code> (BASIC, or any History plan): the record between two players across both halves of the product. Name-keyed; an ambiguous fragment is refused with the candidate list (<code>400 ambiguous_name</code>); totals count meetings with a known winner and <code>undecided</code> counts the rest; every meeting carries <code>outcome</code> so walkovers and retirements can be excluded. New schema <code>HeadToHead</code>.</li>
+<li><strong>Rally construction.</strong> <code>GET /rally/matches</code>, <code>GET /rally/matches/{rallyMatchId}</code> and <code>GET /history/matches/{matchId}/rally</code> (all ULTRA): shot-by-shot charted data — serve direction, every stroke with wing/direction/depth, rally length, how the point ended. Its own id space (<code>rally_match_id</code>); <code>404 not_charted</code> distinguishes "we hold the match but nobody charted it" from "no such match". New schemas <code>RallyMatch</code>, <code>RallyPoint</code>, <code>RallyShot</code>.</li>
+<li><strong>Tournament catalogue.</strong> <code>GET /tournaments</code> and <code>GET /tournaments/{tournamentId}</code> (FREE): the stable id space <code>Match.tournament_id</code> joins, with <code>city</code>/<code>country</code> from a curated table and <code>category</code> only where the catalogues agree unambiguously — never derived from the name. New schema <code>Tournament</code>.</li>
+<li><strong>Rankings listing mode.</strong> <code>GET /rankings</code> without <code>player</code> (PRO) returns the FULL published table in rank order for exactly one <code>system</code> — rows carry <code>player_name</code> as published and a null <code>player_id</code> outside our roster, so a top-N has no silent holes; <code>meta.coverage.effective_date</code> names the week served; <code>utr</code> has no listing. Per-player as-of records stay ULTRA. <code>RankingRecord</code> gains <code>player_name</code>, <code>previous_rank</code> (ATP/WTA) and <code>rank_movement</code> (ITF).</li>
+<li><strong>List filters.</strong> <code>/matches</code> gains <code>player</code> (repeatable, max 50, either-participant), <code>country</code> (IOC-style lowercase 3-letter codes — NOT ISO-3166), <code>from</code>/<code>to</code> (UTC day boundaries, every status) and keeps <code>tour</code>; <code>/history/matches</code> gains <code>tour</code>, <code>player</code> and <code>country</code> alongside its existing <code>from</code>/<code>to</code>/<code>coverage</code>.</li>
+<li><strong>Match fields.</strong> <code>Match</code> gains <code>tour</code> (the same vocabulary as the filter, null never guessed), <code>tournament_id</code>, <code>round_code</code> (controlled vocabulary <code>F</code>…<code>ER</code>, null when unrecognised), a documented <code>event_status</code> enum (<code>Retired</code> | <code>Cancelled</code> | <code>Walk Over</code> | <code>Postponed</code> | <code>Interrupted</code>, with the null-ambiguity caveat) and <code>withdrew</code> (1|2 on <code>Retired</code>/<code>Walk Over</code>, withdrawer = loser by rule); <code>winner</code> is now served for the full archive age.</li>
+<li><strong>Fixture fields.</strong> <code>Fixture</code> gains <code>start_time</code> (null until the order of play assigns a time), <code>player1_id</code>/<code>player2_id</code> (exact-key roster resolution, never a name match) and <code>round_code</code>.</li>
+<li><strong>Tape additions.</strong> Clean-sequence tape rows gain <code>point_winner</code> (derived from single-point transitions, never guessed; absent on raw); the tape detail gains a top-level <code>tiebreaks</code> array (observed terminal tiebreak scores per 7-6 set) and <code>meta.model_rows</code>; the history listing's <code>tape</code> object gains <code>model_rows</code>.</li>
+<li><strong>Archive bulk packages.</strong> <code>/history/packages</code> and <code>/history/packages/{period}</code> gain <code>kind=archive</code> — the results archive (1968–2022) as YEARLY exports (<code>period</code> = <code>YYYY</code>), gzipped, alongside the monthly tape packages; the file manifest documents <code>compression</code>.</li>
+<li><strong>Statistics <code>final</code>.</strong> The in-play statistics coverage vocabulary gains <code>final</code> — the closing figures of a completed match; a finished match cannot be "stale", so <code>age_seconds</code> is null there.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><code>info.version</code> is now <code>1.2.0</code>; <code>info.description</code> states the two history halves and the updated tier deltas (tournaments on FREE, the archive family and <code>/h2h</code> on BASIC/History plans, the rankings listing on PRO, rally and per-player as-of rankings on ULTRA).</li>
+</ul>
+
+
+https://docs.livetennisapi.com/changelog.html#v1.1.0
+1.1.0: Usage endpoint, and 24 more changes
+2026-08-02T00:00:00Z
+2026-08-02T00:00:00Z
+
+
+
+Added: Usage endpoint. As-of rankings. Plus 23 more changes.
+<h3>Added</h3>
+<ul>
+<li><strong>Usage endpoint.</strong> <code>GET /usage</code> (FREE — any tier): your own durable daily usage vs quota — tier, limits, today's calls and a 30-day history. Calls to it are quota-exempt. New schema <code>Usage</code>.</li>
+<li><strong>As-of rankings.</strong> <code>GET /rankings</code> (ULTRA): ranking records as they stood on a date — <code>player</code> (required, repeatable, max 50), <code>as_of</code>, and <code>system</code> (<code>atp</code>, <code>wta</code>, <code>itf_jt</code>, <code>itf_mt</code>, <code>itf_wt</code>, <code>utr</code>). Systems are never collapsed into a single "rank"; UTR carries a rating with null rank/points. <code>meta.coverage.oldest_available</code> gives the earliest date each system can answer for. New schemas <code>RankingRecord</code>, <code>RankingListMeta</code>.</li>
+<li><strong>In-play match statistics.</strong> <code>GET /matches/{matchId}/statistics</code> (ULTRA): two families, deliberately never merged — DERIVED (rebuilt from the point-by-point record: hold/break %, break points, service & return points) and MEASURED (counted during the match: aces, double faults, the serve split, winners/unforced errors). Each family carries its own <code>coverage</code>, <code>as_of</code> and <code>age_seconds</code>; absent measured fields are omitted, never zero-filled; <code>coverage: none</code> on both families is a 200 with null <code>players</code>, not a 404; a divergence guard withholds measured values when the families disagree. New schemas <code>MatchStatistics</code>, <code>MatchStatisticsSide</code>, <code>MatchStatisticsMeasured</code>, <code>MatchStatisticsFreshness</code>, <code>MatchStatisticsFamily</code>.</li>
+<li><strong>Webhooks.</strong> <code>POST /webhooks</code>, <code>GET /webhooks</code>, <code>DELETE /webhooks/{webhookId}</code> (ULTRA, direct keys only): we POST the same frames the WebSocket sends to your HTTPS endpoint. Deliveries carry <code>X-LTAPI-Signature</code> (<code>sha256=<hex></code> HMAC-SHA256 over the raw body), <code>X-LTAPI-Timestamp</code> and <code>X-LTAPI-Event</code>; up to 3 webhooks per key (<code>409 webhook_limit</code>); auto-disable after 25 consecutive failures. The signing secret is returned exactly once, on the 201. New schema <code>Webhook</code>.</li>
+<li><strong>Bare price ticks.</strong> <code>GET /matches/{matchId}/prices</code> (PRO): recent ticks of the mapped match-winner market without the market wrapper — <code>limit</code> caps at 500, <code>minutes</code> bounds the lookback, 404 when the match has no mapped market.</li>
+<li><strong>Tape coverage and sequence.</strong> <code>/history/matches</code> gains <code>?coverage=</code> (items are now <code>HistoryMatch</code> — <code>Match</code> plus a <code>tape</code> coverage object); <code>/history/matches/{matchId}</code> gains <code>?sequence=raw|clean</code>, documents that it works on a LIVE match, and its <code>meta</code> now carries <code>coverage</code>, <code>point_source</code>, <code>raw_rows</code>, <code>unique_states</code> and <code>sequence</code>. Tape rows are now the explicit <code>HistoryTapeRow</code> (null <code>timestamp</code> marks a reconstructed row). New schemas <code>Coverage</code>, <code>HistoryMatch</code>, <code>HistoryTapeRow</code>.</li>
+<li><strong>Rankings packages.</strong> <code>/history/packages</code> and <code>/history/packages/{period}</code> gain <code>?kind=tape|rankings</code> (default <code>tape</code>) and the listing gains <code>?year=YYYY</code> for the year-archive view; <code>HistoryPackage</code> documents the <code>kind</code> field and that JSONL is one line per match while CSV is one row per point.</li>
+<li><strong>List meta.</strong> <code>ListMeta</code> now declares <code>total</code> (nullable — null when the set cannot be counted cheaply) and <code>has_more</code> (page on this, not on <code>count</code>).</li>
+<li><strong>Error hints.</strong> <code>Error</code> now declares <code>detail</code> (human-readable explanation) and <code>allowed</code> (accepted values on a rejected enumerated parameter).</li>
+<li><strong>Price provenance.</strong> <code>Price</code> gains <code>price_source</code> and <code>synthetic</code>, so a quote synthesised from mid is never mistaken for a live order book.</li>
+<li><strong>Profile provenance.</strong> The <code>Analysis</code> profile (and <code>ModelProfile</code>) gains <code>stage</code> (<code>pregame</code> | <code>live</code> | null = unknown), <code>model_version</code>, and — on <code>Analysis</code> — <code>input_state</code>, the score an in-play forecast actually saw.</li>
+<li><strong>CORS documented.</strong> <code>info.description</code> now states that the REST surface sends <code>Access-Control-Allow-Origin: *</code> (GET/OPTIONS, no credentials mode) and that a FREE key in browser code is acceptable while a paid key belongs server-side.</li>
+<li><strong>WebSocket subscribe frame.</strong> <code>info.description</code> now shows the full subscribe frame keys — <code>action</code>, <code>topics</code> and the optional <code>signals</code> list.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><code>info.version</code> is now <code>1.1.0</code>.</li>
+<li><code>GET /matches/{matchId}/score</code> documents that it is a point-in-time snapshot, pointing at <code>/history/matches/{matchId}?sequence=clean</code> for the sequence of states and <code>/matches/{matchId}/statistics</code> for in-play statistics.</li>
+</ul>
+<p>The entries below were previously listed as Unreleased and ship in this release.</p>
+<h3>Added</h3>
+<ul>
+<li><strong>History endpoints documented.</strong> <code>/history/matches/{matchId}</code> (the full point-by-point tape with the per-point model win-probability — BASIC, or Historical Data API Starter+), <code>/history/packages</code> and <code>/history/packages/{period}</code> (pre-built monthly bulk downloads, manifest + <code>?format=jsonl|csv</code> — PRO, Historical Data API Pro+, or a one-off package pass), plus the <code>from</code>/<code>to</code> date-range filter and the 400/403 responses on <code>/history/matches</code>. New schemas <code>HistoryTape</code>, <code>ModelProfile</code>, <code>HistoryPackage</code>. Purely additive.</li>
+<li><strong>Concrete plan deltas.</strong> <code>info.description</code> now states exactly what each tier adds over the one below it, with its rate limits (FREE 30/min · 1,000/day; BASIC 60/min · 10,000/day; PRO 300/min · 100,000/day; ULTRA 600/min · 500,000/day), and documents the standalone Historical Data API plans (Starter / Pro / Business / one-off passes). <code>/matches</code> documents that <code>status=completed</code> requires BASIC — the gate was always enforced, just not written down.</li>
+<li><strong>WebSocket break-point signals.</strong> The <code>/ws</code> subscribe frame now documents an optional <code>signals</code> array; naming <code>break_point</code> opts the connection into two new frames — <code>break_point</code> (schema <code>BreakPoint</code>) the instant a break point arises and <code>break_point_result</code> (schema <code>BreakPointResult</code>) when it resolves. Both are ULTRA-only and purely additive: an existing subscriber that sends no <code>signals</code> sees exactly the frames it saw before. Documented in <code>info.description</code>, the two new component schemas, and the rendered reference.</li>
+<li><strong>FREE tier.</strong> Self-serve with no card at <a href="https://livetennisapi.com/subscribe/free">https://livetennisapi.com/subscribe/free</a> (30 req/min, 1,000 req/day). Covers live and upcoming matches, scores, players and fixtures — the six endpoints now tagged <code>(FREE)</code> in their summary. Purely additive: no endpoint, field, or type changed, and every paid tier keeps exactly the access it had. <code>/history/matches</code> remains BASIC; market prices stay PRO; analysis, live model fields and the WebSocket feed stay ULTRA.</li>
+<li><code>operationId</code> on all 12 operations, so generated clients get stable method names.</li>
+<li><code>info.contact</code>, <code>info.license</code> (MIT) and <code>info.termsOfService</code>.</li>
+<li>Redocly lint + a structural contract check in CI.</li>
+<li>Rendered reference published to <a href="https://docs.livetennisapi.com">https://docs.livetennisapi.com</a>.</li>
+</ul>
+<h3>Changed</h3>
+<ul>
+<li><strong>Rendered reference: correct plan labels + plans/FAQ pages.</strong> The tier parser labelled <code>GET /matches/{matchId}</code> as ULTRA (highest-tier-wins scan over "(FREE; +market PRO, +analysis ULTRA)") and did not recognise FREE at all, so every FREE endpoint showed "Plan required: —". It now takes the first tier named. The Plans section gained the FREE row, per-tier daily caps, the standalone Historical Data API plans (Starter / Pro / Business / one-off passes), the Break-point Alerts plans (Free vs Pro), and a FAQ ("How much data can I access on each plan?", "How far back does history go?", "What's in the point-by-point tape?"). Operation descriptions are now rendered, and <code>llms.txt</code> mirrors all of it.</li>
+<li><code>info.title</code> is now <code>Live Tennis API</code>, matching the product name.</li>
+</ul>
+
+
diff --git a/docs/changelog.html b/docs/changelog.html
index 5ac3a61..4a14519 100644
--- a/docs/changelog.html
+++ b/docs/changelog.html
@@ -4,9 +4,11 @@
Live Tennis API — Changelog
-
+
+
+
@@ -14,7 +16,7 @@
@@ -204,9 +206,20 @@
Live Tennis API · docs
Changelog
-
Every change to the specification, dated and versioned. The API surface is v1; changes within it are additive only. Current version 1.13.88, last change . Also as the full reference, OpenAPI YAML / JSON, and the dated facts file.
+
Every change to the specification, dated and versioned. The API surface is v1; changes within it are additive only. Current version 1.13.89, last change . Also as the full reference, OpenAPI YAML / JSON, and the dated facts file.
Corrections to published data values are not spec changes, so they are listed in the corrections log.
+
To follow new versions in a feed reader, subscribe to the Atom feed or the JSON Feed.
+
1.13.89
+
Added
+
+
The changelog as an Atom feed and a JSON Feed.changelog.atom (Atom 1.0) and changelog.json (JSON Feed 1.1) carry every entry of this changelog. Each one has the version, the date, the items of each section, a summary and the client library versions the entry names. Both are written from the same parse as changelog.html. The build fails when either feed breaks its format or differs from the page. An entry's id is its anchor on the page, so a rebuild never shows an old entry as new. The changelog page and the docs home link both feeds in their head, and llms.txt lists them. Nothing on the wire changed.
+
+
Fixed
+
+
The changelog page shows all 115 entries. Fifteen headings from 1.1.0 to 1.9.3 were written with an em dash. The page folded them into the 1.9.4 entry as plain text. Each is now its own entry, and the build stops on a heading it cannot read.
+
Bold and code that wrap onto a second line render in full. Six older entries showed raw asterisks and backticks there. Two links written in angle brackets showed the brackets. Both now render as on the rest of the page.
+
1.13.88
Changed
@@ -866,14 +879,14 @@
Added
1.9.7
Added
-
Match.outcome is now documented on the Match schema (it has been on every match since 2026-08-18): `completed | retired | walkover | default | abandoned | unresolved | null, derived from status + event_status`.
+
Match.outcome is now documented on the Match schema (it has been on every match since 2026-08-18): completed | retired | walkover | default | abandoned | unresolved | null, derived from status + event_status.
outcome: unresolved — a match every source lost before a result is closed unfinished and says so: score is the last state observed, winner is null, no result is asserted; it flips to completed with the proven final when an authority confirms the result. Until 2026-09-10 such closes were published as completed with a mid-match score, which an ULTRA customer read as a wrong result.
1.9.6
Added
stats.ratings_as_of on GET /players/{id} — the date (UTC) the ratings block was last refreshed. The current Elo is now refreshed every week for every rated player from the published rating tables (Tuesdays); before this the rating carried no date and a May capture read as this week's. Reported by a PRO customer building pre-match research.
-
GET /history/packages/{period}?format=corrections — a CSV keyed by match_id (`match_id, field, before, after, tournament_key, source, corrected_at`) on every package whose stored data was repaired after publication, listed in the manifest as format: corrections; 404 when a package has none. First use: surface corrections on the 2023-02 → 2024-12 tape packages (19,439 Challenger matches whose court surface a tournament-name rule had set to grass; repaired and rebuilt 2026-09-10). Reported by a History Pro customer.
+
GET /history/packages/{period}?format=corrections — a CSV keyed by match_id (match_id, field, before, after, tournament_key, source, corrected_at) on every package whose stored data was repaired after publication, listed in the manifest as format: corrections; 404 when a package has none. First use: surface corrections on the 2023-02 → 2024-12 tape packages (19,439 Challenger matches whose court surface a tournament-name rule had set to grass; repaired and rebuilt 2026-09-10). Reported by a History Pro customer.
Fixed
@@ -899,13 +912,13 @@
Fixed
Clarified
Price.mid is the venue's observed midpoint, never a model estimate; synthetic describes only bid/ask (mid +/- 0.005 when true); pre-match ticks are synthetic: true by design because the live book attaches at match start; timestamp is our observation clock (UTC), not a venue publication time.
-
-
## [1.9.3] — 2026-09-04
+
+
1.9.3
Clarified
Match.tournament_id null cases restated with the real causes and a measured rate: not in our tournament list (UTR), or a match whose name matched no single edition in that list. The API now resolves the latter at ingest by name + event type + date against the published fixtures, and the 2026 backlog was backfilled the same day (2,547 rows); ITF null rate went from about 25% to about 2.5%. No wire change.
-
-
## [1.9.2] — 2026-09-04
+
+
1.9.2
Added (spec catch-up — every one of these has been live on the API; the document lagged)
GET /matches?status=cancelled is now in the status enum, with what it covers (feed-cancelled, walkover with no stated winner, postponed and never played), its tier (BASIC or any History plan, like completed), and the one restriction: it pages with limit/offset and refuses updated_since.
@@ -915,41 +928,41 @@
Added (spec catch-up — every one of these has been live on the API; the do
Clarified
meta.has_more says how to enumerate a filtered set completely (page offset by limit until false) and that total is null on the terminal listings, so has_more is the only end-of-data signal there. Asked by a Basic customer on Discord; no wire change.
-
-
## [1.9.1] — 2026-09-02
+
+
1.9.1
Clarified
Player.ranking / Player.ranking_points now say what they are: the official singles ranking POSITION (the ordinal), ATP table for men and WTA table for women chosen by the player, refreshed ahead of each match the player has with us, null when no ranking is held, and always the CURRENT record even on historical matches (/rankings?as_of= is the as-of surface). Asked by a Basic customer; no wire change.
-
-
## [1.9.0] — 2026-09-02
+
+
1.9.0
Added
has_analysis and has_market on Match — two booleans on every row of GET /matches and on the detail, every tier. They carry the same fact the per-match endpoints answer 404 about, so a slate is filtered in one call instead of one 404 per match. Shipped to production 2026-09-02.
-
**Distinguishable absence on GET /matches/{matchId}/analysis and GET /markets/{matchId}/prices.** The status stays 404 (unchanged, and shipped clients branch on it), but the body now says which absence it is: {"error":"not_found"} for an id that does not exist, and {"error":"no_analysis"|"no_market","match_id":…,"coverage":"none","detail":…} for a real match we hold nothing for — the same coverage: "none" vocabulary /matches/{matchId}/statistics already uses in its 200. New error codes no_analysis, no_market documented on the Error schema.
-
-
## [1.8.0] — 2026-09-01
+
Distinguishable absence on GET /matches/{matchId}/analysis and GET /markets/{matchId}/prices. The status stays 404 (unchanged, and shipped clients branch on it), but the body now says which absence it is: {"error":"not_found"} for an id that does not exist, and {"error":"no_analysis"|"no_market","match_id":…,"coverage":"none","detail":…} for a real match we hold nothing for — the same coverage: "none" vocabulary /matches/{matchId}/statistics already uses in its 200. New error codes no_analysis, no_market documented on the Error schema.
+
+
1.8.0
Added
-
GET /history/archive/matches/{archiveId}/tape (getArchiveTape) — the RECONSTRUCTED 2013–2022 point-by-point tape for one archive result: the score sequence behind the published result: reconstructed point sequences, no capture timestamps. Shipped to production 2026-09-01; the spec was the last place it was missing. Tier: core ULTRA, **or any active History plan including Starter** (which opens it on a FREE core key). The archive RESULT stays on BASIC, so core BASIC and core PRO read the result and are refused the tape — 403 upgrade_required carrying capability: archive_tape.
+
GET /history/archive/matches/{archiveId}/tape (getArchiveTape) — the RECONSTRUCTED 2013–2022 point-by-point tape for one archive result: the score sequence behind the published result: reconstructed point sequences, no capture timestamps. Shipped to production 2026-09-01; the spec was the last place it was missing. Tier: core ULTRA, or any active History plan including Starter (which opens it on a FREE core key). The archive RESULT stays on BASIC, so core BASIC and core PRO read the result and are refused the tape — 403 upgrade_required carrying capability: archive_tape.
ArchiveTape schema. Same envelope as HistoryTape so one parser reads both halves of the tape product, with the differences that are true: match is the winner/loser-shaped archive row (rows are WINNER-FIRST, not p1/p2), profiles is always [], and meta carries archive_match_id rather than match_id — an archive id is not a match id, and passing one to the other's routes resolves a different, real record without erroring. Rows are HistoryTapeRow, reused unchanged.
-
**kind=archive_tape on /history/packages and /history/packages/{period}, and on HistoryPackage.kind.** Ten per-year bulk files, period 2013 through 2022, JSONL + CSV, all ready. The JSONL record is byte-for-byte what the per-match endpoint returns; the CSV is the flat per-row view keyed on archive_match_id and deliberately carries no timestamp, win_probability_p1 or danger column. A SEPARATE gate from the per-match tape: ULTRA, a History Pro/Business subscription, or an active one-off package window — a Starter grant reads tapes one at a time and does not download years of them. Core PRO carries neither gate.
+
kind=archive_tape on /history/packages and /history/packages/{period}, and on HistoryPackage.kind. Ten per-year bulk files, period 2013 through 2022, JSONL + CSV, all ready. The JSONL record is byte-for-byte what the per-match endpoint returns; the CSV is the flat per-row view keyed on archive_match_id and deliberately carries no timestamp, win_probability_p1 or danger column. A SEPARATE gate from the per-match tape: ULTRA, a History Pro/Business subscription, or an active one-off package window — a Starter grant reads tapes one at a time and does not download years of them. Core PRO carries neither gate.
Changed
-
**info.description states the provenance and the coverage, thin spots included.** Nobody watched these matches: timestamp, win_probability_p1 and danger are null on EVERY row and cannot be filled in later — the production table has no timestamp column at all, so this is a structural fact and not a convention. That is the opposite of the 2023→now tape, which is our own recording, where the rows we actually watched carry a real clock and most of them a model probability. The corpus is 97,901 matches / 14,340,663 rows, seasons 2013–2022 only: 977,903 archive results from 1968–2012 have no tape and never will. Coverage of the era is 19.3% overall and 44.9% of tour-level play — main-draw tour buckets 91.6–98.7%, ATP Challenger main 55.3% and Challenger qualifying 33.6%, slam QUALIFYING 16.0% (ATP) / 18.1% (WTA), ITF and futures effectively zero (25 of 116,575 ATP futures). Stated together, because a strong number published without its thin counterpart sells a corpus nobody has.
-
**ArchiveTape.meta.coverage names the measured split rather than glossing the label.** reconstructed_partial (3,594 matches) has TWO causes and does not say which: 3,038 are point-granular tapes of matches that genuinely stopped early (3,027 retirements, 11 defaults) — the larger cause — and the other 556 carry the label only because their tape is per-GAME. Read granularity and the match's own score, not the label, when the question is whether the whole match is there. granularity is point on 99.4% of the corpus; the 556 game tapes are 555 in 2013 and one in 2014.
+
info.description states the provenance and the coverage, thin spots included. Nobody watched these matches: timestamp, win_probability_p1 and danger are null on EVERY row and cannot be filled in later — the production table has no timestamp column at all, so this is a structural fact and not a convention. That is the opposite of the 2023→now tape, which is our own recording, where the rows we actually watched carry a real clock and most of them a model probability. The corpus is 97,901 matches / 14,340,663 rows, seasons 2013–2022 only: 977,903 archive results from 1968–2012 have no tape and never will. Coverage of the era is 19.3% overall and 44.9% of tour-level play — main-draw tour buckets 91.6–98.7%, ATP Challenger main 55.3% and Challenger qualifying 33.6%, slam QUALIFYING 16.0% (ATP) / 18.1% (WTA), ITF and futures effectively zero (25 of 116,575 ATP futures). Stated together, because a strong number published without its thin counterpart sells a corpus nobody has.
+
ArchiveTape.meta.coverage names the measured split rather than glossing the label.reconstructed_partial (3,594 matches) has TWO causes and does not say which: 3,038 are point-granular tapes of matches that genuinely stopped early (3,027 retirements, 11 defaults) — the larger cause — and the other 556 carry the label only because their tape is per-GAME. Read granularity and the match's own score, not the label, when the question is whether the whole match is there. granularity is point on 99.4% of the corpus; the 556 game tapes are 555 in 2013 and one in 2014.
Coverage says where its vocabulary stops. It describes the 2023+ tape; the archive tape reuses two of its five values and derives reconstructed_partial differently.
ArchiveMatch points at the tape, and GET /history/archive/matches/{archiveId} says the RESULT stays on BASIC either way.
The plain-HTML reference gains an FAQ entry — "Is there point-by-point data before 2023?" — and the history FAQ, llms.txt digest and README plan summary carry the same numbers, so an answer engine reading any one of them gets the corpus, the null clock and the coverage floor together.
info.version is now 1.8.0.
-
-
## [1.7.2] — 2026-08-23
+
+
1.7.2
Changed
status on Match now documents the lifecycle rule.completed is asserted only for a match we observed being played or whose match-winner market settled decisively; a closed market alone never finishes a match. cancelled with event_status: null means positive evidence the match was not played as scheduled (the market settled void) and no stated reason for why — outcome / fixture reason stay null rather than guessed, and the row upgrades to a completed walkover if a Walk Over with a stated winner lands later. Before 2026-08-23 a void market closure completed the match and stamped Finished; those rows were corrected. No field added or changed type — prose only.
info.version is now 1.7.2.
-
-
## [1.7.1] — 2026-08-19
+
+
1.7.1
Added
event_status_updated_at on Match. Nullable ISO-8601 UTC (Z) timestamp, right after event_status: when event_status last CHANGED — the instant WE recorded the walkover / retirement / cancellation / postponement / suspension (or its clearing), not when the tournament desk did. It is the field to measure our admin-status latency with. Bumps only on a change of value (a re-read of the same status never moves it; a clear back to null does). Null while event_status has never changed since the field was introduced (2026-08-19) — never backfilled, never guessed. Inherited by every schema built on Match (MatchDetail, HistoryMatch). Additive only — no existing field moved or changed type.
@@ -957,8 +970,8 @@
Added
Changed
info.version is now 1.7.1.
-
-
## [1.7.0] — 2026-08-18
+
+
1.7.0
Added
basis on GET /matches/{matchId}/points responses.live | reconstruction — which base served the page. Live capture is inherently partial: the stream serves what arrived while the match ran, and the match-closing point never streams live. For a COMPLETED match where a measured-complete recorded point sequence exists, the endpoint now serves that complete sequence instead, projected into the same point-frame shape — love-love opener through the match-closing point, seq contiguous 1..N, qualityclean — and says so with basis: "reconstruction". Every other case (every live match, and any completed match without a measured-complete recorded sequence) stays basis: "live" — the persisted stream rows, byte-for-byte what was served before. Completeness beats the partial live capture WHOLESALE — the two sequences are never interleaved (they share no key, so any merge would fabricate an order), the same rule ?points=complete follows on the history read. On projected frames ts is null on every row: the recorded sequence carries no per-point clock and none is fabricated (LivePoint.ts now documents this). after_seq pagination and seq dedup work identically on either basis, but the two bases are different sequences: after a match completes and flips to reconstruction, re-read from after_seq=0 rather than resuming a live cursor into it. Additive only — no existing field moved or changed type.
@@ -966,21 +979,21 @@
Added
Changed
info.version is now 1.7.0.
-
-
## [1.6.0] — 2026-08-18
+
+
1.6.0
Added
The three-valued draw field on Match.singles | doubles | null — same vocabulary as the new ?draw= filter, decided by one shared definition, so filter and field cannot disagree. Evidence order: a doubles-team participant proves doubles over any event type; otherwise the stated event type decides. Null means neither says anything — no stated event type, or a team tie (Davis Cup / BJK Cup / United Cup class), where one event type covers both singles and doubles rubbers and we will not guess which this is. Null is NOT singles. is_doubles stays for compatibility and is now documented as LOSSY, with its evidence order: false also covers "unknown", which is not a claim of singles — prefer draw, whose null says so honestly.
?draw=singles|doubles on four listings — GET /matches, /history/matches, /tournaments and /fixtures: the axis the tour filter deliberately collapses; the two compose (?tour=itf&draw=doubles is the ITF doubles slice). A row whose draw is null matches NEITHER value — null is an answer, not a wildcard. An unknown value is a 400 bad_draw with the allowed values. Two honesty notes carried into the spec: on /tournaments the answer comes from the event type alone (a tournament row has no participants to supply the doubles-team evidence matches have), and ?draw=doubles alone also returns mixed and exhibition doubles that no ?tour= value reaches.
GET /history/coverage (BASIC, or any Historical Data API plan) — the measured completeness rollup per tour × draw bucket: the numbers to read BEFORE choosing what to backtest, in one call instead of paging the archive. A prebuilt snapshot rebuilt nightly right after the completeness ledger reconverges — never computed at read time — so as_of (= built_at) dates every number and ledger_max_computed_at is the newest underlying per-match measurement; 503 coverage_unavailable before the first nightly build. Buckets are atp/wta/challenger/itf/juniors × singles/doubles plus other (team ties, mixed, exhibitions and matches with no stated event type — counted, never dropped, so the totals cannot lie); a bucket with zero completed matches is OMITTED, never emitted as zeros. Each bucket carries the five verifiable numbers (completed, any_tape, point_complete, complete_on_default_read, share), and method states the full measurement rule so every number carries its own definition. As of 2026-08-18: 174,393 completed matches; 91,318 point-complete on the best basis (52.4%) against 81,196 on the default read alone (46.6%); ITF singles 51.1% against ITF doubles 3.5% — do not extrapolate a completeness rate across a tour group.
-
**tape.starts_at_love and tape.computed_at on /history/matches items** (both nullable, present where enabled). starts_at_love follows the SAME best-basis rule as points_complete — true when either measured basis opens at the 0-0 state, so if any on-disk sequence opens at love you can obtain one that does. computed_at is when the ledger last measured the match: every field in the tape's measured block is a nightly-reconverged cache, and this is the as-of to quote with any of them. Null on either means the match has not been measured — never a guess.
+
tape.starts_at_love and tape.computed_at on /history/matches items (both nullable, present where enabled). starts_at_love follows the SAME best-basis rule as points_complete — true when either measured basis opens at the 0-0 state, so if any on-disk sequence opens at love you can obtain one that does. computed_at is when the ledger last measured the match: every field in the tape's measured block is a nightly-reconverged cache, and this is the as-of to quote with any of them. Null on either means the match has not been measured — never a guess.
The complete-basis addendum package files. An affected month's manifest may also list tennis_history_points_complete_<period>.jsonl.gz / .csv.gz (marked compression: gzip; bytes/sha256 cover the compressed bytes — exactly the download): for exactly the matches whose complete point sequence exists only as the on-disk reconstruction, the same tape ?points=complete serves (reconstruction contract: null timestamps and null model fields). The base files keep carrying every match's DEFAULT read — already the complete tape for most point-complete matches — and are never rewritten by the addendum; their bytes and sha256 values do not move.
Changed
info.version is now 1.6.0.
-
-
## [1.5.0] — 2026-08-17
+
+
1.5.0
Added
The as-of Elo tape.GET /rankings?system=elo (ULTRA, both modes) — our own computed Elo as point-in-time records: per-player as-of (also via archive_player for the ~62,000 rated people outside the roster, tour required there) and a leaderboard (tour required — the ATP and WTA walks are disjoint; exactly one surface, default overall; min_matches default 20 and activity_weeks default 52 / max 104, both echoed in meta.coverage.qualified). Four independent ladders (overall/hard/clay/grass), ATP from 1877, WTA from 1968, main tours plus challengers plus the futures tier. rating always; rank leaderboard-only; points always null; matches is the ladder-scoped count. meta.coverage gains newest_available, players_rated / players_linked, qualified and the model block (publication_lag_days: 14 — a week's results become effective 14 days after the week begins, so the failure direction is staleness, never look-ahead). The corpus head is frozen at 2026-05-25 and does not advance; elo is never included implicitly. The free current Elo on GET /players/{id} is a DIFFERENT scale (~150 Elo of per-player standard deviation apart) — never present a rating from one scale against a rating from the other. kind=elo yearly bulk packages join /history/packages.
@@ -994,8 +1007,8 @@
Changed
UTR honesty.system=utr is documented as observed from UTR's public search: withheld ratings are ABSENT, never 0; per-player as-of only — no listing by design (a table of only the players we happen to track would be a fake leaderboard); history since 2026-07-29; and the sweep's deliberate rankless / no-Elo (ITF-skewed) bias with its measured coverage (2026-08-17, players active in the last 60 days): ITF 931 of 5,606 (16.6%), Challenger 197 of 1,903 (10.4%), WTA 43 of 573 (7.5%), ATP 15 of 525 (2.9%).
*(recorded retroactively)* On 2026-08-16 the /ws-token description was expanded in place to teach the push-feed protocol (Centrifugo v2 connect/subscribe/heartbeat, token-per-reconnect, the SDK PushStream pointer) without a version bump; this entry is the changelog record of that change.
info.version is now 1.5.0.
-
-
## [1.4.0] — 2026-08-16
+
+
1.4.0
Added
Live per-point events.GET /matches/{matchId}/points (ULTRA) — the live per-point event stream of one match in seq order, paged with ?after_seq= (the resume cursor; up to 500 rows per page, continue on last_seq while has_more). It is the REST catch-up for the WebSocket point frames, which are best-effort with no replay. Per-match pbp_coverage states honestly whether the match has a true per-point stream (point) or only the snapshot score path (game — points empty, an answer rather than an error); per-point coverage is never promised slate-wide. New schemas MatchPoints, LivePoint and PointFrame; new error codes bad_after_seq and points_disabled (the surface switched off server-side).
@@ -1008,13 +1021,13 @@
Changed
Model win-probability copy tells the truth about density. The plan and FAQ copy no longer claims the tape carries the model win-probability "at every point" / "per point": the model stamp is best-effort and rides the rows where the model ran — meta.model_rows is the count, and null is the honest value elsewhere. The current-score snapshot is described as overwritten on every score commit, not "on every point".
HistoryTapeRow.point_winner is documented on ?points=complete rows as well as ?sequence=clean (there the served order IS point order).
info.version is now 1.4.0.
-
-
## [1.3.1] — 2026-08-07
+
+
1.3.1
Added
kind=rally bulk packages documented. The /history/packageskind enum gains rally — the charted rally corpus (shot-by-shot) as YEARLY exports, ULTRA, period = YYYY like archive. The kind has been live in the API (it shipped alongside the per-match rally endpoints); the spec simply did not list it. Additive only: the enum, the HistoryPackage.kind field, and the package-shape prose now match the served surface — kind accepts tape (default), rankings (ULTRA), rally (ULTRA) and archive.
-
-
## [1.3.0] — 2026-08-07
+
+
1.3.0
Added
Shot-level charting.GET /charting/players (ULTRA) — career serve/return profile: serve placement (deuce/ad × wide/body/T), return depth and outcomes, net and serve-and-volley conversion, clutch break/game/set-point serving and returning, winners and unforced errors by wing, rally-length and shot-direction tendencies, summed over the player's charted matches (name keyed, gender=men|women disambiguates, ambiguous fragments refused with candidates). GET /charting/matches/{chartingMatchId} (ULTRA) — every stat family for one charted match, both players, per-set split. Coverage is curated: 11,646 charted matches back to the 1960s, concentrated on the majors, not full-slate.
@@ -1028,8 +1041,8 @@
Changed
Tours. Coverage phrasing is now the five tours everywhere — ATP, WTA, Challenger, ITF and juniors — matching the tour filter enum (atp, wta, challenger, itf, juniors).
WebSocket copy. The subscribe frame is documented as {"topics":["live-scores"]} (+ optional signals) — the previously shown action key is not read by the server. Score frames are documented as carrying the ULTRA model fields (win_probability_p1, danger) live — a null means the model had no output for that point. The 2-connections-per- key limit is stated.
info.version is now 1.3.0.
-
-
## [1.2.0] — 2026-08-03
+
+
1.2.0
Added
Results archive (1968–2022). The history product now runs in two continuous, non-overlapping halves: the point-by-point tape (2023→now) and the results archive (1968–2022). Four new endpoints, all BASIC (or any Historical Data API plan): GET /history/archive/matches (winner/loser- shaped results — ATP and WTA, main draws, qualifying and the ITF/futures tiers, 1968 through 2022, with final score, seeds, ranks at the time; filters tour, name, from/to, round, level; its own id space; event_date is the TOURNAMENT START date; ends 2022-12-31, exactly where the tape begins), GET /history/archive/matches/{archiveId} (one result, with per-match serve statistics where the era recorded them — null before 1991 mostly, never synthesised), GET /history/archive/players (bios + career-high rank and the week it was first reached), and GET /history/archive/career (career aggregates — sums and ratios of sums only; serve.matches_with_stats states the serve-stat coverage). New schemas ArchiveMatch, ArchivePlayer, ArchivePlayerBio, ArchiveCareer.
@@ -1047,8 +1060,8 @@
Added
Changed
info.version is now 1.2.0; info.description states the two history halves and the updated tier deltas (tournaments on FREE, the archive family and /h2h on BASIC/History plans, the rankings listing on PRO, rally and per-player as-of rankings on ULTRA).
-
-
## [1.1.0] — 2026-08-02
+
+
1.1.0
Added
Usage endpoint.GET /usage (FREE — any tier): your own durable daily usage vs quota — tier, limits, today's calls and a 30-day history. Calls to it are quota-exempt. New schema Usage.
@@ -1076,11 +1089,11 @@
Added
History endpoints documented./history/matches/{matchId} (the full point-by-point tape with the per-point model win-probability — BASIC, or Historical Data API Starter+), /history/packages and /history/packages/{period} (pre-built monthly bulk downloads, manifest + ?format=jsonl|csv — PRO, Historical Data API Pro+, or a one-off package pass), plus the from/to date-range filter and the 400/403 responses on /history/matches. New schemas HistoryTape, ModelProfile, HistoryPackage. Purely additive.
Concrete plan deltas.info.description now states exactly what each tier adds over the one below it, with its rate limits (FREE 30/min · 1,000/day; BASIC 60/min · 10,000/day; PRO 300/min · 100,000/day; ULTRA 600/min · 500,000/day), and documents the standalone Historical Data API plans (Starter / Pro / Business / one-off passes). /matches documents that status=completed requires BASIC — the gate was always enforced, just not written down.
WebSocket break-point signals. The /ws subscribe frame now documents an optional signals array; naming break_point opts the connection into two new frames — break_point (schema BreakPoint) the instant a break point arises and break_point_result (schema BreakPointResult) when it resolves. Both are ULTRA-only and purely additive: an existing subscriber that sends no signals sees exactly the frames it saw before. Documented in info.description, the two new component schemas, and the rendered reference.
-
FREE tier. Self-serve with no card at <https://livetennisapi.com/subscribe/free> (30 req/min, 1,000 req/day). Covers live and upcoming matches, scores, players and fixtures — the six endpoints now tagged (FREE) in their summary. Purely additive: no endpoint, field, or type changed, and every paid tier keeps exactly the access it had. /history/matches remains BASIC; market prices stay PRO; analysis, live model fields and the WebSocket feed stay ULTRA.
+
FREE tier. Self-serve with no card at https://livetennisapi.com/subscribe/free (30 req/min, 1,000 req/day). Covers live and upcoming matches, scores, players and fixtures — the six endpoints now tagged (FREE) in their summary. Purely additive: no endpoint, field, or type changed, and every paid tier keeps exactly the access it had. /history/matches remains BASIC; market prices stay PRO; analysis, live model fields and the WebSocket feed stay ULTRA.
operationId on all 12 operations, so generated clients get stable method names.
info.contact, info.license (MIT) and info.termsOfService.
diff --git a/docs/changelog.json b/docs/changelog.json
new file mode 100644
index 0000000..be7f74d
--- /dev/null
+++ b/docs/changelog.json
@@ -0,0 +1,3613 @@
+{
+ "version": "https://jsonfeed.org/version/1.1",
+ "title": "Live Tennis API changelog",
+ "home_page_url": "https://docs.livetennisapi.com/changelog.html",
+ "feed_url": "https://docs.livetennisapi.com/changelog.json",
+ "description": "Every change to the Live Tennis API specification, dated and versioned, newest first.",
+ "icon": "https://docs.livetennisapi.com/icon-256.png",
+ "favicon": "https://docs.livetennisapi.com/icon-256.png",
+ "language": "en",
+ "authors": [
+ {
+ "name": "Live Tennis API",
+ "url": "https://livetennisapi.com"
+ }
+ ],
+ "items": [
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.89",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.89",
+ "title": "1.13.89: The changelog as an Atom feed and a JSON Feed, and 2 more changes",
+ "summary": "Added: The changelog as an Atom feed and a JSON Feed. Fixed: The changelog page shows all 115 entries. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
The changelog as an Atom feed and a JSON Feed.changelog.atom (Atom 1.0) and changelog.json (JSON Feed 1.1) carry every entry of this changelog. Each one has the version, the date, the items of each section, a summary and the client library versions the entry names. Both are written from the same parse as changelog.html. The build fails when either feed breaks its format or differs from the page. An entry's id is its anchor on the page, so a rebuild never shows an old entry as new. The changelog page and the docs home link both feeds in their head, and llms.txt lists them. Nothing on the wire changed.
\n
\n
Fixed
\n
\n
The changelog page shows all 115 entries. Fifteen headings from 1.1.0 to 1.9.3 were written with an em dash. The page folded them into the 1.9.4 entry as plain text. Each is now its own entry, and the build stops on a heading it cannot read.
\n
Bold and code that wrap onto a second line render in full. Six older entries showed raw asterisks and backticks there. Two links written in angle brackets showed the brackets. Both now render as on the rest of the page.
\n
",
+ "date_published": "2026-11-20T00:00:00Z",
+ "tags": [
+ "Added",
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.89",
+ "date": "2026-11-20",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "The changelog as an Atom feed and a JSON Feed. changelog.atom (Atom 1.0) and changelog.json (JSON Feed 1.1) carry every entry of this changelog. Each one has the version, the date, the items of each section, a summary and the client library versions the entry names. Both are written from the same parse as changelog.html. The build fails when either feed breaks its format or differs from the page. An entry's id is its anchor on the page, so a rebuild never shows an old entry as new. The changelog page and the docs home link both feeds in their head, and llms.txt lists them. Nothing on the wire changed."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Fixed",
+ "items": [
+ "The changelog page shows all 115 entries. Fifteen headings from 1.1.0 to 1.9.3 were written with an em dash. The page folded them into the 1.9.4 entry as plain text. Each is now its own entry, and the build stops on a heading it cannot read.",
+ "Bold and code that wrap onto a second line render in full. Six older entries showed raw asterisks and backticks there. Two links written in angle brackets showed the brackets. Both now render as on the rest of the page."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.88",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.88",
+ "title": "1.13.88: The client libraries page and the code samples name JavaScript 1.10.3",
+ "summary": "Changed: The client libraries page and the code samples name JavaScript 1.10.3.",
+ "content_html": "
Changed
\n
\n
The client libraries page and the code samples name JavaScript 1.10.3. It fixes requests in a browser, which all failed with "Illegal invocation" in 1.10.2. Its timeout now covers the response body as well as the headers. No JavaScript method changed, so every JavaScript sample keeps its call. The samples now link the source at the v1.10.3 tag. The page also links the live scores web starter, a static page built on this client. Nothing on the wire changed.
\n
",
+ "date_published": "2026-11-19T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.88",
+ "date": "2026-11-19",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The client libraries page and the code samples name JavaScript 1.10.3. It fixes requests in a browser, which all failed with \"Illegal invocation\" in 1.10.2. Its timeout now covers the response body as well as the headers. No JavaScript method changed, so every JavaScript sample keeps its call. The samples now link the source at the v1.10.3 tag. The page also links the live scores web starter, a static page built on this client. Nothing on the wire changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": [
+ {
+ "client": "JavaScript",
+ "package": "livetennisapi",
+ "version": "1.10.3"
+ },
+ {
+ "client": "JavaScript",
+ "package": "livetennisapi",
+ "version": "1.10.2"
+ }
+ ]
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.87",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.87",
+ "title": "1.13.87: The client libraries page and the code samples name R 0.2.1",
+ "summary": "Changed: The client libraries page and the code samples name R 0.2.1.",
+ "content_html": "
Changed
\n
\n
The client libraries page and the code samples name R 0.2.1. It now installs jsonlite, so a clean install can read responses. No R function changed, so every R sample keeps its call. The R samples now link the source at the v0.2.1 tag. Nothing on the wire changed.
\n
",
+ "date_published": "2026-11-17T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.87",
+ "date": "2026-11-17",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The client libraries page and the code samples name R 0.2.1. It now installs jsonlite, so a clean install can read responses. No R function changed, so every R sample keeps its call. The R samples now link the source at the v0.2.1 tag. Nothing on the wire changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": [
+ {
+ "client": "R",
+ "package": "livetennisapi",
+ "version": "0.2.1"
+ }
+ ]
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.86",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.86",
+ "title": "1.13.86: The client libraries page and the code samples name Swift 1.2.1",
+ "summary": "Changed: The client libraries page and the code samples name Swift 1.2.1.",
+ "content_html": "
Changed
\n
\n
The client libraries page and the code samples name Swift 1.2.1. It adds Match.hasAnalysis, Match.hasMarket and Match.eventStatusUpdatedAt. It also builds on Linux with Swift 5.10 again. No Swift method changed, so every Swift sample keeps its call. The Swift samples now link the source at the v1.2.1 tag. Nothing on the wire changed.
\n
",
+ "date_published": "2026-11-17T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.86",
+ "date": "2026-11-17",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The client libraries page and the code samples name Swift 1.2.1. It adds Match.hasAnalysis, Match.hasMarket and Match.eventStatusUpdatedAt. It also builds on Linux with Swift 5.10 again. No Swift method changed, so every Swift sample keeps its call. The Swift samples now link the source at the v1.2.1 tag. Nothing on the wire changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": [
+ {
+ "client": "Swift",
+ "package": "livetennisapi-swift",
+ "version": "1.2.1"
+ }
+ ]
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.85",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.85",
+ "title": "1.13.85: Code samples for the official clients",
+ "summary": "Added: Code samples for the official clients.",
+ "content_html": "
Added
\n
\n
Code samples for the official clients. Every operation now carries an x-codeSamples list. It starts with a curl call, then has one sample for each official client with a method for that operation. The samples cover Python 1.8.0, JavaScript 1.10.2, Go 1.3.0 and PHP 1.2.0. They also cover .NET 1.2.1, Dart 1.2.1, Swift 1.1.1 and R 0.2.0. The interactive reference lists them in the language picker of each operation. Of its own generated snippets it keeps only curl, which follows the values typed into Test Request. code-samples/sdk-methods.yaml records the method and the call of each client for each operation, read from the source of that version. scripts/build_code_samples.py writes the samples into the spec. The build fails when an operation has no entry in that file. It also fails when an entry names an operation the spec does not define, or when the samples are stale. Eleven operations have no client method yet, so they show curl only. Nothing on the wire changed.
\n
",
+ "date_published": "2026-11-17T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.85",
+ "date": "2026-11-17",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "Code samples for the official clients. Every operation now carries an x-codeSamples list. It starts with a curl call, then has one sample for each official client with a method for that operation. The samples cover Python 1.8.0, JavaScript 1.10.2, Go 1.3.0 and PHP 1.2.0. They also cover .NET 1.2.1, Dart 1.2.1, Swift 1.1.1 and R 0.2.0. The interactive reference lists them in the language picker of each operation. Of its own generated snippets it keeps only curl, which follows the values typed into Test Request. code-samples/sdk-methods.yaml records the method and the call of each client for each operation, read from the source of that version. scripts/build_code_samples.py writes the samples into the spec. The build fails when an operation has no entry in that file. It also fails when an entry names an operation the spec does not define, or when the samples are stale. Eleven operations have no client method yet, so they show curl only. Nothing on the wire changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": [
+ {
+ "client": "Python",
+ "package": "livetennisapi",
+ "version": "1.8.0"
+ },
+ {
+ "client": "JavaScript",
+ "package": "livetennisapi",
+ "version": "1.10.2"
+ },
+ {
+ "client": "Go",
+ "package": "github.com/livetennisapi/livetennisapi-go",
+ "version": "1.3.0"
+ },
+ {
+ "client": "PHP",
+ "package": "livetennisapi/livetennisapi",
+ "version": "1.2.0"
+ },
+ {
+ "client": ".NET",
+ "package": "LiveTennisApi",
+ "version": "1.2.1"
+ },
+ {
+ "client": "Dart",
+ "package": "livetennisapi",
+ "version": "1.2.1"
+ },
+ {
+ "client": "Swift",
+ "package": "livetennisapi-swift",
+ "version": "1.1.1"
+ },
+ {
+ "client": "R",
+ "package": "livetennisapi",
+ "version": "0.2.0"
+ }
+ ]
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.84",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.84",
+ "title": "1.13.84: A SportsML mapping page and test fixtures",
+ "summary": "Added: A SportsML mapping page and test fixtures.",
+ "content_html": "
Added
\n
\n
A SportsML mapping page and test fixtures.sportsml.html maps every field of the Score object to the IPTC SportsML 3.1 element or attribute it becomes. Where nothing fits, it says so. Each SportsML name links to IPTC's schema documentation. A worked example shows a match decided by a match tie-break as score states and as a SportsML document. examples/sportsml/ holds six synthetic matches under CC BY 4.0: a deciding-set tie-break, a match tie-break, no-ad games, a retirement, a walkover and a default. check_fixtures.py validates each document against the vendored IPTC schema and replays every point against the JSON states. The build fails when the Score schema gains a field the page does not map. The docs menu, the docs home, the sitemap, llms.txt and the README link the page. Nothing on the wire changed.
\n
",
+ "date_published": "2026-11-15T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.84",
+ "date": "2026-11-15",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "A SportsML mapping page and test fixtures. sportsml.html maps every field of the Score object to the IPTC SportsML 3.1 element or attribute it becomes. Where nothing fits, it says so. Each SportsML name links to IPTC's schema documentation. A worked example shows a match decided by a match tie-break as score states and as a SportsML document. examples/sportsml/ holds six synthetic matches under CC BY 4.0: a deciding-set tie-break, a match tie-break, no-ad games, a retirement, a walkover and a default. check_fixtures.py validates each document against the vendored IPTC schema and replays every point against the JSON states. The build fails when the Score schema gains a field the page does not map. The docs menu, the docs home, the sitemap, llms.txt and the README link the page. Nothing on the wire changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.83",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.83",
+ "title": "1.13.83: A client libraries page",
+ "summary": "Added: A client libraries page.",
+ "content_html": "
Added
\n
\n
A client libraries page.client-libraries.html lists the eight official language clients and seven packages built on them. Each row has the install command, the latest published version, the minimum runtime, the licence and the source repository. A second table says whether each package covers REST, the live stream, file export and MCP. A third shows which endpoint groups each language client reaches, read from the source of each published version. The page ends with the smallest working call in each language, copied from that library's README. scripts/check_client_versions.py reads the registries again and names any row that is behind. The build fails when the page names an operation the spec does not define. The docs menu, the docs home, the sitemap, llms.txt and the README link the page. Nothing on the wire changed.
\n
",
+ "date_published": "2026-11-12T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.83",
+ "date": "2026-11-12",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "A client libraries page. client-libraries.html lists the eight official language clients and seven packages built on them. Each row has the install command, the latest published version, the minimum runtime, the licence and the source repository. A second table says whether each package covers REST, the live stream, file export and MCP. A third shows which endpoint groups each language client reaches, read from the source of each published version. The page ends with the smallest working call in each language, copied from that library's README. scripts/check_client_versions.py reads the registries again and names any row that is behind. The build fails when the page names an operation the spec does not define. The docs menu, the docs home, the sitemap, llms.txt and the README link the page. Nothing on the wire changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.82",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.82",
+ "title": "1.13.82: The Fixture schema lists every field a fixture row carries",
+ "summary": "Fixed: The Fixture schema lists every field a fixture row carries.",
+ "content_html": "
Fixed
\n
\n
The Fixture schema lists every field a fixture row carries.reason, match_id, gender, is_qualifying, created_at, updated_at and pregame_forecasts have been on /fixtures rows since the dates in their descriptions. The schema named 13 of the 20 fields, so a generated client dropped the other seven. Match gains pregame_forecasts for the same reason. Schema only, the responses are unchanged.
\n
",
+ "date_published": "2026-11-12T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.82",
+ "date": "2026-11-12",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "The Fixture schema lists every field a fixture row carries. reason, match_id, gender, is_qualifying, created_at, updated_at and pregame_forecasts have been on /fixtures rows since the dates in their descriptions. The schema named 13 of the 20 fields, so a generated client dropped the other seven. Match gains pregame_forecasts for the same reason. Schema only, the responses are unchanged."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.81",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.81",
+ "title": "1.13.81: An export formats page",
+ "summary": "Added: An export formats page.",
+ "content_html": "
Added
\n
\n
An export formats page.export-formats.html explains how to export matches, fixtures, players, rankings and tournaments to CSV, JSON Lines or Parquet with livetennisapi-export 0.1.0. It says when each format fits and lists the six datasets with the endpoint and the plan each needs. It also covers paging, how nested fields become dotted columns and when the file is written. The read-back examples for pandas, DuckDB and R were run against files the package wrote. The build fails when the page names an endpoint, a column or a page size the spec does not have. The docs menu, the docs home, the sitemap and llms.txt link the page. Nothing on the wire changed.
\n
",
+ "date_published": "2026-11-11T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.81",
+ "date": "2026-11-11",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "An export formats page. export-formats.html explains how to export matches, fixtures, players, rankings and tournaments to CSV, JSON Lines or Parquet with livetennisapi-export 0.1.0. It says when each format fits and lists the six datasets with the endpoint and the plan each needs. It also covers paging, how nested fields become dotted columns and when the file is written. The read-back examples for pandas, DuckDB and R were run against files the package wrote. The build fails when the page names an endpoint, a column or a page size the spec does not have. The docs menu, the docs home, the sitemap and llms.txt link the page. Nothing on the wire changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": [
+ {
+ "client": "livetennisapi-export",
+ "package": "livetennisapi-export",
+ "version": "0.1.0"
+ }
+ ]
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.80",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.80",
+ "title": "1.13.80: An errors and rate limits page",
+ "summary": "Added: An errors and rate limits page.",
+ "content_html": "
Added
\n
\n
An errors and rate limits page.errors-and-rate-limits.html lists every non-2xx status the spec documents and every error code. Each row says what causes it, what to do and which operations name it. It also covers the three 429 bodies and the WebSocket close codes. Statuses, codes and per-plan limits are read from the spec at build time, and the build fails when the written half disagrees with it. The retry section matches the official clients, Python 1.8.0 and JS 1.10.2. They retry a per-minute 429, any 5xx and dropped connections, and wait Retry-After when it is sent. They never retry a daily 429 or abuse_throttled. The docs menu, the docs home, the sitemap and llms.txt link the page. Nothing on the wire changed.
\n
",
+ "date_published": "2026-11-09T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.80",
+ "date": "2026-11-09",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "An errors and rate limits page. errors-and-rate-limits.html lists every non-2xx status the spec documents and every error code. Each row says what causes it, what to do and which operations name it. It also covers the three 429 bodies and the WebSocket close codes. Statuses, codes and per-plan limits are read from the spec at build time, and the build fails when the written half disagrees with it. The retry section matches the official clients, Python 1.8.0 and JS 1.10.2. They retry a per-minute 429, any 5xx and dropped connections, and wait Retry-After when it is sent. They never retry a daily 429 or abuse_throttled. The docs menu, the docs home, the sitemap and llms.txt link the page. Nothing on the wire changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": [
+ {
+ "client": "Python",
+ "package": "livetennisapi",
+ "version": "1.8.0"
+ },
+ {
+ "client": "JavaScript",
+ "package": "livetennisapi",
+ "version": "1.10.2"
+ }
+ ]
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.79",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.79",
+ "title": "1.13.79: The super-tiebreak example in Score.games reads right, and 1 more change",
+ "summary": "Fixed: The super-tiebreak example in Score.games reads right. The gender parameter description lost a stray line.",
+ "content_html": "
Fixed
\n
\n
The super-tiebreak example in Score.games reads right.[[6,4,10],[4,6,5]] is 6-4 4-6 [10-5]. The description said 6-4 3-6 [10-5], which does not match the array. Text only, the field is unchanged.
\n
The gender parameter description lost a stray line. A run of error codes from another parameter had been pasted at its end. Text only, the parameter is unchanged.
\n
",
+ "date_published": "2026-11-09T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.79",
+ "date": "2026-11-09",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "The super-tiebreak example in Score.games reads right. [[6,4,10],[4,6,5]] is 6-4 4-6 [10-5]. The description said 6-4 3-6 [10-5], which does not match the array. Text only, the field is unchanged.",
+ "The gender parameter description lost a stray line. A run of error codes from another parameter had been pasted at its end. Text only, the parameter is unchanged."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.78",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.78",
+ "title": "1.13.78: Shot-level winners and errors: where they are and when they are withheld",
+ "summary": "Changed: Shot-level winners and errors: where they are and when they are withheld.",
+ "content_html": "
Changed
\n
\n
Shot-level winners and errors: where they are and when they are withheld. The family is present where the statistics source counts it: the 2026 Grand Slam main draws (Wimbledon and the US Open near complete, the Australian Open and Roland Garros filling as the completed-match statistics fetch walks back) and the odd regular-tour match. The earlier note that it had not been available since 2026-07-12 described live payloads of the regular tour and is replaced. A family the source filled with zeros is now withheld: a side that won 20 points or more with every by-shot winner and error at 0 is not measured, and the whole family is omitted, totals included. 33 of 33 Roland Garros 2026 qualifying payloads carried such zeros.
\n
",
+ "date_published": "2026-10-29T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.78",
+ "date": "2026-10-29",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "Shot-level winners and errors: where they are and when they are withheld. The family is present where the statistics source counts it: the 2026 Grand Slam main draws (Wimbledon and the US Open near complete, the Australian Open and Roland Garros filling as the completed-match statistics fetch walks back) and the odd regular-tour match. The earlier note that it had not been available since 2026-07-12 described live payloads of the regular tour and is replaced. A family the source filled with zeros is now withheld: a side that won 20 points or more with every by-shot winner and error at 0 is not measured, and the whole family is omitted, totals included. 33 of 33 Roland Garros 2026 qualifying payloads carried such zeros."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.77",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.77",
+ "title": "1.13.77: /fixtures and /matches tell one story, and 1 more change",
+ "summary": "Fixed: /fixtures and /matches tell one story. A bare digit in the schedule's status field is no longer a match in play.",
+ "content_html": "
Fixed
\n
\n
/fixtures and /matches tell one story. A fixture bound to a match (match_id set) is served with the match's status, in the fixture vocabulary, and the match's start_time, read at request time. The match row moves within minutes of a schedule change; the fixture row followed the 6-hourly refresher, so the two endpoints could disagree on the same match for hours. A customer found a Vienna match live at 09:00Z on one and upcoming at 12:30Z on the other. A vendor-stated cancelled or postponed on the fixture is kept. A null is_qualifying on the fixture takes the match's value when the match states one.
\n
A bare digit in the schedule's status field is no longer a match in play. The day-ahead listing carries 1 on matches that have not started. It was read as a set in play, so 298 fixtures for 29 October were published as live. The digit is live only when the schedule's own in-play flag says so. The stored rows were restated the same day.
\n
",
+ "date_published": "2026-10-29T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.77",
+ "date": "2026-10-29",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "/fixtures and /matches tell one story. A fixture bound to a match (match_id set) is served with the match's status, in the fixture vocabulary, and the match's start_time, read at request time. The match row moves within minutes of a schedule change; the fixture row followed the 6-hourly refresher, so the two endpoints could disagree on the same match for hours. A customer found a Vienna match live at 09:00Z on one and upcoming at 12:30Z on the other. A vendor-stated cancelled or postponed on the fixture is kept. A null is_qualifying on the fixture takes the match's value when the match states one.",
+ "A bare digit in the schedule's status field is no longer a match in play. The day-ahead listing carries 1 on matches that have not started. It was read as a set in play, so 298 fixtures for 29 October were published as live. The digit is live only when the schedule's own in-play flag says so. The stored rows were restated the same day."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.76",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.76",
+ "title": "1.13.76: X-LTAPI-Signature-V1 on every webhook delivery, and 2 more changes",
+ "summary": "Added: X-LTAPI-Signature-V1 on every webhook delivery. Fixed: The webhook events enum matches what registration accepts. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
X-LTAPI-Signature-V1 on every webhook delivery. The value is t=<unix seconds>,v1=<hex>, and the hex is HMAC-SHA256 of <t>.<raw request body> with your webhook secret. The timestamp is inside the signed string, so a captured delivery re-sent later under a new time no longer verifies. Each body also carries a top-level delivery_id, unique per delivery, to dedupe on. The POST /webhooks description lists the receiver checks: a 5-minute timestamp window, a constant-time compare and a dedupe on delivery_id. It also lists score_withdrawn among the X-LTAPI-Event values, which it had left out.
\n
\n
Fixed
\n
\n
The webhook events enum matches what registration accepts.POST /webhooks listed stoppages, which registration refuses with a 400, and the Webhook schema listed only score and break_point. Both now read score, break_point, point, status.
\n
\n
Deprecated
\n
\n
X-LTAPI-Signature, the body-only webhook signature. It does not cover the timestamp. We keep sending it unchanged until 2027-02-01, and deliveries sent from 00:00 UTC on that date do not carry it. Both headers use the same secret, so moving to the new header needs no re-registration.
\n
",
+ "date_published": "2026-10-28T00:00:00Z",
+ "tags": [
+ "Added",
+ "Fixed",
+ "Deprecated"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.76",
+ "date": "2026-10-28",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "X-LTAPI-Signature-V1 on every webhook delivery. The value is t=,v1=, and the hex is HMAC-SHA256 of . with your webhook secret. The timestamp is inside the signed string, so a captured delivery re-sent later under a new time no longer verifies. Each body also carries a top-level delivery_id, unique per delivery, to dedupe on. The POST /webhooks description lists the receiver checks: a 5-minute timestamp window, a constant-time compare and a dedupe on delivery_id. It also lists score_withdrawn among the X-LTAPI-Event values, which it had left out."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Fixed",
+ "items": [
+ "The webhook events enum matches what registration accepts. POST /webhooks listed stoppages, which registration refuses with a 400, and the Webhook schema listed only score and break_point. Both now read score, break_point, point, status."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Deprecated",
+ "items": [
+ "X-LTAPI-Signature, the body-only webhook signature. It does not cover the timestamp. We keep sending it unchanged until 2027-02-01, and deliveries sent from 00:00 UTC on that date do not carry it. Both headers use the same secret, so moving to the new header needs no re-registration."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.75",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.75",
+ "title": "1.13.75: Fields are described by what they hold",
+ "summary": "Changed: Fields are described by what they hold.",
+ "content_html": "
Changed
\n
\n
Fields are described by what they hold. Rounds, stoppages, per-point tags, statistics and surfaces now say what each value means and when it is null. Earlier entries in this changelog were reworded the same way. The 2013 to 2022 archive tape is described as reconstructed point sequences with no capture timestamps. Its point_source_detail value is unchanged. The charting endpoints describe the charted corpus by its coverage. Nothing on the wire changed.
\n
",
+ "date_published": "2026-10-28T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.75",
+ "date": "2026-10-28",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "Fields are described by what they hold. Rounds, stoppages, per-point tags, statistics and surfaces now say what each value means and when it is null. Earlier entries in this changelog were reworded the same way. The 2013 to 2022 archive tape is described as reconstructed point sequences with no capture timestamps. Its point_source_detail value is unchanged. The charting endpoints describe the charted corpus by its coverage. Nothing on the wire changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.74",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.74",
+ "title": "1.13.74: The point-by-point page describes the tape by what it holds",
+ "summary": "Changed: The point-by-point page describes the tape by what it holds.",
+ "content_html": "
Changed
\n
\n
The point-by-point page describes the tape by what it holds. The tape is the ordered score states we hold for a completed match, with coverage declared per match. The model's win probability sits only on rows that were observed live.
\n
",
+ "date_published": "2026-10-26T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.74",
+ "date": "2026-10-26",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The point-by-point page describes the tape by what it holds. The tape is the ordered score states we hold for a completed match, with coverage declared per match. The model's win probability sits only on rows that were observed live."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.73",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.73",
+ "title": "1.13.73: Source links point at GitHub again",
+ "summary": "Changed: Source links point at GitHub again.",
+ "content_html": "
Changed
\n
\n
Source links point at GitHub again. The github.com/livetennisapi organisation is public again, so every page, the README and llms.txt link each package's repository on GitHub beside its registry page. The Organization's sameAs names the GitHub org instead of the GitLab mirror.
\n
",
+ "date_published": "2026-10-26T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.73",
+ "date": "2026-10-26",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "Source links point at GitHub again. The github.com/livetennisapi organisation is public again, so every page, the README and llms.txt link each package's repository on GitHub beside its registry page. The Organization's sameAs names the GitHub org instead of the GitLab mirror."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.72",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.72",
+ "title": "1.13.72: duplicates_collapsed on GET /matches/{matchId}/points",
+ "summary": "Added: duplicates_collapsed on GET /matches/{matchId}/points.",
+ "content_html": "
Added
\n
\n
duplicates_collapsed on GET /matches/{matchId}/points. How many rows of the page repeated a point the match had already published and were left out. The writer no longer stores such repeats; the count covers rows written before that. last_seq still advances past them.
\n
",
+ "date_published": "2026-10-26T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.72",
+ "date": "2026-10-26",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "duplicates_collapsed on GET /matches/{matchId}/points. How many rows of the page repeated a point the match had already published and were left out. The writer no longer stores such repeats; the count covers rows written before that. last_seq still advances past them."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.71",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.71",
+ "title": "1.13.71: Every docs page head carries the X (Twitter) card tag",
+ "summary": "Fixed: Every docs page head carries the X (Twitter) card tag.",
+ "content_html": "
Fixed
\n
\n
Every docs page head carries the X (Twitter) card tag. The 2026-10-25 Ahrefs crawl listed twelve docs pages under "X card missing". The heads already carried the Open Graph tags, so the card tag names the card type and the rest falls back to Open Graph.
\n
",
+ "date_published": "2026-10-26T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.71",
+ "date": "2026-10-26",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "Every docs page head carries the X (Twitter) card tag. The 2026-10-25 Ahrefs crawl listed twelve docs pages under \"X card missing\". The heads already carried the Open Graph tags, so the card tag names the card type and the rest falls back to Open Graph."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.70",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.70",
+ "title": "1.13.70: index.html's JSON-LD description is its meta description",
+ "summary": "Fixed: index.html's JSON-LD description is its meta description.",
+ "content_html": "
Fixed
\n
\n
index.html's JSON-LD description is its meta description. The hub page's TechArticle carried a different sentence from its meta tag. The meta text is the one that fits the result-page budget, so the JSON-LD now repeats it. The generator's JSON-LD check covers index.html as well as the generated pages.
\n
",
+ "date_published": "2026-10-23T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.70",
+ "date": "2026-10-23",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "index.html's JSON-LD description is its meta description. The hub page's TechArticle carried a different sentence from its meta tag. The meta text is the one that fits the result-page budget, so the JSON-LD now repeats it. The generator's JSON-LD check covers index.html as well as the generated pages."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.69",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.69",
+ "title": "1.13.69: JSON-LD carries raw text, not HTML entities",
+ "summary": "Fixed: JSON-LD carries raw text, not HTML entities.",
+ "content_html": "
Fixed
\n
\n
JSON-LD carries raw text, not HTML entities. The players page's TechArticle description read player's inside its JSON-LD while the meta tag decoded to player's: the topic, reference and changelog templates formatted the block with the HTML escaper, and a script element's text is never decoded. Every JSON-LD block is now built as a dict and serialized once with json.dumps, </ escaped as <\\/. The reference and changelog descriptions in JSON-LD are the same text as their meta descriptions. The generator refuses a page whose JSON-LD does not parse, contains an entity, or whose TechArticle description differs from the decoded meta description.
\n
",
+ "date_published": "2026-10-23T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.69",
+ "date": "2026-10-23",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "JSON-LD carries raw text, not HTML entities. The players page's TechArticle description read player's inside its JSON-LD while the meta tag decoded to player's: the topic, reference and changelog templates formatted the block with the HTML escaper, and a script element's text is never decoded. Every JSON-LD block is now built as a dict and serialized once with json.dumps, escaped as <\\/. The reference and changelog descriptions in JSON-LD are the same text as their meta descriptions. The generator refuses a page whose JSON-LD does not parse, contains an entity, or whose TechArticle description differs from the decoded meta description."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.68",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.68",
+ "title": "1.13.68: The docs' Organization node matches the apex page's, and 1 more change",
+ "summary": "Changed: The docs' Organization node matches the apex page's. Titles fit 60 characters and descriptions 110 to 155.",
+ "content_html": "
Changed
\n
\n
The docs' Organization node matches the apex page's. Every docs page publishes the same Organization as livetennisapi.com: alternateName and the ten sameAs profiles (RapidAPI, Apify, API.market, Postman, GitLab, PyPI, the two npm packages, the docs site and X). Ahrefs flagged the docs node as incomplete while it carried only the name, url and logo.
\n
Titles fit 60 characters and descriptions 110 to 155. Three topic pages carry a shorter head title than their h1, the authentication page's description says what GET /usage is for, and a description the budget cuts no longer ends on a connective. The generator refuses a page outside these budgets, index.html included.
\n
",
+ "date_published": "2026-10-23T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.68",
+ "date": "2026-10-23",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The docs' Organization node matches the apex page's. Every docs page publishes the same Organization as livetennisapi.com: alternateName and the ten sameAs profiles (RapidAPI, Apify, API.market, Postman, GitLab, PyPI, the two npm packages, the docs site and X). Ahrefs flagged the docs node as incomplete while it carried only the name, url and logo.",
+ "Titles fit 60 characters and descriptions 110 to 155. Three topic pages carry a shorter head title than their h1, the authentication page's description says what GET /usage is for, and a description the budget cuts no longer ends on a connective. The generator refuses a page outside these budgets, index.html included."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.67",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.67",
+ "title": "1.13.67: /players/{playerId}/stats keeps a UTC instant as given, and 3 more changes",
+ "summary": "Fixed: /players/{playerId}/stats keeps a UTC instant as given. An omitted to is the request instant. Plus 2 more changes.",
+ "content_html": "
Fixed
\n
\n
/players/{playerId}/stats keeps a UTC instant as given.from and to are compared as instants, and window echoes what was applied. An instant inside a repeated daylight-saving hour was read through a local wall clock. to=2025-10-25T22:30:00Z came back as 23:30Z with from an hour late, from=22:30Z&to=23:15Z was refused and the reversed window was accepted. The same ordering applies to from and to on GET /matches and GET /history/matches.
\n
An omitted to is the request instant. It used to be rounded down to the minute, so a from a few seconds earlier answered 400 bad_date. Polls without to inside one minute still share a cached response, and its window.to is the first poll's instant.
\n
freshness.measured.complete and serve_return.matches_with_stats are one judgement. A deciding match tiebreak written as its points in the stamp and as one game in the served final, or the other way round, is the same finished set. The statistics endpoint compared the totals first and answered null one way and false the other, while the player aggregate counted the match complete. The field text now says what the judgement is.
\n
PlayerRecentMatch.score.games are the games the match object publishes. The recent scores go through the same projection as GET /matches/{matchId}, so a deciding match tiebreak the row keeps in points is rendered into the deciding-set games slot. Before this, the stored lists were read raw and a match that ended in the breaker showed one set fewer.
\n
",
+ "date_published": "2026-10-22T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.67",
+ "date": "2026-10-22",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "/players/{playerId}/stats keeps a UTC instant as given. from and to are compared as instants, and window echoes what was applied. An instant inside a repeated daylight-saving hour was read through a local wall clock. to=2025-10-25T22:30:00Z came back as 23:30Z with from an hour late, from=22:30Z&to=23:15Z was refused and the reversed window was accepted. The same ordering applies to from and to on GET /matches and GET /history/matches.",
+ "An omitted to is the request instant. It used to be rounded down to the minute, so a from a few seconds earlier answered 400 bad_date. Polls without to inside one minute still share a cached response, and its window.to is the first poll's instant.",
+ "freshness.measured.complete and serve_return.matches_with_stats are one judgement. A deciding match tiebreak written as its points in the stamp and as one game in the served final, or the other way round, is the same finished set. The statistics endpoint compared the totals first and answered null one way and false the other, while the player aggregate counted the match complete. The field text now says what the judgement is.",
+ "PlayerRecentMatch.score.games are the games the match object publishes. The recent scores go through the same projection as GET /matches/{matchId}, so a deciding match tiebreak the row keeps in points is rendered into the deciding-set games slot. Before this, the stored lists were read raw and a match that ended in the breaker showed one set fewer."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.66",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.66",
+ "title": "1.13.66: Player.tour describes a doubles team the way the API decides it, and 2 more changes",
+ "summary": "Fixed: Player.tour describes a doubles team the way the API decides it. Parameter tables show numeric bounds. Plus 1 more change.",
+ "content_html": "
Fixed
\n
\n
Player.tour describes a doubles team the way the API decides it. A team filed under a circuit of its own carries that circuit. Only a team filed under a level word or nothing falls back to the circuit both its members share, and it is null when they disagree or either is unknown. The 1.13.63 text put the fallback first. The field for singles or doubles is the match's draw, not its gender.
\n
Parameter tables show numeric bounds. Every parameter with a minimum or maximum now says so next to its default. /players/{playerId}/stats?last= reads as 1 to 50, where it used to show only the default of 20.
\n
The AI SDK guide says how check_api_status detects the plan. livetennisapi-ai 1.3.0 reads /usage and reports the key's own tier, so a FREE key with History access reports as FREE. The BASIC estimate applies only to the fallback probe, used when /usage cannot be read.
\n
",
+ "date_published": "2026-10-22T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.66",
+ "date": "2026-10-22",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "Player.tour describes a doubles team the way the API decides it. A team filed under a circuit of its own carries that circuit. Only a team filed under a level word or nothing falls back to the circuit both its members share, and it is null when they disagree or either is unknown. The 1.13.63 text put the fallback first. The field for singles or doubles is the match's draw, not its gender.",
+ "Parameter tables show numeric bounds. Every parameter with a minimum or maximum now says so next to its default. /players/{playerId}/stats?last= reads as 1 to 50, where it used to show only the default of 20.",
+ "The AI SDK guide says how check_api_status detects the plan. livetennisapi-ai 1.3.0 reads /usage and reports the key's own tier, so a FREE key with History access reports as FREE. The BASIC estimate applies only to the fallback probe, used when /usage cannot be read."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": [
+ {
+ "client": "livetennisapi-ai",
+ "package": "livetennisapi-ai",
+ "version": "1.3.0"
+ }
+ ]
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.65",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.65",
+ "title": "1.13.65: The AI SDK guide now pins livetennisapi-ai 1.3.0, which adds player_stats (PRO, with serve_return on ULTRA)",
+ "summary": "Changed: The AI SDK guide now pins livetennisapi-ai 1.3.0, which adds player_stats (PRO, with serve_return on ULTRA).",
+ "content_html": "
Changed
\n
\n
The AI SDK guide now pins livetennisapi-ai 1.3.0, which adds player_stats (PRO, with serve_return on ULTRA). The guide and llms.txt now say 25 tools, the same set as livetennisapi-mcp 1.6.1.
\n
",
+ "date_published": "2026-10-22T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.65",
+ "date": "2026-10-22",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The AI SDK guide now pins livetennisapi-ai 1.3.0, which adds player_stats (PRO, with serve_return on ULTRA). The guide and llms.txt now say 25 tools, the same set as livetennisapi-mcp 1.6.1."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": [
+ {
+ "client": "livetennisapi-ai",
+ "package": "livetennisapi-ai",
+ "version": "1.3.0"
+ },
+ {
+ "client": "livetennisapi-mcp",
+ "package": "livetennisapi-mcp",
+ "version": "1.6.1"
+ }
+ ]
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.64",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.64",
+ "title": "1.13.64: The AI SDK guide now pins livetennisapi-ai 1.2.1",
+ "summary": "Changed: The AI SDK guide now pins livetennisapi-ai 1.2.1.",
+ "content_html": "
Changed
\n
\n
The AI SDK guide now pins livetennisapi-ai 1.2.1. That release fixes check_api_status for FREE keys, so the known-issue note and its workaround are gone.
\n
",
+ "date_published": "2026-10-22T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.64",
+ "date": "2026-10-22",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The AI SDK guide now pins livetennisapi-ai 1.2.1. That release fixes check_api_status for FREE keys, so the known-issue note and its workaround are gone."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": [
+ {
+ "client": "livetennisapi-ai",
+ "package": "livetennisapi-ai",
+ "version": "1.2.1"
+ }
+ ]
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.63",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.63",
+ "title": "1.13.63: serve_return.matches_incomplete is documented, and 6 more changes",
+ "summary": "Fixed: serve_return.matches_incomplete is documented. serve_return.matches_with_stats says what it counts. Plus 5 more changes.",
+ "content_html": "
Fixed
\n
\n
serve_return.matches_incomplete is documented. The API already returns it. It counts matches whose measured statistics stop short of the final. Those matches are left out of every sum.
\n
serve_return.matches_with_stats says what it counts. It counts only matches whose measured statistics reach the final, and only those are summed.
\n
GET /players documents its 400 response. A restated_since that is not an ISO-8601 instant answers bad_restated_since with the Error body.
\n
Player.tour lists the values the API sends. They are atp, wta and null. The old text described capitalised and level values that have not been served since 2026-09-24.
\n
Field tables show nested fields. Inline objects are spelled out as dotted paths, such as sets.tiebreaks and serve.first_serve_in. A field typed by a named schema links to that schema's table.
\n
Changelog list items keep their continuation lines. An unindented line under a bullet rendered as a paragraph inside the list. That is invalid HTML.
\n
The AI SDK guide states a known issue with FREE keys. In livetennisapi-ai 1.1.0, check_api_status reports a FREE key as unreachable. The guide names the workaround until 1.2.1 is on npm.
\n
",
+ "date_published": "2026-10-22T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.63",
+ "date": "2026-10-22",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "serve_return.matches_incomplete is documented. The API already returns it. It counts matches whose measured statistics stop short of the final. Those matches are left out of every sum.",
+ "serve_return.matches_with_stats says what it counts. It counts only matches whose measured statistics reach the final, and only those are summed.",
+ "GET /players documents its 400 response. A restated_since that is not an ISO-8601 instant answers bad_restated_since with the Error body.",
+ "Player.tour lists the values the API sends. They are atp, wta and null. The old text described capitalised and level values that have not been served since 2026-09-24.",
+ "Field tables show nested fields. Inline objects are spelled out as dotted paths, such as sets.tiebreaks and serve.first_serve_in. A field typed by a named schema links to that schema's table.",
+ "Changelog list items keep their continuation lines. An unindented line under a bullet rendered as a paragraph inside the list. That is invalid HTML.",
+ "The AI SDK guide states a known issue with FREE keys. In livetennisapi-ai 1.1.0, check_api_status reports a FREE key as unreachable. The guide names the workaround until 1.2.1 is on npm."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": [
+ {
+ "client": "livetennisapi-ai",
+ "package": "livetennisapi-ai",
+ "version": "1.1.0"
+ },
+ {
+ "client": "livetennisapi-ai",
+ "package": "livetennisapi-ai",
+ "version": "1.2.1"
+ }
+ ]
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.62",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.62",
+ "title": "1.13.62: PlayerRecentMatch.opponent is nullable, and 6 more changes",
+ "summary": "Fixed: PlayerRecentMatch.opponent is nullable. PlayerRecentMatch.score.games values are nullable integers. Plus 5 more changes.",
+ "content_html": "
Fixed
\n
\n
PlayerRecentMatch.opponent is nullable. The endpoint returns null when the match row holds no opponent id. The schema said it was always an object.
\n
PlayerRecentMatch.score.games values are nullable integers. The per-set lists are passed through from the stored row as GET /matches/{matchId} serves them, and nothing on that path coerces the values.
\n
records.last_12_months now says what window it covers. It is the 365 days ending at to, whatever from says. With the default window it equals overall.
\n
The /players/{playerId}/stats 400 response references the Error schema. It names the five codes and says that bad_surface and bad_level list the accepted values in allowed.
\n
The Player schema gains seven fields the API already returns. They are ranking_status, ranking_checked_at, itf_ranking, profile_restated_at, profile_version, archive_player_id and itf_id (ULTRA only). GET /players gains the restated_since parameter that profile_restated_at points to. Nothing changed on the wire.
\n
llms.txt and the plans table name the new endpoint./players/{playerId}/stats is listed under PRO and its serve_return block under ULTRA.
\n
The players topic page no longer says /rankings per-player as-of records are PRO. The rank-ordered listing is PRO, and the as-of records are ULTRA.
\n
",
+ "date_published": "2026-10-22T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.62",
+ "date": "2026-10-22",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "PlayerRecentMatch.opponent is nullable. The endpoint returns null when the match row holds no opponent id. The schema said it was always an object.",
+ "PlayerRecentMatch.score.games values are nullable integers. The per-set lists are passed through from the stored row as GET /matches/{matchId} serves them, and nothing on that path coerces the values.",
+ "records.last_12_months now says what window it covers. It is the 365 days ending at to, whatever from says. With the default window it equals overall.",
+ "The /players/{playerId}/stats 400 response references the Error schema. It names the five codes and says that bad_surface and bad_level list the accepted values in allowed.",
+ "The Player schema gains seven fields the API already returns. They are ranking_status, ranking_checked_at, itf_ranking, profile_restated_at, profile_version, archive_player_id and itf_id (ULTRA only). GET /players gains the restated_since parameter that profile_restated_at points to. Nothing changed on the wire.",
+ "llms.txt and the plans table name the new endpoint. /players/{playerId}/stats is listed under PRO and its serve_return block under ULTRA.",
+ "The players topic page no longer says /rankings per-player as-of records are PRO. The rank-ordered listing is PRO, and the as-of records are ULTRA."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.61",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.61",
+ "title": "1.13.61: GET /players/{playerId}/stats (PRO; serve_return ULTRA), and 2 more changes",
+ "summary": "Added: GET /players/{playerId}/stats (PRO; serve_return ULTRA). height_cm and doubles_ranking on GET /players/{playerId}. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
GET /players/{playerId}/stats (PRO; serve_return ULTRA). One player's record over our own completed singles matches, 2023 onward. The default window is the 365 days ending now, and ?from=, ?to=, ?surface=, ?level= and ?last= narrow it. PRO gets records: W/L overall, over the last 12 months, and by surface, venue, level and season. It also carries the set-based record (tiebreaks, deciding sets, after winning or losing the first set, comebacks, sets won and lost), withdrawals given and received, and the last N matches with the score turned round to the player. Every count carries its sample size, and a percentage on an empty sample is null, not 0. Winners and scores are read the same way the match object publishes them, so this endpoint cannot disagree with GET /matches/{matchId}. ULTRA adds serve_return, the per-match statistics of /matches/{matchId}/statistics summed over the window. Each ratio is published as numerator, denominator and pct, with the number of matches behind it. Below ULTRA the key is absent. A doubles-team id answers 400 doubles_not_supported, and a merged or retired id answers 410 as /players/{playerId} does. For 1968–2022, /history/archive/career remains the endpoint.
\n
height_cm and doubles_ranking on GET /players/{playerId}. Both are on the single-player read only, on paid tiers, and null on a FREE key. height_cm comes from the results archive's record of the same person and is never inferred. doubles_ranking is {rank, points, observed_on}, the current position on the official ATP/WTA doubles table. It is null when the player is off the table or was last seen there more than 21 days ago. It is never copied into ranking, which stays the singles position. Nothing existing changed type or meaning.
\n
\n
Changed
\n
\n
The licence is named by its SPDX identifier.info.license now carries identifier: MIT in place of a URL. The old URL pointed at a GitHub page that no longer loads. No endpoint, field or type changed.
\n
",
+ "date_published": "2026-10-22T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.61",
+ "date": "2026-10-22",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /players/{playerId}/stats (PRO; serve_return ULTRA). One player's record over our own completed singles matches, 2023 onward. The default window is the 365 days ending now, and ?from=, ?to=, ?surface=, ?level= and ?last= narrow it. PRO gets records: W/L overall, over the last 12 months, and by surface, venue, level and season. It also carries the set-based record (tiebreaks, deciding sets, after winning or losing the first set, comebacks, sets won and lost), withdrawals given and received, and the last N matches with the score turned round to the player. Every count carries its sample size, and a percentage on an empty sample is null, not 0. Winners and scores are read the same way the match object publishes them, so this endpoint cannot disagree with GET /matches/{matchId}. ULTRA adds serve_return, the per-match statistics of /matches/{matchId}/statistics summed over the window. Each ratio is published as numerator, denominator and pct, with the number of matches behind it. Below ULTRA the key is absent. A doubles-team id answers 400 doubles_not_supported, and a merged or retired id answers 410 as /players/{playerId} does. For 1968–2022, /history/archive/career remains the endpoint.",
+ "height_cm and doubles_ranking on GET /players/{playerId}. Both are on the single-player read only, on paid tiers, and null on a FREE key. height_cm comes from the results archive's record of the same person and is never inferred. doubles_ranking is {rank, points, observed_on}, the current position on the official ATP/WTA doubles table. It is null when the player is off the table or was last seen there more than 21 days ago. It is never copied into ranking, which stays the singles position. Nothing existing changed type or meaning."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "The licence is named by its SPDX identifier. info.license now carries identifier: MIT in place of a URL. The old URL pointed at a GitHub page that no longer loads. No endpoint, field or type changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.60",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.60",
+ "title": "1.13.60: The hosted MCP endpoint, which this reference had never mentioned",
+ "summary": "Added: The hosted MCP endpoint, which this reference had never mentioned.",
+ "content_html": "
Added
\n
\n
The hosted MCP endpoint, which this reference had never mentioned. A remote Streamable HTTP server has run at https://mcp.livetennisapi.com/mcp for weeks. The only MCP line here named the self-run package, npx livetennisapi-mcp. A reader who wanted an LLM agent over this API was therefore told to install and host one. The string mcp.livetennisapi.com appeared zero times in the reference, zero times in llms.txt and zero times in the README. An engineer evaluating the API found the endpoint by going looking for it, which is what prompted this. A new MCP server section now sets the hosted and self-run routes side by side. The client-libraries table, llms.txt and the README name the endpoint too. Everything in the section was read off the live server on 2026-10-03, not taken from a release note. initialize reports serverInfo version 1.5.0. tools/list returns 24 tools, set-identical to the 24 the section lists by name. Both documented auth headers were called and both work: Authorization: Bearer twjp_... and X-API-Key: twjp_.... The rate headers read limit=300 on a key and limit=60 without one. The section states both, and states that the caller's own plan limit still applies underneath. That second sentence is load-bearing. The hosted cap sits above a free key's 30 a minute, so publishing it alone would read as a quota raise. With no key at all only check_api_status answers, and it returns the free-key URL. GET /mcp answers 405 "use POST", so the section says POST only. GET /health reports the running version. Every plan reaches MCP, free keys included, and each tool reaches exactly what that plan reaches. No price is quoted anywhere, because the endpoint takes the key a customer already has. No field, parameter or type changed, and no endpoint of the REST API changed.
\n
",
+ "date_published": "2026-10-03T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.60",
+ "date": "2026-10-03",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "The hosted MCP endpoint, which this reference had never mentioned. A remote Streamable HTTP server has run at https://mcp.livetennisapi.com/mcp for weeks. The only MCP line here named the self-run package, npx livetennisapi-mcp. A reader who wanted an LLM agent over this API was therefore told to install and host one. The string mcp.livetennisapi.com appeared zero times in the reference, zero times in llms.txt and zero times in the README. An engineer evaluating the API found the endpoint by going looking for it, which is what prompted this. A new MCP server section now sets the hosted and self-run routes side by side. The client-libraries table, llms.txt and the README name the endpoint too. Everything in the section was read off the live server on 2026-10-03, not taken from a release note. initialize reports serverInfo version 1.5.0. tools/list returns 24 tools, set-identical to the 24 the section lists by name. Both documented auth headers were called and both work: Authorization: Bearer twjp_... and X-API-Key: twjp_.... The rate headers read limit=300 on a key and limit=60 without one. The section states both, and states that the caller's own plan limit still applies underneath. That second sentence is load-bearing. The hosted cap sits above a free key's 30 a minute, so publishing it alone would read as a quota raise. With no key at all only check_api_status answers, and it returns the free-key URL. GET /mcp answers 405 \"use POST\", so the section says POST only. GET /health reports the running version. Every plan reaches MCP, free keys included, and each tool reaches exactly what that plan reaches. No price is quoted anywhere, because the endpoint takes the key a customer already has. No field, parameter or type changed, and no endpoint of the REST API changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.59",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.59",
+ "title": "1.13.59: ?is_qualifying= on /matches, /history/matches and /fixtures, and ?gender= documented on the same three, and 1 more change",
+ "summary": "Added: ?is_qualifying= on /matches, /history/matches and /fixtures, and ?gender= documented on the same three. Fixed: ?gender= was accepted and ignored on /matches?status=live and ?status=upcoming.",
+ "content_html": "
Added
\n
\n
?is_qualifying= on /matches, /history/matches and /fixtures, and ?gender= documented on the same three. The is_qualifying field has been published on every match and fixture since 2026-08-12 and this reference already called it "the only field that separates a qualifying draw from the main draw" and told callers to "filter accordingly" — while the parameter they would reach for did not exist and was accepted and ignored. Measured on prod before the change: /matches?status=completed&limit=50 returned the same fifty ids under is_qualifying=true, is_qualifying=false and no filter at all, every one of the fifty rows carrying is_qualifying: false, while the ?draw= control changed the page. The filter is applied in SQL before the page limit on status=live, upcoming, completed and cancelled and on the archive listing, so a deep crawl can page a single draw rather than filtering a full corpus client-side. It is three-valued like the field: a row whose flag is null matches neither true nor false and is returned only when the filter is omitted, because returning it under false would assert "main draw" on the strength of nothing. Coverage, measured 2026-10-02 over completed matches: the flag is stated on 98.9% of the last thirty days and 98.3% of 2026 outside the UTR circuit, which is where almost all unstated rows are (1,695 of the 1,831 in the last thirty days). Anything other than true/false is a 400 bad_is_qualifying. Asked for by a customer splitting qualifying from main draw for a match-prediction model. ?gender=men|women has worked on the match and archive listings since 2026-09-09 and was never documented here; it is now, with the same null contract — a mixed-doubles or team-tie row matches neither value. No field or type changed.
\n
\n
Fixed
\n
\n
?gender= was accepted and ignored on /matches?status=live and ?status=upcoming. Those two statuses are served from a short-lived slate cache keyed only on the status, and the gate that decides whether a request may use it never learned about gender, so a filtered request was answered with the whole slate under a 200. Measured on prod before the fix: status=live returned the same twenty-five rows for gender=men, gender=women and no filter — six of them women's matches and seven with no gender stated — and status=upcoming the same sixty rows of a hundred and forty-three under every value. status=completed and status=cancelled are not cached and answered correctly throughout (sixty of sixty, pure), so the filter itself was never wrong: the request did not reach it. The live listing is polled continuously, so the cache was effectively always warm and the filter was dropped on every call rather than occasionally. Present from 2026-09-09 to 2026-10-02. Callers who filtered the live or upcoming slate by gender and trusted the result were served the unfiltered slate and should re-read anything derived from it; every row carried a correct gender field throughout, so a client that filtered on the field rather than the parameter was unaffected.
\n
",
+ "date_published": "2026-10-02T00:00:00Z",
+ "tags": [
+ "Added",
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.59",
+ "date": "2026-10-02",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "?is_qualifying= on /matches, /history/matches and /fixtures, and ?gender= documented on the same three. The is_qualifying field has been published on every match and fixture since 2026-08-12 and this reference already called it \"the only field that separates a qualifying draw from the main draw\" and told callers to \"filter accordingly\" — while the parameter they would reach for did not exist and was accepted and ignored. Measured on prod before the change: /matches?status=completed&limit=50 returned the same fifty ids under is_qualifying=true, is_qualifying=false and no filter at all, every one of the fifty rows carrying is_qualifying: false, while the ?draw= control changed the page. The filter is applied in SQL before the page limit on status=live, upcoming, completed and cancelled and on the archive listing, so a deep crawl can page a single draw rather than filtering a full corpus client-side. It is three-valued like the field: a row whose flag is null matches neither true nor false and is returned only when the filter is omitted, because returning it under false would assert \"main draw\" on the strength of nothing. Coverage, measured 2026-10-02 over completed matches: the flag is stated on 98.9% of the last thirty days and 98.3% of 2026 outside the UTR circuit, which is where almost all unstated rows are (1,695 of the 1,831 in the last thirty days). Anything other than true/false is a 400 bad_is_qualifying. Asked for by a customer splitting qualifying from main draw for a match-prediction model. ?gender=men|women has worked on the match and archive listings since 2026-09-09 and was never documented here; it is now, with the same null contract — a mixed-doubles or team-tie row matches neither value. No field or type changed."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Fixed",
+ "items": [
+ "?gender= was accepted and ignored on /matches?status=live and ?status=upcoming. Those two statuses are served from a short-lived slate cache keyed only on the status, and the gate that decides whether a request may use it never learned about gender, so a filtered request was answered with the whole slate under a 200. Measured on prod before the fix: status=live returned the same twenty-five rows for gender=men, gender=women and no filter — six of them women's matches and seven with no gender stated — and status=upcoming the same sixty rows of a hundred and forty-three under every value. status=completed and status=cancelled are not cached and answered correctly throughout (sixty of sixty, pure), so the filter itself was never wrong: the request did not reach it. The live listing is polled continuously, so the cache was effectively always warm and the filter was dropped on every call rather than occasionally. Present from 2026-09-09 to 2026-10-02. Callers who filtered the live or upcoming slate by gender and trusted the result were served the unfiltered slate and should re-read anything derived from it; every row carried a correct gender field throughout, so a client that filtered on the field rather than the parameter was unaffected."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.58",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.58",
+ "title": "1.13.58: The archived tape's accept clock now begins 2026-08-11, six weeks earlier than 1.13.57 published, because the missing values were recovered rather than written…",
+ "summary": "Changed: The archived tape's accept clock now begins 2026-08-11, six weeks earlier than 1.13.57 published, because the missing values were recovered rather than written off.",
+ "content_html": "
Changed
\n
\n
The archived tape's accept clock now begins 2026-08-11, six weeks earlier than 1.13.57 published, because the missing values were recovered rather than written off. 1.13.57 measured the boundary honestly at 2026-09-23, the day history_archived_scores.content_changed_at landed as a catalogue-only column with no backfill. What that measurement did not say is that the values were still reachable: for every archive row copied before the column existed, the live row it was copied from may still be inside the 90-day retention, and that row holds the value the archiver would itself have written. A backfill carried them across on 2026-10-01 in six bounded tranches, 1,413,471 accept clocks and 1,393,458 accept counters, inside 1,538,176 rows over 14,491 matches, and left fillable=0. Nothing is derived, rounded or taken from a neighbouring row, and a row whose live row had already been trimmed stays null for good. Measured after the run: the archive carries the clock on 100.0% of every day from 2026-08-11 to 2026-10-01, 92.5% of 2026-08-10 (the live column's own first instant is 2026-08-10T08:07:42Z) and 0.0% of every day before it, which is 1,721,102 of 4,663,181 archived observed rows against 307,631 before. HistoryTapeRow.sequence, the accept counter, has its own later floor because live_scores.match_seq landed after the clock: nothing to 2026-08-13, 68.2% of 08-14, 99.3% of 08-15, then 95.9% or more of every day from 2026-08-16 and 100% from 2026-09-01. Two boundaries, not one. received_at could not move at all and the reference now says so: its live column landed 2026-09-24, a day AFTER the archive column, so every live row carrying an arrival clock was archived with it and the recoverable count is zero. Wire-verified both sides: match 2168 (26 April 2026), before the boundary, still answers 131 observed rows with timestamp on all 131 and accepted_at on none; match 189679's archived copy carried the clock on 0 of its 171 observed rows before the run and on 171 of 171 after it. The practical consequence for a timing study is that the clock no longer disappears when a match ages out of live retention. No field, endpoint or type changed.
\n
",
+ "date_published": "2026-10-01T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.58",
+ "date": "2026-10-01",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The archived tape's accept clock now begins 2026-08-11, six weeks earlier than 1.13.57 published, because the missing values were recovered rather than written off. 1.13.57 measured the boundary honestly at 2026-09-23, the day history_archived_scores.content_changed_at landed as a catalogue-only column with no backfill. What that measurement did not say is that the values were still reachable: for every archive row copied before the column existed, the live row it was copied from may still be inside the 90-day retention, and that row holds the value the archiver would itself have written. A backfill carried them across on 2026-10-01 in six bounded tranches, 1,413,471 accept clocks and 1,393,458 accept counters, inside 1,538,176 rows over 14,491 matches, and left fillable=0. Nothing is derived, rounded or taken from a neighbouring row, and a row whose live row had already been trimmed stays null for good. Measured after the run: the archive carries the clock on 100.0% of every day from 2026-08-11 to 2026-10-01, 92.5% of 2026-08-10 (the live column's own first instant is 2026-08-10T08:07:42Z) and 0.0% of every day before it, which is 1,721,102 of 4,663,181 archived observed rows against 307,631 before. HistoryTapeRow.sequence, the accept counter, has its own later floor because live_scores.match_seq landed after the clock: nothing to 2026-08-13, 68.2% of 08-14, 99.3% of 08-15, then 95.9% or more of every day from 2026-08-16 and 100% from 2026-09-01. Two boundaries, not one. received_at could not move at all and the reference now says so: its live column landed 2026-09-24, a day AFTER the archive column, so every live row carrying an arrival clock was archived with it and the recoverable count is zero. Wire-verified both sides: match 2168 (26 April 2026), before the boundary, still answers 131 observed rows with timestamp on all 131 and accepted_at on none; match 189679's archived copy carried the clock on 0 of its 171 observed rows before the run and on 171 of 171 after it. The practical consequence for a timing study is that the clock no longer disappears when a match ages out of live retention. No field, endpoint or type changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.57",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.57",
+ "title": "1.13.57: HistoryTapeRow.accepted_at was documented as covering the whole observed corpus, and on an archived tape it begins 2026-09-23",
+ "summary": "Fixed: HistoryTapeRow.accepted_at was documented as covering the whole observed corpus, and on an archived tape it begins 2026-09-23.",
+ "content_html": "
Fixed
\n
\n
HistoryTapeRow.accepted_at was documented as covering the whole observed corpus, and on an archived tape it begins 2026-09-23. The field is the one this reference tells a timing study to use, and its description read "It rides the whole observed corpus, so a study that spans 2026-09-24 must use this one and not received_at", while received_at said "THIS CLOCK HAS A START DATE AND THE OTHER TWO DO NOT". Both sentences were wrong about accepted_at. Measured over the whole archived observed corpus: 307,512 of 4,659,024 rows carry it, which is 0.0% of every month to August 2026 (286 rows in all of August), 27.3% of September and 100% of October. Wire-verified on match 2168 (26 April 2026), answered with meta.from_archive: true: 131 observed rows, timestamp on all 131 and accepted_at on none. The measurement quoted in 1.13.5x — match 189679, accepted_at on all 171 observed rows — is correct and does not generalise, because that match is still inside live retention and is answered from its live rows (from_archive: false); its own archived copy carries the clock on 0 of 367 rows. So coverage depends on which copy answers the read, meta.from_archive is how you tell, and a null here is a boundary of the record rather than a missing value. timestamp is the only one of the three clocks with no start date. Found while answering a customer who asked for "a timestamp precise enough to line up with exchange quotes" — the exact question the wrong sentence was there to answer. No field, endpoint or type changed.
\n
",
+ "date_published": "2026-10-01T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.57",
+ "date": "2026-10-01",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "HistoryTapeRow.accepted_at was documented as covering the whole observed corpus, and on an archived tape it begins 2026-09-23. The field is the one this reference tells a timing study to use, and its description read \"It rides the whole observed corpus, so a study that spans 2026-09-24 must use this one and not received_at\", while received_at said \"THIS CLOCK HAS A START DATE AND THE OTHER TWO DO NOT\". Both sentences were wrong about accepted_at. Measured over the whole archived observed corpus: 307,512 of 4,659,024 rows carry it, which is 0.0% of every month to August 2026 (286 rows in all of August), 27.3% of September and 100% of October. Wire-verified on match 2168 (26 April 2026), answered with meta.from_archive: true: 131 observed rows, timestamp on all 131 and accepted_at on none. The measurement quoted in 1.13.5x — match 189679, accepted_at on all 171 observed rows — is correct and does not generalise, because that match is still inside live retention and is answered from its live rows (from_archive: false); its own archived copy carries the clock on 0 of 367 rows. So coverage depends on which copy answers the read, meta.from_archive is how you tell, and a null here is a boundary of the record rather than a missing value. timestamp is the only one of the three clocks with no start date. Found while answering a customer who asked for \"a timestamp precise enough to line up with exchange quotes\" — the exact question the wrong sentence was there to answer. No field, endpoint or type changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.56",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.56",
+ "title": "1.13.56: An exact player name that sat inside a longer name could not be asked about on any of the three name endpoints",
+ "summary": "Fixed: An exact player name that sat inside a longer name could not be asked about on any of the three name endpoints.",
+ "content_html": "
Fixed
\n
\n
An exact player name that sat inside a longer name could not be asked about on any of the three name endpoints. Every name lookup is a substring match, so there was no way to say "exactly this one", and a person whose name is a prefix of somebody else's was unreachable: GET /h2h?p1=Carlos%20Alcaraz answered 400 ambiguous_name listing Carlos Alcaraz and Carlos Alcaraz Gonzalez, with no string that could pick between them. Measured on the live roster and corpus: 14 of the 1,177 ranked players on /h2h, 21 of 1,500 distinct archive names on /history/archive/career and 4 of 1,739 charted names on /charting/players, the men's world #2 and #3 among them. Passing a name exactly now selects that person on all three. A name given in part is still a fragment and still refuses with the candidate list, so Sinner continues to list Jannik and Martin rather than guessing, and a single name charted under both genders on /charting/players still needs gender. The refusal also reports the candidates it would have shown before the exact match was applied, so a hint never gets shorter than it was. Verified on the wire: Alcaraz against Sinner returns 18 meetings 11-7 with the surface split and the ULTRA aggregates. No field, endpoint, parameter or error code changed.
\n
",
+ "date_published": "2026-10-01T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.56",
+ "date": "2026-10-01",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "An exact player name that sat inside a longer name could not be asked about on any of the three name endpoints. Every name lookup is a substring match, so there was no way to say \"exactly this one\", and a person whose name is a prefix of somebody else's was unreachable: GET /h2h?p1=Carlos%20Alcaraz answered 400 ambiguous_name listing Carlos Alcaraz and Carlos Alcaraz Gonzalez, with no string that could pick between them. Measured on the live roster and corpus: 14 of the 1,177 ranked players on /h2h, 21 of 1,500 distinct archive names on /history/archive/career and 4 of 1,739 charted names on /charting/players, the men's world #2 and #3 among them. Passing a name exactly now selects that person on all three. A name given in part is still a fragment and still refuses with the candidate list, so Sinner continues to list Jannik and Martin rather than guessing, and a single name charted under both genders on /charting/players still needs gender. The refusal also reports the candidates it would have shown before the exact match was applied, so a hint never gets shorter than it was. Verified on the wire: Alcaraz against Sinner returns 18 meetings 11-7 with the surface split and the ULTRA aggregates. No field, endpoint, parameter or error code changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.55",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.55",
+ "title": "1.13.55: ITF World Tennis Tour doubles began stating outcome, and the reference still said no doubles draw carries either field",
+ "summary": "Fixed: ITF World Tennis Tour doubles began stating outcome, and the reference still said no doubles draw carries either field.",
+ "content_html": "
Fixed
\n
\n
ITF World Tennis Tour doubles began stating outcome, and the reference still said no doubles draw carries either field. The 1.13.42 note measured 454 completed doubles matches and found no tagged row. That was true when it was written. The first doubles row carried a tag on 2026-09-23, and by 2026-09-29 GET /matches/{id}/points published outcome on 104 of 596 completed ITF doubles matches, 17%, against 0 of 371 for the ATP, WTA and Challenger doubles draws in the same window. The vocabulary is ace and double fault, as on the ITF singles draws. serve is still unstated on every doubles draw, measured 0, and that sentence is unchanged. The cause sat before the tagger: a dedup change let a doubles fixture be adopted and oriented by its member sets instead of being refused, so the ITF per-point join reached rows it had never been able to match. 17% falls between "almost none" and "a clear majority", so the description states it as a measured fraction and sends the reader to enrichment per match instead of to the draw. WTA main-tour qualifying has carried that treatment since 1.13.42. Found by the every-sweep tag-coverage check. No field, endpoint or behaviour changed.
\n
",
+ "date_published": "2026-09-29T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.55",
+ "date": "2026-09-29",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "ITF World Tennis Tour doubles began stating outcome, and the reference still said no doubles draw carries either field. The 1.13.42 note measured 454 completed doubles matches and found no tagged row. That was true when it was written. The first doubles row carried a tag on 2026-09-23, and by 2026-09-29 GET /matches/{id}/points published outcome on 104 of 596 completed ITF doubles matches, 17%, against 0 of 371 for the ATP, WTA and Challenger doubles draws in the same window. The vocabulary is ace and double fault, as on the ITF singles draws. serve is still unstated on every doubles draw, measured 0, and that sentence is unchanged. The cause sat before the tagger: a dedup change let a doubles fixture be adopted and oriented by its member sets instead of being refused, so the ITF per-point join reached rows it had never been able to match. 17% falls between \"almost none\" and \"a clear majority\", so the description states it as a measured fraction and sends the reader to enrichment per match instead of to the draw. WTA main-tour qualifying has carried that treatment since 1.13.42. Found by the every-sweep tag-coverage check. No field, endpoint or behaviour changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.54",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.54",
+ "title": "1.13.54: The two broadcast surfaces, live since 2026-09-24 and documented nowhere a customer could read, and 1 more change",
+ "summary": "Added: The two broadcast surfaces, live since 2026-09-24 and documented nowhere a customer could read. The 1 Hz polling contract, which is the part a live scoreboard gets wrong.",
+ "content_html": "
Added
\n
\n
The two broadcast surfaces, live since 2026-09-24 and documented nowhere a customer could read.GET /broadcast/match/{matchId} and GET /broadcast/live have served every ULTRA key for four days and appeared in no published reference: the write-up went into the tennis repository, which nothing serves. Both are now specified, with the 54-field BroadcastObject schema in the same order the API returns it, verified field by field against a live response rather than transcribed from the source. The object exists because a graphics template binds each field to a fixed path, so three properties are contractual rather than incidental and are documented as such: it is FLAT, every key is PRESENT on every read (null, never missing — a missing key breaks a binding), and the score is already in DISPLAY form (p1_points reads "40" or "AD", set_line is one string, each set has its own column). It is a projection of what the API already publishes plus the break-, set- and match-point flags the scoring pipeline derives; nothing in it is a new fact and nothing is guessed.
\n
The 1 Hz polling contract, which is the part a live scoreboard gets wrong. Every response carries a strong ETag and Cache-Control: no-cache, max-age=1, so If-None-Match turns an unchanged state into a 304 with no body; ULTRA's 600-a-minute burst window means ten single-match pollers at 1 Hz fit on one key, and /broadcast/live stays one request however many matches are live. A new state is detected on sequence, never on served_at, which changes on every read — stated on both routes because polling on served_at is the mistake that looks like it works.
\n
",
+ "date_published": "2026-09-28T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.54",
+ "date": "2026-09-28",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "The two broadcast surfaces, live since 2026-09-24 and documented nowhere a customer could read. GET /broadcast/match/{matchId} and GET /broadcast/live have served every ULTRA key for four days and appeared in no published reference: the write-up went into the tennis repository, which nothing serves. Both are now specified, with the 54-field BroadcastObject schema in the same order the API returns it, verified field by field against a live response rather than transcribed from the source. The object exists because a graphics template binds each field to a fixed path, so three properties are contractual rather than incidental and are documented as such: it is FLAT, every key is PRESENT on every read (null, never missing — a missing key breaks a binding), and the score is already in DISPLAY form (p1_points reads \"40\" or \"AD\", set_line is one string, each set has its own column). It is a projection of what the API already publishes plus the break-, set- and match-point flags the scoring pipeline derives; nothing in it is a new fact and nothing is guessed.",
+ "The 1 Hz polling contract, which is the part a live scoreboard gets wrong. Every response carries a strong ETag and Cache-Control: no-cache, max-age=1, so If-None-Match turns an unchanged state into a 304 with no body; ULTRA's 600-a-minute burst window means ten single-match pollers at 1 Hz fit on one key, and /broadcast/live stays one request however many matches are live. A new state is detected on sequence, never on served_at, which changes on every read — stated on both routes because polling on served_at is the mistake that looks like it works."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.53",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.53",
+ "title": "1.13.53: The tape now publishes the accept counter the withdrawal ledger points at, and 1 more change",
+ "summary": "Added: The tape now publishes the accept counter the withdrawal ledger points at. Fixed: Score.sequence said the tape does not carry it, and the stated reason was the gap.",
+ "content_html": "
Added
\n
\n
The tape now publishes the accept counter the withdrawal ledger points at.GET /matches/{matchId}/withdrawals tells you a sequence was withdrawn and to resume from replaced_by_sequence, and until today no served history-tape row carried that number, so the pointer named a row nothing identified. HistoryTapeRow.sequence is the same per-match accept counter as Score.sequence on a live read. Because a withdrawn state never enters the record at all, a withdrawal now reads as a GAP in the tape's run of counters, which is what makes "resume from the predecessor" resolvable against the served history. Reported by an Ultra customer who verified three withdrawal records against the tape and wrote that "the history rows lack sequence IDs, so these responses alone cannot reconstruct exactly what a live client received". Null, never fabricated, on a row archived before 2026-09-26 (the archive did not keep the counter, so it died with the live row at retention) and on the pre-2023 archive tape, which has no such counter. Unlike accepted_at and received_at it is NOT null on a reconstructed row: a counter is a fact about our own accept order.
\n
\n
Fixed
\n
\n
Score.sequence said the tape does not carry it, and the stated reason was the gap. The description read "absent on history-tape rows, which are already served in order". Being served in order does not resolve a replaced_by_sequence, which is a pointer at a counter rather than a position in a list. Corrected, and it now points at HistoryTapeRow.sequence.
\n
",
+ "date_published": "2026-09-26T00:00:00Z",
+ "tags": [
+ "Added",
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.53",
+ "date": "2026-09-26",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "The tape now publishes the accept counter the withdrawal ledger points at. GET /matches/{matchId}/withdrawals tells you a sequence was withdrawn and to resume from replaced_by_sequence, and until today no served history-tape row carried that number, so the pointer named a row nothing identified. HistoryTapeRow.sequence is the same per-match accept counter as Score.sequence on a live read. Because a withdrawn state never enters the record at all, a withdrawal now reads as a GAP in the tape's run of counters, which is what makes \"resume from the predecessor\" resolvable against the served history. Reported by an Ultra customer who verified three withdrawal records against the tape and wrote that \"the history rows lack sequence IDs, so these responses alone cannot reconstruct exactly what a live client received\". Null, never fabricated, on a row archived before 2026-09-26 (the archive did not keep the counter, so it died with the live row at retention) and on the pre-2023 archive tape, which has no such counter. Unlike accepted_at and received_at it is NOT null on a reconstructed row: a counter is a fact about our own accept order."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Fixed",
+ "items": [
+ "Score.sequence said the tape does not carry it, and the stated reason was the gap. The description read \"absent on history-tape rows, which are already served in order\". Being served in order does not resolve a replaced_by_sequence, which is a pointer at a counter rather than a position in a list. Corrected, and it now points at HistoryTapeRow.sequence."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.52",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.52",
+ "title": "1.13.52: The tape's two accept clocks are published, and 3 more changes",
+ "summary": "Added: The tape's two accept clocks are published. The arrival clock's start date, which is the part that bites. Plus 2 more changes.",
+ "content_html": "
Added
\n
\n
The tape's two accept clocks are published.HistoryTapeRow.accepted_at and HistoryTapeRow.received_at have ridden every observed row of GET /history/matches/{matchId} since 2026-09-24 and were documented nowhere, so a customer doing timing research on a tape had no published field to read. accepted_at is the instant our arbiter accepted the state, stamped once and never refreshed, which is the clock to difference against: timestamp starts equal to it and is then refreshed while the owning source keeps re-asserting the unchanged state, so on most raw rows timestamp is a later re-assertion instant (on match 189679, 164 of its 171 observed rows carry a timestamp more than a second after accepted_at). received_at is the instant the state reached our edge before arbitration, so accepted_at minus received_at is our own arbitration step. Neither figure is a claim about speed before our edge and neither is the court's clock. Both are null on every reconstructed row, which never had a clock.
\n
The arrival clock's start date, which is the part that bites.received_at is stamped only as a state arrives, so it can never be filled in afterwards. The earliest one we hold is 2026-09-24T02:00:26Z; on every observed row accepted before that instant it is null while timestamp and accepted_at are populated. A study that spans the boundary has to read accepted_at, which rides the whole observed corpus. Measured on match 189679 (12 September) after a Basic customer reported it: 171 observed rows, accepted_at on all 171, received_at on none.
\n
What server guarantees on a tiebreak row, and what it does not. Tiebreak rows are never settled on either sequence. The server genuinely changes during a breaker, so there is no constant to take a majority of, and the field is the source's own assertion carried through. Where a breaker's rows skip points the rotation cannot be reconstructed from what is served: you have the server of a state, not of a numbered point, and meta.points.transitions_legal against transitions_total tells you whether that happened. On the set roll-up row carrying tiebreak_final, server is the value of that committed set-end state and is NOT a preserved copy of the final breaker point's server. Reported by a Basic customer who could not reconcile 21 tiebreak transitions against the standard rotation and was right not to.
\n
\n
Fixed
\n
\n
Score.accepted_at said "absent on history-tape rows". It has been present on them since 2026-09-24, so the description had been wrong for a day, on the one field a timing study is told to use. Corrected, and the tape row now documents its own copy.
\n
",
+ "date_published": "2026-09-25T00:00:00Z",
+ "tags": [
+ "Added",
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.52",
+ "date": "2026-09-25",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "The tape's two accept clocks are published. HistoryTapeRow.accepted_at and HistoryTapeRow.received_at have ridden every observed row of GET /history/matches/{matchId} since 2026-09-24 and were documented nowhere, so a customer doing timing research on a tape had no published field to read. accepted_at is the instant our arbiter accepted the state, stamped once and never refreshed, which is the clock to difference against: timestamp starts equal to it and is then refreshed while the owning source keeps re-asserting the unchanged state, so on most raw rows timestamp is a later re-assertion instant (on match 189679, 164 of its 171 observed rows carry a timestamp more than a second after accepted_at). received_at is the instant the state reached our edge before arbitration, so accepted_at minus received_at is our own arbitration step. Neither figure is a claim about speed before our edge and neither is the court's clock. Both are null on every reconstructed row, which never had a clock.",
+ "The arrival clock's start date, which is the part that bites. received_at is stamped only as a state arrives, so it can never be filled in afterwards. The earliest one we hold is 2026-09-24T02:00:26Z; on every observed row accepted before that instant it is null while timestamp and accepted_at are populated. A study that spans the boundary has to read accepted_at, which rides the whole observed corpus. Measured on match 189679 (12 September) after a Basic customer reported it: 171 observed rows, accepted_at on all 171, received_at on none.",
+ "What server guarantees on a tiebreak row, and what it does not. Tiebreak rows are never settled on either sequence. The server genuinely changes during a breaker, so there is no constant to take a majority of, and the field is the source's own assertion carried through. Where a breaker's rows skip points the rotation cannot be reconstructed from what is served: you have the server of a state, not of a numbered point, and meta.points.transitions_legal against transitions_total tells you whether that happened. On the set roll-up row carrying tiebreak_final, server is the value of that committed set-end state and is NOT a preserved copy of the final breaker point's server. Reported by a Basic customer who could not reconcile 21 tiebreak transitions against the standard rotation and was right not to."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Fixed",
+ "items": [
+ "Score.accepted_at said \"absent on history-tape rows\". It has been present on them since 2026-09-24, so the description had been wrong for a day, on the one field a timing study is told to use. Corrected, and the tape row now documents its own copy."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.51",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.51",
+ "title": "1.13.51: The score_withdrawn frame is documented on the push feed page, where a streaming consumer looks for it",
+ "summary": "Added: The score_withdrawn frame is documented on the push feed page, where a streaming consumer looks for it.",
+ "content_html": "
Added
\n
\n
The score_withdrawn frame is documented on the push feed page, where a streaming consumer looks for it. 1.13.50 put the record and the frame's fields on the score read and in the schema list, and the frame is a STREAM artifact: it arrives on the same channel as the score frame it retracts. GET /ws-token now carries the frame itself, the reason it exists (the feed publishes every accepted state and runs no trust deference, while the score read runs the legality gate and can serve an older state), what each kind asks you to do, and the three reads that recover a judgement you were offline for. No spec surface changed.
\n
",
+ "date_published": "2026-09-25T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.51",
+ "date": "2026-09-25",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "The score_withdrawn frame is documented on the push feed page, where a streaming consumer looks for it. 1.13.50 put the record and the frame's fields on the score read and in the schema list, and the frame is a STREAM artifact: it arrives on the same channel as the score frame it retracts. GET /ws-token now carries the frame itself, the reason it exists (the feed publishes every accepted state and runs no trust deference, while the score read runs the legality gate and can serve an older state), what each kind asks you to do, and the three reads that recover a judgement you were offline for. No spec surface changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.50",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.50",
+ "title": "1.13.50: A withdrawn score state is now recoverable, and GET /matches/{matchId}/withdrawals serves the record, and 2 more changes",
+ "summary": "Added: A withdrawn score state is now recoverable, and GET /matches/{matchId}/withdrawals serves the record. The score_withdrawn frame's own fields are published for the first time. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
A withdrawn score state is now recoverable, and GET /matches/{matchId}/withdrawals serves the record. The verdict object has told a stream consumer since 2026-09-23 that a state it holds was taken back, but it is recomputed on every read over the newest state, so it returned null again as soon as the next state was accepted. In a live match that is seconds. A consumer that disconnected and missed the score_withdrawn frame had no way back to it: the publisher's duplicate ledgers live inside the publishing process, and nothing else recorded the judgement. Every withdrawal is now written down when the judgement is made, before the frame goes out. Three reads use it. GET /matches/{matchId}/withdrawals (ULTRA, the tier that receives the frame) returns every refused sequence on a match, oldest first, each with kind, reason, withdrawn_at, published_at, the sequence served in its place, and the frame exactly as the feed sent it. GET /matches?withdrawn_since=<instant> lists the match ids with a withdrawal since then, with a count per match, so a consumer coming back after a gap knows which matches to ask about. And ?sequence=N on GET /matches/{matchId}/score makes verdict answer for the one sequence you hold, even after later states have landed. Records are kept 90 days from the withdrawal.
\n
The score_withdrawn frame's own fields are published for the first time.type, match_id, kind, superseded_sequence, reason, safe_to_resume, ts and published_at. It carries NO replacement score and no new sequence, which is the question a customer had to ask us: the state to resume from is the one on the next score frame, or on a fresh GET /matches/{matchId}/score, and the sequence served in the refused state's place is on the withdrawal record rather than on the frame. ts is when the judgement was made, published_at when the frame was handed to the feed.
\n
\n
Changed
\n
\n
Only a discarded withdrawal is re-asserted by ?sequence=. A deferred state may since have been corroborated and a withheld read described that moment's scan, so neither answers from the record; a verdict the read produces itself always wins, because it describes the state actually being served. ?sequence= changes verdict and nothing else on the object. A non-integer is 400 bad_sequence rather than an accepted-and-ignored filter, because a silently dropped lookup would read as "not withdrawn". ?withdrawn_since= likewise takes only limit and offset beside it: any match filter, status or updated_since alongside it is 400 bad_withdrawn_since.
\n
",
+ "date_published": "2026-09-25T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.50",
+ "date": "2026-09-25",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "A withdrawn score state is now recoverable, and GET /matches/{matchId}/withdrawals serves the record. The verdict object has told a stream consumer since 2026-09-23 that a state it holds was taken back, but it is recomputed on every read over the newest state, so it returned null again as soon as the next state was accepted. In a live match that is seconds. A consumer that disconnected and missed the score_withdrawn frame had no way back to it: the publisher's duplicate ledgers live inside the publishing process, and nothing else recorded the judgement. Every withdrawal is now written down when the judgement is made, before the frame goes out. Three reads use it. GET /matches/{matchId}/withdrawals (ULTRA, the tier that receives the frame) returns every refused sequence on a match, oldest first, each with kind, reason, withdrawn_at, published_at, the sequence served in its place, and the frame exactly as the feed sent it. GET /matches?withdrawn_since= lists the match ids with a withdrawal since then, with a count per match, so a consumer coming back after a gap knows which matches to ask about. And ?sequence=N on GET /matches/{matchId}/score makes verdict answer for the one sequence you hold, even after later states have landed. Records are kept 90 days from the withdrawal.",
+ "The score_withdrawn frame's own fields are published for the first time. type, match_id, kind, superseded_sequence, reason, safe_to_resume, ts and published_at. It carries NO replacement score and no new sequence, which is the question a customer had to ask us: the state to resume from is the one on the next score frame, or on a fresh GET /matches/{matchId}/score, and the sequence served in the refused state's place is on the withdrawal record rather than on the frame. ts is when the judgement was made, published_at when the frame was handed to the feed."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Only a discarded withdrawal is re-asserted by ?sequence=. A deferred state may since have been corroborated and a withheld read described that moment's scan, so neither answers from the record; a verdict the read produces itself always wins, because it describes the state actually being served. ?sequence= changes verdict and nothing else on the object. A non-integer is 400 bad_sequence rather than an accepted-and-ignored filter, because a silently dropped lookup would read as \"not withdrawn\". ?withdrawn_since= likewise takes only limit and offset beside it: any match filter, status or updated_since alongside it is 400 bad_withdrawn_since."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.49",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.49",
+ "title": "1.13.49: The tournament catalogue takes ?tier=, which it had been accepting and ignoring",
+ "summary": "Fixed: The tournament catalogue takes ?tier=, which it had been accepting and ignoring.",
+ "content_html": "
Fixed
\n
\n
The tournament catalogue takes ?tier=, which it had been accepting and ignoring.GET /tournaments publishes each row's tier and tier_source, and the two match listings have filtered on the same vocabulary since 1.13.28. The catalogue did not: ?tier=atp_500, ?tier=laver_cup and an outright invalid value each returned the entire 10,302-row catalogue under a 200, byte-identical to no filter at all, because the level was attached to a page after it had already been selected. It now selects. The one difference from the match listings is the season, and it is stated on the parameter: a catalogue row has no season of its own, so ?tier= here reads the CURRENT season, the same season the row's own tier field reports, while on /matches and /history/matches it reads each match's own season. A tournament with no row for the current season matches no value. An invalid value is now a 400 bad_tier carrying the vocabulary, as it always was elsewhere.
\n
",
+ "date_published": "2026-09-24T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.49",
+ "date": "2026-09-24",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "The tournament catalogue takes ?tier=, which it had been accepting and ignoring. GET /tournaments publishes each row's tier and tier_source, and the two match listings have filtered on the same vocabulary since 1.13.28. The catalogue did not: ?tier=atp_500, ?tier=laver_cup and an outright invalid value each returned the entire 10,302-row catalogue under a 200, byte-identical to no filter at all, because the level was attached to a page after it had already been selected. It now selects. The one difference from the match listings is the season, and it is stated on the parameter: a catalogue row has no season of its own, so ?tier= here reads the CURRENT season, the same season the row's own tier field reports, while on /matches and /history/matches it reads each match's own season. A tournament with no row for the current season matches no value. An invalid value is now a 400 bad_tier carrying the vocabulary, as it always was elsewhere."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.48",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.48",
+ "title": "1.13.48: tiebreak_final on the history tape row is documented for the first time, and the page-first rule the reference stated is no longer true, and 1 more change",
+ "summary": "Fixed: tiebreak_final on the history tape row is documented for the first time, and the page-first rule the reference stated is no longer true. The row that closes a tiebreak set carries the score whichever row the set counter moves on.",
+ "content_html": "
Fixed
\n
\n
tiebreak_final on the history tape row is documented for the first time, and the page-first rule the reference stated is no longer true. A PRO customer wired a bot to the closing score and reported that match 195134 served a tiebreaks array of {p1: 2, p2: 7} while no tape row carried tiebreak_final, and that it stayed that way on every later read. Two things were wrong here. The field had never appeared in this reference on HistoryTapeRow at all, though it is served on GET /history/matches/{matchId} and a customer had been pointed at it in writing; it is now described beside is_tiebreak. And the MatchPoints text said a page that *starts* on a roll-up row "has no previous row to read and uses the recorded finals alone", which was true when written and is not true now: since 2026-09-24 such a page reads the one row before it, so the value does not depend on where the page was cut. Both endpoints resolve the score the same way and agree with the response's own tiebreaks array.
\n
The row that closes a tiebreak set carries the score whichever row the set counter moves on. The tape's set counter lags the games array by a row: the row that took 195134's first set to 6-7 still read sets[0, 0], and the counter reached [0, 1] on the row after it. The stamp only fired when the counter grew on that same row, so a set whose counter lagged was never stamped, and nothing revisits a tape. The games on the row now decide it, judged by the tournament's own set test so the Next Gen first-to-4 format is read by its own rule. Measured on the live endpoint before the change: of 9 live matches whose tiebreaks array stated a final, 2 carried it on no row, both first-set breakers.
\n
",
+ "date_published": "2026-09-24T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.48",
+ "date": "2026-09-24",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "tiebreak_final on the history tape row is documented for the first time, and the page-first rule the reference stated is no longer true. A PRO customer wired a bot to the closing score and reported that match 195134 served a tiebreaks array of {p1: 2, p2: 7} while no tape row carried tiebreak_final, and that it stayed that way on every later read. Two things were wrong here. The field had never appeared in this reference on HistoryTapeRow at all, though it is served on GET /history/matches/{matchId} and a customer had been pointed at it in writing; it is now described beside is_tiebreak. And the MatchPoints text said a page that *starts* on a roll-up row \"has no previous row to read and uses the recorded finals alone\", which was true when written and is not true now: since 2026-09-24 such a page reads the one row before it, so the value does not depend on where the page was cut. Both endpoints resolve the score the same way and agree with the response's own tiebreaks array.",
+ "The row that closes a tiebreak set carries the score whichever row the set counter moves on. The tape's set counter lags the games array by a row: the row that took 195134's first set to 6-7 still read sets [0, 0], and the counter reached [0, 1] on the row after it. The stamp only fired when the counter grew on that same row, so a set whose counter lagged was never stamped, and nothing revisits a tape. The games on the row now decide it, judged by the tournament's own set test so the Next Gen first-to-4 format is read by its own rule. Measured on the live endpoint before the change: of 9 live matches whose tiebreaks array stated a final, 2 carried it on no row, both first-set breakers."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.47",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.47",
+ "title": "1.13.47: A set completion that does not end the match is assigned a sequence only once confirmed",
+ "summary": "Changed: A set completion that does not end the match is assigned a sequence only once confirmed.",
+ "content_html": "
Changed
\n
\n
A set completion that does not end the match is assigned a sequence only once confirmed. Live since 2026-09-24 and stated here for the first time. A live scoring display can render the set-ending frame ON set point, before the point is decided, and step back when the point is not won; a customer measured exactly that twice. A rise in sets that does not end the match, arriving while the state we hold is set point for the side that would gain the set, is now held until the same source keeps presenting it for 10 s or more, a source at least as trusted presents it, or the same source shows play continuing inside the new set. The unconfirmed frame is never stored, so no sequence ever carries it and nothing is published as provisional — a sequence that carries a set completion is one that met the rule. Until then GET /matches/{matchId}/score, the push feed and the WebSocket all keep serving the set-point state; a frame nobody confirms within 120 s is discarded. A set completion that ends the match is exempt and follows the completion rules instead. Measured over the 30 days to 2026-09-24: a median 10 s added between the point and the published set-end, and on a feed that does not re-present states the set-end lands with the first point of the next set (90th percentile 133 s). It does not cover a set genuinely won and then corrected on court, which can still be published and rewound — read corroborated beside it, and verdict says so when a state is taken back.
\n
",
+ "date_published": "2026-09-24T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.47",
+ "date": "2026-09-24",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "A set completion that does not end the match is assigned a sequence only once confirmed. Live since 2026-09-24 and stated here for the first time. A live scoring display can render the set-ending frame ON set point, before the point is decided, and step back when the point is not won; a customer measured exactly that twice. A rise in sets that does not end the match, arriving while the state we hold is set point for the side that would gain the set, is now held until the same source keeps presenting it for 10 s or more, a source at least as trusted presents it, or the same source shows play continuing inside the new set. The unconfirmed frame is never stored, so no sequence ever carries it and nothing is published as provisional — a sequence that carries a set completion is one that met the rule. Until then GET /matches/{matchId}/score, the push feed and the WebSocket all keep serving the set-point state; a frame nobody confirms within 120 s is discarded. A set completion that ends the match is exempt and follows the completion rules instead. Measured over the 30 days to 2026-09-24: a median 10 s added between the point and the published set-end, and on a feed that does not re-present states the set-end lands with the first point of the next set (90th percentile 133 s). It does not cover a set genuinely won and then corrected on court, which can still be published and rewound — read corroborated beside it, and verdict says so when a state is taken back."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.46",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.46",
+ "title": "1.13.46: corroborated and corroborated_age_seconds on the Score: has an independent source vouched for THIS EXACT state?, and 1 more change",
+ "summary": "Added: corroborated and corroborated_age_seconds on the Score: has an independent source vouched for THIS EXACT state? changing_sources_count on the Score is documented.",
+ "content_html": "
Added
\n
\n
corroborated and corroborated_age_seconds on the Score: has an independent source vouched for THIS EXACT state? Raised by a customer who acted on a completed set that was taken back seconds later. corroborated is true when a source other than the one that wrote the state has since presented us the same state, false when none has — including a brand-new state nobody has had a chance to agree with yet — and null on the archive fallback of a retention-trimmed completed match, where age_seconds and sources_count are null too. corroborated_age_seconds is the seconds since that vouch, null when there is none. This is the field to gate a set-end on.sources_count and changing_sources_count are window measures over the match and say nothing about the state in front of you, so a set completion that only one source has announced can sit beside a perfectly healthy count — and a set-end is the one state whose correction changes a result rather than a number. It is evidence, not proof: a vouched state can still be corrected, so compare corroborated_age_seconds with age_seconds and treat a vouch younger than the state it vouches for as the weaker signal it is. When a state is taken back the sequence goes down and verdict says so. Present wherever sources_count is; a yes/no, never a name.
\n
changing_sources_count on the Score is documented. Live since 2026-09-24 and absent from this reference. How many distinct sources CHANGED the score in the last 300 seconds, counted on the content clock alone, so a source re-asserting an unchanged state is not counted — the difference from sources_count, which a feed stuck re-sending its last state holds at 1 for as long as the echo lasts while this reads 0. The window is wider on purpose: a point lands every 30-60 s and a changeover is about 90 s. A cross-confirmation is agreement, not a change, so this can read below sources_count and never above it.
\n
",
+ "date_published": "2026-09-24T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.46",
+ "date": "2026-09-24",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "corroborated and corroborated_age_seconds on the Score: has an independent source vouched for THIS EXACT state? Raised by a customer who acted on a completed set that was taken back seconds later. corroborated is true when a source other than the one that wrote the state has since presented us the same state, false when none has — including a brand-new state nobody has had a chance to agree with yet — and null on the archive fallback of a retention-trimmed completed match, where age_seconds and sources_count are null too. corroborated_age_seconds is the seconds since that vouch, null when there is none. This is the field to gate a set-end on. sources_count and changing_sources_count are window measures over the match and say nothing about the state in front of you, so a set completion that only one source has announced can sit beside a perfectly healthy count — and a set-end is the one state whose correction changes a result rather than a number. It is evidence, not proof: a vouched state can still be corrected, so compare corroborated_age_seconds with age_seconds and treat a vouch younger than the state it vouches for as the weaker signal it is. When a state is taken back the sequence goes down and verdict says so. Present wherever sources_count is; a yes/no, never a name.",
+ "changing_sources_count on the Score is documented. Live since 2026-09-24 and absent from this reference. How many distinct sources CHANGED the score in the last 300 seconds, counted on the content clock alone, so a source re-asserting an unchanged state is not counted — the difference from sources_count, which a feed stuck re-sending its last state holds at 1 for as long as the echo lasts while this reads 0. The window is wider on purpose: a point lands every 30-60 s and a changeover is about 90 s. A cross-confirmation is agreement, not a change, so this can read below sources_count and never above it."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.45",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.45",
+ "title": "1.13.45: GET /history/incidents and GET /history/incidents/{incidentId}/matches are documented for the first time, and 1 more change",
+ "summary": "Added: GET /history/incidents and GET /history/incidents/{incidentId}/matches are documented for the first time. The verdict object on the Score is documented.",
+ "content_html": "
Added
\n
\n
GET /history/incidents and GET /history/incidents/{incidentId}/matches are documented for the first time. Both have been live and neither appeared anywhere in this reference, so the register a customer is meant to reconcile against could not be found from the documentation. /history/incidents returns the published data-quality incidents, oldest window first, each with its window_utc (start, end and the basis it was read from), rule_before, rule_after, defect, affected_surfaces, how_to_tell, corrections and matches. The register is part of a release and is not inferred at read time, so a record does not change between releases. /history/incidents/{incidentId}/matches streams one row per affected match as JSONL, or CSV with ?format=csv, with the columns incident_id, match_id, tournament, tour, round, draw_stage, scheduled_time, affected_field, served and expected. A match whose derivation would be a guess is left out and the incident's matches.note says which, so the export is exact rather than complete. Measured on the live API 2026-09-24: five incidents across the kinds pricing_rule, version_stamp, field_orientation and field_repair.
\n
The verdict object on the Score is documented. Live since 2026-09-23 and absent from this reference. It is null on an ordinary read. When the legality gate refuses a state a stream has already published, it carries kind (withdrawn, deferred or withheld), superseded_sequence (the sequence a stream consumer is holding), reason and safe_to_resume, which is true exactly when kind is withdrawn. The push feed, the WebSocket and webhooks announce the same judgement as a score_withdrawn frame naming the same sequence, so a stream consumer learns a state was taken back without polling. GET /matches/{matchId}/score now names the field and points at the incident register.
\n
\n
No field, endpoint or behaviour changed by this release; both were already served.
",
+ "date_published": "2026-09-24T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.45",
+ "date": "2026-09-24",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /history/incidents and GET /history/incidents/{incidentId}/matches are documented for the first time. Both have been live and neither appeared anywhere in this reference, so the register a customer is meant to reconcile against could not be found from the documentation. /history/incidents returns the published data-quality incidents, oldest window first, each with its window_utc (start, end and the basis it was read from), rule_before, rule_after, defect, affected_surfaces, how_to_tell, corrections and matches. The register is part of a release and is not inferred at read time, so a record does not change between releases. /history/incidents/{incidentId}/matches streams one row per affected match as JSONL, or CSV with ?format=csv, with the columns incident_id, match_id, tournament, tour, round, draw_stage, scheduled_time, affected_field, served and expected. A match whose derivation would be a guess is left out and the incident's matches.note says which, so the export is exact rather than complete. Measured on the live API 2026-09-24: five incidents across the kinds pricing_rule, version_stamp, field_orientation and field_repair.",
+ "The verdict object on the Score is documented. Live since 2026-09-23 and absent from this reference. It is null on an ordinary read. When the legality gate refuses a state a stream has already published, it carries kind (withdrawn, deferred or withheld), superseded_sequence (the sequence a stream consumer is holding), reason and safe_to_resume, which is true exactly when kind is withdrawn. The push feed, the WebSocket and webhooks announce the same judgement as a score_withdrawn frame naming the same sequence, so a stream consumer learns a state was taken back without polling. GET /matches/{matchId}/score now names the field and points at the incident register."
+ ],
+ "notes": [
+ "No field, endpoint or behaviour changed by this release; both were already served."
+ ]
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.44",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.44",
+ "title": "1.13.44: GET /matches/{matchId}/points: the row that closes a tiebreak set now carries tiebreak_final, and the reference did not say so",
+ "summary": "GET /matches/{matchId}/points: the row that closes a tiebreak set now carries tiebreak_final, and the reference did not say so.",
+ "content_html": "
\n
GET /matches/{matchId}/points: the row that closes a tiebreak set now carries tiebreak_final, and the reference did not say so. Raised by a Pro customer running trading bots, who reported that the tape stops one point short of every tiebreak and that a field filling once a day was no use to a decision taken while the match is live. A tiebreak's last point is carried, like every set-winning point, by the set roll-up row (tiebreak false, the set banked 7-6, score 0-0), and that row never stated the breaker's score at closure. It now carries tiebreak_final: [p1, p2] in our player order, e.g. [7, 5]. Exact when the previous row is the decided score or one point from it, so it is derived from the stream itself and is present as soon as the set closes; otherwise taken from the match's recorded tiebreak finals, the same finals GET /history/matches/{matchId} publishes as tiebreaks; null when neither can state it, never guessed. Absent on every other row, including the breaker rows. A page that *starts* on a roll-up row has no previous row to read and uses the recorded finals alone. Measured at the endpoint over the 30 hours to 2026-09-23 20:30Z: 60 completed matches held 67 tiebreak sets and 36 of them served a closing score, only 7 of those matches having a recorded final at the time. Documentation of an already-served field; no behaviour changed by this release.
\n
",
+ "date_published": "2026-09-23T00:00:00Z",
+ "tags": [],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.44",
+ "date": "2026-09-23",
+ "sections": [
+ {
+ "name": null,
+ "items": [
+ "GET /matches/{matchId}/points: the row that closes a tiebreak set now carries tiebreak_final, and the reference did not say so. Raised by a Pro customer running trading bots, who reported that the tape stops one point short of every tiebreak and that a field filling once a day was no use to a decision taken while the match is live. A tiebreak's last point is carried, like every set-winning point, by the set roll-up row (tiebreak false, the set banked 7-6, score 0-0), and that row never stated the breaker's score at closure. It now carries tiebreak_final: [p1, p2] in our player order, e.g. [7, 5]. Exact when the previous row is the decided score or one point from it, so it is derived from the stream itself and is present as soon as the set closes; otherwise taken from the match's recorded tiebreak finals, the same finals GET /history/matches/{matchId} publishes as tiebreaks; null when neither can state it, never guessed. Absent on every other row, including the breaker rows. A page that *starts* on a roll-up row has no previous row to read and uses the recorded finals alone. Measured at the endpoint over the 30 hours to 2026-09-23 20:30Z: 60 completed matches held 67 tiebreak sets and 36 of them served a closing score, only 7 of those matches having a recorded final at the time. Documentation of an already-served field; no behaviour changed by this release."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.43",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.43",
+ "title": "1.13.43: The reference said a FREE key is refused the history endpoints outright. It is not, and has not been since 2026-08-07: a FREE key is served 20 history calls…",
+ "summary": "Fixed: The reference said a FREE key is refused the history endpoints outright. It is not, and has not been since 2026-08-07: a FREE key is served 20 history calls per calendar month.",
+ "content_html": "
Fixed
\n
\n
The reference said a FREE key is refused the history endpoints outright. It is not, and has not been since 2026-08-07: a FREE key is served 20 history calls per calendar month. Three separate statements carried the wrong rule — the FREE tier summary ("No historical results"), the /matches description ("requires BASIC ... and returns 403 upgrade_required on a FREE key"), and the status parameter. Measured on the live API 2026-09-23 with a fresh FREE key: GET /history/matches answered 200 and kept answering until the monthly allowance was spent, after which it answered 403 upgrade_required carrying free_history_taste: "used" and the detail "your 20 free history calls this month are used". A reader planning against the old text would have concluded the endpoints were closed to them and either bought a tier they did not yet need or abandoned the evaluation; a reader who tried anyway got a 200 the reference said was impossible and could not tell an allowance from a private grant. /history/matches, /history/coverage, status=completed on /matches and the FREE tier summary now all state the allowance and the refusal that follows it.
\n
\n
No field, endpoint or behaviour changed; every correction is to a description.
",
+ "date_published": "2026-09-23T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.43",
+ "date": "2026-09-23",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "The reference said a FREE key is refused the history endpoints outright. It is not, and has not been since 2026-08-07: a FREE key is served 20 history calls per calendar month. Three separate statements carried the wrong rule — the FREE tier summary (\"No historical results\"), the /matches description (\"requires BASIC ... and returns 403 upgrade_required on a FREE key\"), and the status parameter. Measured on the live API 2026-09-23 with a fresh FREE key: GET /history/matches answered 200 and kept answering until the monthly allowance was spent, after which it answered 403 upgrade_required carrying free_history_taste: \"used\" and the detail \"your 20 free history calls this month are used\". A reader planning against the old text would have concluded the endpoints were closed to them and either bought a tier they did not yet need or abandoned the evaluation; a reader who tried anyway got a 200 the reference said was impossible and could not tell an allowance from a private grant. /history/matches, /history/coverage, status=completed on /matches and the FREE tier summary now all state the allowance and the refusal that follows it."
+ ],
+ "notes": [
+ "No field, endpoint or behaviour changed; every correction is to a description."
+ ]
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.42",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.42",
+ "title": "1.13.42: Every per-point coverage count in this reference was measured in our own tables and published as though it described the API. It did not, and the gap is large, and 1 more change",
+ "summary": "Fixed: Every per-point coverage count in this reference was measured in our own tables and published as though it described the API. It did not, and the gap is large. enrichment per match was correct throughout, on every one of these matches, and basis has always named which sequence you were given.",
+ "content_html": "
Fixed
\n
\n
Every per-point coverage count in this reference was measured in our own tables and published as though it described the API. It did not, and the gap is large.GET /matches/{id}/points serves a COMPLETED match's stored per-point stream only while that stream passes our quality bar. Where it does not, the endpoint serves a measured-complete RECONSTRUCTION instead, and a reconstruction carries no clock and no serve / outcome tags at all. So the share of matches whose tags a reader can fetch is lower, sometimes far lower, than the share of matches a source tagged. Re-measured 2026-09-23 on what the endpoint publishes, beside what the previous text claimed: Davis Cup World Group serve 2 of 22 (published as 22 of 22), World Group I 1 of 42 (published as 41 of 42); WTA qualifying serve18 of 48 (published as 48 of 48); ATP Challenger qualifying serve137 of 197 (195 of 197) and outcome134 of 197 (191 of 197); WTA 125 qualifying serve47 of 63 (61 of 63) and outcome30 of 63 (24 of 63); ITF qualifying outcome181 of 215 men's and 142 of 176 women's (189 of 209, 143 of 168); WTA main-tour qualifying outcome5 of 48, against 47 of 88 in the same events' main draws (published as 3 of 48 against 82 of 88). Both descriptions now state which layer they count and name basis as the per-response answer.
\n
enrichment per match was correct throughout, on every one of these matches, and basis has always named which sequence you were given. Only the prose was wrong.
\n
\n
No field, endpoint or behaviour changed; every correction is to a description.
",
+ "date_published": "2026-09-23T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.42",
+ "date": "2026-09-23",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "Every per-point coverage count in this reference was measured in our own tables and published as though it described the API. It did not, and the gap is large. GET /matches/{id}/points serves a COMPLETED match's stored per-point stream only while that stream passes our quality bar. Where it does not, the endpoint serves a measured-complete RECONSTRUCTION instead, and a reconstruction carries no clock and no serve / outcome tags at all. So the share of matches whose tags a reader can fetch is lower, sometimes far lower, than the share of matches a source tagged. Re-measured 2026-09-23 on what the endpoint publishes, beside what the previous text claimed: Davis Cup World Group serve 2 of 22 (published as 22 of 22), World Group I 1 of 42 (published as 41 of 42); WTA qualifying serve 18 of 48 (published as 48 of 48); ATP Challenger qualifying serve 137 of 197 (195 of 197) and outcome 134 of 197 (191 of 197); WTA 125 qualifying serve 47 of 63 (61 of 63) and outcome 30 of 63 (24 of 63); ITF qualifying outcome 181 of 215 men's and 142 of 176 women's (189 of 209, 143 of 168); WTA main-tour qualifying outcome 5 of 48, against 47 of 88 in the same events' main draws (published as 3 of 48 against 82 of 88). Both descriptions now state which layer they count and name basis as the per-response answer.",
+ "enrichment per match was correct throughout, on every one of these matches, and basis has always named which sequence you were given. Only the prose was wrong."
+ ],
+ "notes": [
+ "No field, endpoint or behaviour changed; every correction is to a description."
+ ]
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.41",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.41",
+ "title": "1.13.41: The outcome description named the wrong tour as the qualifying gap, and it was wrong in both directions, and 2 more changes",
+ "summary": "Fixed: The outcome description named the wrong tour as the qualifying gap, and it was wrong in both directions. tour: challenger pools two populations with different outcome coverage, and the reference described only one of them. Plus 1 more change.",
+ "content_html": "
Fixed
\n
\n
The outcome description named the wrong tour as the qualifying gap, and it was wrong in both directions. It read "Qualifying draws follow their own tour, with one gap: ... Challenger (men) qualifying states no outcome at all." Measured 2026-09-22 over completed matches carrying a per-point stream since 2026-09-12, ATP Challenger qualifying states outcome on 191 of 197 matches, with the full five-value vocabulary (6,881 unforced errors, 4,275 winners, 4,267 forced errors, 1,026 aces, 774 double faults) — a reader who segregated a corpus on that sentence discarded the single richest qualifying population in the product. The gap is on the WTA main tour, which the old sentence implicitly cleared: WTA qualifying states outcome on 3 of 48, against 82 of 88 in the same events' main draws. ITF qualifying is unchanged in direction (189 of 209 men's, 143 of 168 women's) and WTA 125 qualifying is partial (24 of 63). enrichment per match was correct on every one of these matches throughout — only the prose was wrong, and the prose is what a corpus is planned from.
\n
tour: challenger pools two populations with different outcome coverage, and the reference described only one of them. It read "Challenger: all five". That is true of ATP Challenger (tier: challenger_50 … challenger_175): 252 of 255 completed main-draw matches with a stream carry the full five. It is not true of the WTA 125 events, which are also served as tour: challenger (tier: wta_125) and state ace and double fault only — 137 of 141 main-draw matches carry outcome, none of them a winner, forced error or unforced error. Read tier, not tour, to tell the two apart. Both descriptions now say so.
\n
The serve qualifying denominators are re-measured (WTA qualifying 48 of 48, ATP Challenger qualifying 195 of 197, WTA 125 qualifying 61 of 63) and name the populations by tier rather than by "Challenger (men)" / "Challenger (women)", which the served fields do not distinguish.
\n
\n
No field, endpoint or behaviour changed; every correction is to a description.
",
+ "date_published": "2026-09-22T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.41",
+ "date": "2026-09-22",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "The outcome description named the wrong tour as the qualifying gap, and it was wrong in both directions. It read \"Qualifying draws follow their own tour, with one gap: ... Challenger (men) qualifying states no outcome at all.\" Measured 2026-09-22 over completed matches carrying a per-point stream since 2026-09-12, ATP Challenger qualifying states outcome on 191 of 197 matches, with the full five-value vocabulary (6,881 unforced errors, 4,275 winners, 4,267 forced errors, 1,026 aces, 774 double faults) — a reader who segregated a corpus on that sentence discarded the single richest qualifying population in the product. The gap is on the WTA main tour, which the old sentence implicitly cleared: WTA qualifying states outcome on 3 of 48, against 82 of 88 in the same events' main draws. ITF qualifying is unchanged in direction (189 of 209 men's, 143 of 168 women's) and WTA 125 qualifying is partial (24 of 63). enrichment per match was correct on every one of these matches throughout — only the prose was wrong, and the prose is what a corpus is planned from.",
+ "tour: challenger pools two populations with different outcome coverage, and the reference described only one of them. It read \"Challenger: all five\". That is true of ATP Challenger (tier: challenger_50 … challenger_175): 252 of 255 completed main-draw matches with a stream carry the full five. It is not true of the WTA 125 events, which are also served as tour: challenger (tier: wta_125) and state ace and double fault only — 137 of 141 main-draw matches carry outcome, none of them a winner, forced error or unforced error. Read tier, not tour, to tell the two apart. Both descriptions now say so.",
+ "The serve qualifying denominators are re-measured (WTA qualifying 48 of 48, ATP Challenger qualifying 195 of 197, WTA 125 qualifying 61 of 63) and name the populations by tier rather than by \"Challenger (men)\" / \"Challenger (women)\", which the served fields do not distinguish."
+ ],
+ "notes": [
+ "No field, endpoint or behaviour changed; every correction is to a description."
+ ]
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.40",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.40",
+ "title": "1.13.40: tier and tier_source on every match and tournament object, with ?tier= on GET /matches and GET /history/matches, and 5 more changes",
+ "summary": "Added: tier and tier_source on every match and tournament object, with ?tier= on GET /matches and GET /history/matches. A correction to a published result is signalled, never a silent edit: basis: restatement on the status history, result_restated_at / result_version on every match object, and ?restated_since= on GET /history/matches. Plus 4 more changes.",
+ "content_html": "
Added
\n
\n
tier and tier_source on every match and tournament object, with ?tier= on GET /matches and GET /history/matches.tier is the level the tournament was played at in the season of the match — the official category as the tour publishes it — and answers the question category cannot: is this a WTA 125, an ITF W35 or a Challenger 75? The vocabulary is closed: grand_slam; the ATP levels (atp_finals, atp_1000, atp_500, atp_250, next_gen_finals); the WTA levels (wta_finals, wta_elite_trophy, wta_1000, wta_500, wta_250, wta_125); challenger_175 … challenger_50; the ITF World Tennis Tour categories (itf_m15, itf_m25, itf_w15 … itf_w100, with the 2023 women's categories kept as printed that season); the team events (united_cup, davis_cup, bjk_cup, laver_cup, olympics); juniors; utr; exhibition; or null — nothing we hold names the level, never guessed. The tier is per season and it moves under one tournament_id: Dallas was atp_250 in 2023 and 2024 and atp_500 from 2025, so a 2024 Dallas match reads atp_250 and a 2025 one atp_500 under the same id. A match's tier is resolved by the calendar year of its scheduled_time, never copied from the tournament row; /tournaments rows carry the current season's, null when this season's calendar does not list the event. tier_source says how the level was established — calendar (the official per-season tour calendar), name (an unambiguous name rule: the ITF category is in the event's official name, team and UTR events are named as such), wikipedia (the season's schedule page, only where the official calendar could not state the level for that season), resolver (resolved after the seed dataset by the same rules) — and is null exactly when tier is null. The 2023–2026 seasons were verified event by event against the official calendars (99.1% of tournament-seasons resolved; the rest deliberately null). ?tier= takes comma-separated exact values, read from the same per-season table so filter and field cannot disagree; a null tier matches no value; an unknown value is a 400 bad_tier with the offending values in bad and the full vocabulary in allowed. category is unchanged and remains the coarse class.
\n
A correction to a published result is signalled, never a silent edit: basis: restatement on the status history, result_restated_at / result_version on every match object, and ?restated_since= on GET /history/matches. Raised by a licensed results customer settling off our results. Once a match has been published as completed (or as a cancelled walkover with a winner), any later change to status, event_status, winner or the final score is appended to GET /matches/{matchId}/status-history as a new row with basis: restatement, at the instant we made the change: status / event_status carry before and after as usual (the same value twice when only the winner or the score moved) and score is the result after the correction. This holds for every path that can change a result — a source's late final, a second authority's correction, an operator's repair, a completion reopened and re-closed — because it is enforced where the result is written, not by each path remembering to say so; a re-assertion of the same result, or a change to points alone, writes nothing. result_restated_at is the instant of the newest such row (null while the result stands as first published); result_version is 1 plus the number of corrections — record it with the result you settle on, and a higher number on re-read means the one you hold was superseded. ?restated_since=<ISO instant> keeps only the matches corrected after that instant — the poll to run after each settlement pass; a Z or an offset, a naive value read as UTC, and a bare date refused as 400 bad_restated_since, because a day is not an instant. Rows with this basis exist from 2026-09-22; earlier corrections were not signalled and nothing is reconstructed for them.
\n
winner on every row of every history tape: who won the point that produced the row. Raised by a modelling customer initialising serve states from the tape — server and serve were stated per row, and the third fact a model needs was only implicit in the score step. winner (1 | 2 | null) is judged from the score step between the row and the previous served row (the previous raw row on the raw sequence, the previous clean row under ?sequence=clean), by the same rule the completeness ledger counts a legal transition with — and from nothing else: never the serve, the server, the outcome tag or the pattern of play. Null on the first row and wherever the step is not one attributable point: a re-sent row, a set opener, a backward correction, a multi-game jump. A step whose games total rises by exactly one attributes to the side whose count rose — the game point was theirs — whatever in-game points the step skipped, a missed poll or a game withheld whole (meta.points.games_withheld). Present on the raw and clean sequences, on ?points=complete reads and on the pre-2023 archive tape; equal to point_winner wherever that older key is present, which is kept unchanged for existing readers. Derived once per response, never stored.
\n
schema_version on the packages listing, and bulk tape rows carry the enriched tape. Under schema_version 2 a kind=tape JSONL line is exactly what GET /history/matches/{matchId} returns for that match on its default basis, produced by the same code — the same rows with origin, serve, outcome and winner beside the score columns, and the full meta (enrichment, reconstructed_at, observed_span, meta.points where served) — so a model reads its serve states from the bulk file just as it would from the endpoint; the CSV appends origin, serve, outcome, winner after danger, in that order, and its column order grows only at the end. 1 is the shape every month built before 22 September 2026 carries: score columns only, no meta.enrichment, and a CSV that ends at danger. Months built before that date keep shape 1 until they are rebuilt — newest month first, then backwards, built_at and sha256 moving as each flips — so check the manifest rather than assuming; a consumer that needs the enriched fields should require schema_version >= 2 and, for a month still at 1, read the per-match endpoint for the matches it needs. null on the non-tape kinds.
\n
\n
Changed
\n
\n
A completed match on the tours with a complete per-point record is re-joined after the match against that complete sequence. ATP 250/500/1000, ATP Challenger and WTA 1000/500/250/125: the live join paired serve / outcome onto points as they were captured, so a point tagged after our capture went untagged. A completed match is now re-joined against the complete sequence by the same exact-state rule, and its serve/outcome coverage exceeds what was captured live. The late tags arrive as point_update frames and tagged_at revisions like any other — collect them with ?changed_since=. Nothing is inferred; a point that is never stated stays null.
\n
The pre-2023 archive tape no longer claims point_winner is null throughout a per-game tape. On a per-game tape (556 matches, 555 of them 2013) consecutive rows differ by a whole game, and the reference said no point was attributable there. Each row in fact reads the side whose game count rose — the game point was theirs — and says nothing about the game's other points; winner reads the same. No field changed shape or value.
\n
",
+ "date_published": "2026-09-22T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.40",
+ "date": "2026-09-22",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "tier and tier_source on every match and tournament object, with ?tier= on GET /matches and GET /history/matches. tier is the level the tournament was played at in the season of the match — the official category as the tour publishes it — and answers the question category cannot: is this a WTA 125, an ITF W35 or a Challenger 75? The vocabulary is closed: grand_slam; the ATP levels (atp_finals, atp_1000, atp_500, atp_250, next_gen_finals); the WTA levels (wta_finals, wta_elite_trophy, wta_1000, wta_500, wta_250, wta_125); challenger_175 … challenger_50; the ITF World Tennis Tour categories (itf_m15, itf_m25, itf_w15 … itf_w100, with the 2023 women's categories kept as printed that season); the team events (united_cup, davis_cup, bjk_cup, laver_cup, olympics); juniors; utr; exhibition; or null — nothing we hold names the level, never guessed. The tier is per season and it moves under one tournament_id: Dallas was atp_250 in 2023 and 2024 and atp_500 from 2025, so a 2024 Dallas match reads atp_250 and a 2025 one atp_500 under the same id. A match's tier is resolved by the calendar year of its scheduled_time, never copied from the tournament row; /tournaments rows carry the current season's, null when this season's calendar does not list the event. tier_source says how the level was established — calendar (the official per-season tour calendar), name (an unambiguous name rule: the ITF category is in the event's official name, team and UTR events are named as such), wikipedia (the season's schedule page, only where the official calendar could not state the level for that season), resolver (resolved after the seed dataset by the same rules) — and is null exactly when tier is null. The 2023–2026 seasons were verified event by event against the official calendars (99.1% of tournament-seasons resolved; the rest deliberately null). ?tier= takes comma-separated exact values, read from the same per-season table so filter and field cannot disagree; a null tier matches no value; an unknown value is a 400 bad_tier with the offending values in bad and the full vocabulary in allowed. category is unchanged and remains the coarse class.",
+ "A correction to a published result is signalled, never a silent edit: basis: restatement on the status history, result_restated_at / result_version on every match object, and ?restated_since= on GET /history/matches. Raised by a licensed results customer settling off our results. Once a match has been published as completed (or as a cancelled walkover with a winner), any later change to status, event_status, winner or the final score is appended to GET /matches/{matchId}/status-history as a new row with basis: restatement, at the instant we made the change: status / event_status carry before and after as usual (the same value twice when only the winner or the score moved) and score is the result after the correction. This holds for every path that can change a result — a source's late final, a second authority's correction, an operator's repair, a completion reopened and re-closed — because it is enforced where the result is written, not by each path remembering to say so; a re-assertion of the same result, or a change to points alone, writes nothing. result_restated_at is the instant of the newest such row (null while the result stands as first published); result_version is 1 plus the number of corrections — record it with the result you settle on, and a higher number on re-read means the one you hold was superseded. ?restated_since= keeps only the matches corrected after that instant — the poll to run after each settlement pass; a Z or an offset, a naive value read as UTC, and a bare date refused as 400 bad_restated_since, because a day is not an instant. Rows with this basis exist from 2026-09-22; earlier corrections were not signalled and nothing is reconstructed for them.",
+ "winner on every row of every history tape: who won the point that produced the row. Raised by a modelling customer initialising serve states from the tape — server and serve were stated per row, and the third fact a model needs was only implicit in the score step. winner (1 | 2 | null) is judged from the score step between the row and the previous served row (the previous raw row on the raw sequence, the previous clean row under ?sequence=clean), by the same rule the completeness ledger counts a legal transition with — and from nothing else: never the serve, the server, the outcome tag or the pattern of play. Null on the first row and wherever the step is not one attributable point: a re-sent row, a set opener, a backward correction, a multi-game jump. A step whose games total rises by exactly one attributes to the side whose count rose — the game point was theirs — whatever in-game points the step skipped, a missed poll or a game withheld whole (meta.points.games_withheld). Present on the raw and clean sequences, on ?points=complete reads and on the pre-2023 archive tape; equal to point_winner wherever that older key is present, which is kept unchanged for existing readers. Derived once per response, never stored.",
+ "schema_version on the packages listing, and bulk tape rows carry the enriched tape. Under schema_version 2 a kind=tape JSONL line is exactly what GET /history/matches/{matchId} returns for that match on its default basis, produced by the same code — the same rows with origin, serve, outcome and winner beside the score columns, and the full meta (enrichment, reconstructed_at, observed_span, meta.points where served) — so a model reads its serve states from the bulk file just as it would from the endpoint; the CSV appends origin, serve, outcome, winner after danger, in that order, and its column order grows only at the end. 1 is the shape every month built before 22 September 2026 carries: score columns only, no meta.enrichment, and a CSV that ends at danger. Months built before that date keep shape 1 until they are rebuilt — newest month first, then backwards, built_at and sha256 moving as each flips — so check the manifest rather than assuming; a consumer that needs the enriched fields should require schema_version >= 2 and, for a month still at 1, read the per-match endpoint for the matches it needs. null on the non-tape kinds."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "A completed match on the tours with a complete per-point record is re-joined after the match against that complete sequence. ATP 250/500/1000, ATP Challenger and WTA 1000/500/250/125: the live join paired serve / outcome onto points as they were captured, so a point tagged after our capture went untagged. A completed match is now re-joined against the complete sequence by the same exact-state rule, and its serve/outcome coverage exceeds what was captured live. The late tags arrive as point_update frames and tagged_at revisions like any other — collect them with ?changed_since=. Nothing is inferred; a point that is never stated stays null.",
+ "The pre-2023 archive tape no longer claims point_winner is null throughout a per-game tape. On a per-game tape (556 matches, 555 of them 2013) consecutive rows differ by a whole game, and the reference said no point was attributable there. Each row in fact reads the side whose game count rose — the game point was theirs — and says nothing about the game's other points; winner reads the same. No field changed shape or value."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.39",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.39",
+ "title": "1.13.39: is_qualifying states the one case where a round label DOES decide the draw",
+ "summary": "Changed: is_qualifying states the one case where a round label DOES decide the draw.",
+ "content_html": "
Changed
\n
\n
is_qualifying states the one case where a round label DOES decide the draw. The field said it is "never inferred from the round label", which is the right instinct and slightly too strong: main-draw vocabulary (Semi-finals, Final) is indeed never read that way, but a round label that itself reads Qualification Round 1 is STATING the draw, and it reads true — a false beside such a round is a payload contradicting itself rather than asserting main draw. Three matches (Chengdu, 22 Sep) were briefly served is_qualifying: false with round_code: "Q1" beside them and were repaired the same morning. Readers segregating a corpus by draw need to know which way that conflict resolves. No field changed shape or type.
\n
",
+ "date_published": "2026-09-22T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.39",
+ "date": "2026-09-22",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "is_qualifying states the one case where a round label DOES decide the draw. The field said it is \"never inferred from the round label\", which is the right instinct and slightly too strong: main-draw vocabulary (Semi-finals, Final) is indeed never read that way, but a round label that itself reads Qualification Round 1 is STATING the draw, and it reads true — a false beside such a round is a payload contradicting itself rather than asserting main draw. Three matches (Chengdu, 22 Sep) were briefly served is_qualifying: false with round_code: \"Q1\" beside them and were repaired the same morning. Readers segregating a corpus by draw need to know which way that conflict resolves. No field changed shape or type."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.38",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.38",
+ "title": "1.13.38: GET /markets states that the scope is MATCH-WINNER ONLY, and returns at most one market",
+ "summary": "Changed: GET /markets states that the scope is MATCH-WINNER ONLY, and returns at most one market.",
+ "content_html": "
Changed
\n
\n
GET /markets states that the scope is MATCH-WINNER ONLY, and returns at most one market. A prospect asked whether the odds endpoints carry game spreads / handicaps (A -3.5 @ 1.90) in addition to match-winner prices. Every description in this reference already said "match-winner market", but nowhere said what that EXCLUDES, and the endpoint's own summary said "market(s)". Both are now explicit: no handicaps or game spreads, no totals, no set-winner books, no per-game or per-set derivative; data holds at most one object and meta.count is 0 when nothing is mapped. Venues do list tennis derivatives beside the match-winner book and this API publishes none of them — a derivative is refused at the mapping step. No field changed shape or value.
\n
",
+ "date_published": "2026-09-22T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.38",
+ "date": "2026-09-22",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "GET /markets states that the scope is MATCH-WINNER ONLY, and returns at most one market. A prospect asked whether the odds endpoints carry game spreads / handicaps (A -3.5 @ 1.90) in addition to match-winner prices. Every description in this reference already said \"match-winner market\", but nowhere said what that EXCLUDES, and the endpoint's own summary said \"market(s)\". Both are now explicit: no handicaps or game spreads, no totals, no set-winner books, no per-game or per-set derivative; data holds at most one object and meta.count is 0 when nothing is mapped. Venues do list tennis derivatives beside the match-winner book and this API publishes none of them — a derivative is refused at the mapping step. No field changed shape or value."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.37",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.37",
+ "title": "1.13.37: is_qualifying is documented on the match object, and 1 more change",
+ "summary": "Added: is_qualifying is documented on the match object. Changed: round and round_code now state that they are DRAW-RELATIVE.",
+ "content_html": "
Added
\n
\n
is_qualifying is documented on the match object. The field has been served on GET /matches, GET /matches/{matchId} and GET /history/matches for some time and was absent from this reference entirely, which left the one field that separates a qualifying draw from the main draw invisible to anyone reading the docs. Three-valued: true/false are the source's own assertion, null means no source has ever stated it, and null is not false.
\n
\n
Changed
\n
\n
round and round_code now state that they are DRAW-RELATIVE. Both describe the round *within* the draw the match belongs to, and the draw is named by is_qualifying — not by the round. A qualifying semi-final carries round: "... - Semi-finals" and round_code: SF, exactly as a main-draw semi-final does; Q/Q1..Q4 appear only where the label itself names the round as qualifying, which most labels do not. round_code previously said "this is the field to branch on" without that limit, and two customers independently read a qualifying match as a main-draw one and reported it as a data fault. The data was correct in every case; the documentation was not. No field changed shape or value.
\n
",
+ "date_published": "2026-09-21T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.37",
+ "date": "2026-09-21",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "is_qualifying is documented on the match object. The field has been served on GET /matches, GET /matches/{matchId} and GET /history/matches for some time and was absent from this reference entirely, which left the one field that separates a qualifying draw from the main draw invisible to anyone reading the docs. Three-valued: true/false are the source's own assertion, null means no source has ever stated it, and null is not false."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "round and round_code now state that they are DRAW-RELATIVE. Both describe the round *within* the draw the match belongs to, and the draw is named by is_qualifying — not by the round. A qualifying semi-final carries round: \"... - Semi-finals\" and round_code: SF, exactly as a main-draw semi-final does; Q/Q1..Q4 appear only where the label itself names the round as qualifying, which most labels do not. round_code previously said \"this is the field to branch on\" without that limit, and two customers independently read a qualifying match as a main-draw one and reported it as a data fault. The data was correct in every case; the documentation was not. No field changed shape or value."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.36",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.36",
+ "title": "1.13.36: GET /charting/players states a sample of 11,803 charted matches, not 11,646",
+ "summary": "Changed: GET /charting/players states a sample of 11,803 charted matches, not 11,646.",
+ "content_html": "
Changed
\n
\n
GET /charting/players states a sample of 11,803 charted matches, not 11,646. The charted corpus was refreshed on 2026-09-19 and 184 matches — 133 of them played after 2026-05-24, including the US Open women's draw to the quarter-finals — were loaded into the product on 2026-09-21. matches_charted on every response has always been the true per-player denominator and was never affected; the curated-coverage note beside it was quoting an older total and understating the sample. GET /rally/matches grew in the same load: 11,822 charted matches and 1,875,132 shot-by-shot points, with the newest women's chart now 2026-09-09. The men's corpus is unchanged because no men's match has been charted since 2026-05-21. No field changed shape.
\n
",
+ "date_published": "2026-09-21T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.36",
+ "date": "2026-09-21",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "GET /charting/players states a sample of 11,803 charted matches, not 11,646. The charted corpus was refreshed on 2026-09-19 and 184 matches — 133 of them played after 2026-05-24, including the US Open women's draw to the quarter-finals — were loaded into the product on 2026-09-21. matches_charted on every response has always been the true per-player denominator and was never affected; the curated-coverage note beside it was quoting an older total and understating the sample. GET /rally/matches grew in the same load: 11,822 charted matches and 1,875,132 shot-by-shot points, with the newest women's chart now 2026-09-09. The men's corpus is unchanged because no men's match has been charted since 2026-05-21. No field changed shape."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.35",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.35",
+ "title": "1.13.35: meta.points.games_short on GET /history/matches/{matchId} counts the completed games the tape holds too few points for, and 5 more changes",
+ "summary": "Added: meta.points.games_short on GET /history/matches/{matchId} counts the completed games the tape holds too few points for. basis_reason on GET /matches/{matchId}/points says why a reconstruction was served instead of the stream. Plus 4 more changes.",
+ "content_html": "
Added
\n
\n
meta.points.games_short on GET /history/matches/{matchId} counts the completed games the tape holds too few points for. Raised by a prospect evaluating per-point histories. A game the tape skips whole is a legal 0-0 → 0-0 boundary with the games counter up by one, and the transition test alone cannot see it — so a tape could read complete: true with a whole game missing. The measurement now walks the sequence by game boundary and compares what is held for each completed game with the fewest points that game can have contained given the last state it shows (from 0-0 at least 4; from 15-30 at least 5; from 40-40 at least 8; a tiebreak at least 7) — a lower bound, never an estimate. A game the tape skips whole counts; so does a game that ended with no further rows after its last stored point. > 0 forces complete to false; nothing is inferred or repaired, and the game's boundary rows are still served exactly as stored. null = nothing measured (an empty sequence).
\n
basis_reason on GET /matches/{matchId}/points says why a reconstruction was served instead of the stream. Present on the reconstruction basis only: stream_absent (no stored stream rows), stream_incomplete (the stream is legal but does not measure complete — it joined mid-match or stopped short — and carries no tags) or stream_illegal (at least one transition is not attributable to one point: a gap or a torn row). A stream that is complete, or tagged and legal end to end, is always served on the live basis and the key is absent. Additive; nothing else in the response changes shape.
\n
\n
Changed
\n
\n
meta.points.complete now means the served sequence holds at least every point the score requires — not merely that consecutive rows are consistent with each other. True only when the sequence opens at 0-0, every transition is a legal single-point step, it carries at least as many point transitions as the final scoreline implies the match contained, every completed game holds at least as many point transitions as it must have contained (games_short is 0), it reaches a finished final scoreline at love (or the match ended early — retirement/walkover), it is not known-truncated, and no game was withheld (games_withheld). ends_at_final on the same object now also requires that the last row shows no game in progress — points at love or null, not a tiebreak in progress: a row carrying a finished spine with a game still underway ([[6,6],[3,4]] at 15-0) reads false, where before the spine alone decided. Some tapes that read complete: true until now will read false; no row is changed.
\n
A recurring score state on the history tape pairs its serve/outcome by nearest clock, instead of reading null. Tags on GET /history/matches/{matchId} rows are joined from the match's per-point stream by exact score state, and a state the stream asserts more than once (a deuce cycle revisiting deuce) cannot be told apart by the score, so until now it read null outright. Where a state recurs the join now pairs by nearest clock within 90 seconds — the tape row's observed clock against the stream point's, each stream point used at most once — and reads null otherwise (no candidate within the window, or two at the same distance). Every tag is an observed pairing, state and where needed clock, never an inference from the pattern of play. Reconstructed rows and rows no stream point matches still read null.
\n
\n
Fixed
\n
\n
A stored live stream that is complete, or tagged and legal end to end, is served on a completed match; a reconstruction is projected only when the stream falls short. Until now a measured-complete recorded sequence displaced the stream unconditionally, so a completed match could lose its serve/outcome tags and every per-point clock the moment a reconstruction landed — a projection carries neither, so a complete tagged stream is strictly more information than any reconstruction of the same match. GET /matches/{matchId}/points now keeps the live basis whenever the stream is itself measured complete (match-closing point included) or carries tags and is legal end to end (every transition one attributable point, judged in playing order), and serves the projection only when the stream has no rows, is incomplete and untagged, or holds a transition nobody can attribute — basis_reason says which. When the reconstruction serves it still serves wholesale; the two sequences are never interleaved. If a completed match reads reconstruction, re-read from after_seq=0 rather than resuming a live cursor into it.
\n
A double fault stated on a first serve is a contradiction; the point is served untagged. A source point whose serve number and outcome contradict each other — a double fault on serve 1 — was carried through as stated. serve and outcome on such a point now both read null on GET /matches/{matchId}/points and on the WebSocket frames rather than tagged wrongly; nothing is corrected to a second serve. Stored tags that carried the contradiction have been cleared.
\n
",
+ "date_published": "2026-09-21T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed",
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.35",
+ "date": "2026-09-21",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "meta.points.games_short on GET /history/matches/{matchId} counts the completed games the tape holds too few points for. Raised by a prospect evaluating per-point histories. A game the tape skips whole is a legal 0-0 → 0-0 boundary with the games counter up by one, and the transition test alone cannot see it — so a tape could read complete: true with a whole game missing. The measurement now walks the sequence by game boundary and compares what is held for each completed game with the fewest points that game can have contained given the last state it shows (from 0-0 at least 4; from 15-30 at least 5; from 40-40 at least 8; a tiebreak at least 7) — a lower bound, never an estimate. A game the tape skips whole counts; so does a game that ended with no further rows after its last stored point. > 0 forces complete to false; nothing is inferred or repaired, and the game's boundary rows are still served exactly as stored. null = nothing measured (an empty sequence).",
+ "basis_reason on GET /matches/{matchId}/points says why a reconstruction was served instead of the stream. Present on the reconstruction basis only: stream_absent (no stored stream rows), stream_incomplete (the stream is legal but does not measure complete — it joined mid-match or stopped short — and carries no tags) or stream_illegal (at least one transition is not attributable to one point: a gap or a torn row). A stream that is complete, or tagged and legal end to end, is always served on the live basis and the key is absent. Additive; nothing else in the response changes shape."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "meta.points.complete now means the served sequence holds at least every point the score requires — not merely that consecutive rows are consistent with each other. True only when the sequence opens at 0-0, every transition is a legal single-point step, it carries at least as many point transitions as the final scoreline implies the match contained, every completed game holds at least as many point transitions as it must have contained (games_short is 0), it reaches a finished final scoreline at love (or the match ended early — retirement/walkover), it is not known-truncated, and no game was withheld (games_withheld). ends_at_final on the same object now also requires that the last row shows no game in progress — points at love or null, not a tiebreak in progress: a row carrying a finished spine with a game still underway ([[6,6],[3,4]] at 15-0) reads false, where before the spine alone decided. Some tapes that read complete: true until now will read false; no row is changed.",
+ "A recurring score state on the history tape pairs its serve/outcome by nearest clock, instead of reading null. Tags on GET /history/matches/{matchId} rows are joined from the match's per-point stream by exact score state, and a state the stream asserts more than once (a deuce cycle revisiting deuce) cannot be told apart by the score, so until now it read null outright. Where a state recurs the join now pairs by nearest clock within 90 seconds — the tape row's observed clock against the stream point's, each stream point used at most once — and reads null otherwise (no candidate within the window, or two at the same distance). Every tag is an observed pairing, state and where needed clock, never an inference from the pattern of play. Reconstructed rows and rows no stream point matches still read null."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Fixed",
+ "items": [
+ "A stored live stream that is complete, or tagged and legal end to end, is served on a completed match; a reconstruction is projected only when the stream falls short. Until now a measured-complete recorded sequence displaced the stream unconditionally, so a completed match could lose its serve/outcome tags and every per-point clock the moment a reconstruction landed — a projection carries neither, so a complete tagged stream is strictly more information than any reconstruction of the same match. GET /matches/{matchId}/points now keeps the live basis whenever the stream is itself measured complete (match-closing point included) or carries tags and is legal end to end (every transition one attributable point, judged in playing order), and serves the projection only when the stream has no rows, is incomplete and untagged, or holds a transition nobody can attribute — basis_reason says which. When the reconstruction serves it still serves wholesale; the two sequences are never interleaved. If a completed match reads reconstruction, re-read from after_seq=0 rather than resuming a live cursor into it.",
+ "A double fault stated on a first serve is a contradiction; the point is served untagged. A source point whose serve number and outcome contradict each other — a double fault on serve 1 — was carried through as stated. serve and outcome on such a point now both read null on GET /matches/{matchId}/points and on the WebSocket frames rather than tagged wrongly; nothing is corrected to a second serve. Stored tags that carried the contradiction have been cleared."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.34",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.34",
+ "title": "1.13.34: The reference said Davis Cup and the Grand Slams \"read null throughout\" for the per-point tags; serve is in fact stated on the top Davis Cup tiers",
+ "summary": "Changed: The reference said Davis Cup and the Grand Slams \"read null throughout\" for the per-point tags; serve is in fact stated on the top Davis Cup tiers.",
+ "content_html": "
Changed
\n
\n
The reference said Davis Cup and the Grand Slams "read null throughout" for the per-point tags; serve is in fact stated on the top Davis Cup tiers.serve and outcome are stated separately, and the old sentence excluded both on the strength of one of them: team ties and Slams carry no per-point outcome, so outcome is genuinely null there — but the serve number is stated on them. Measured over completed matches with a per-point stream since 2026-09-12: Davis Cup World Group states serve on 22 of 22 matches and World Group I on 41 of 42, while World Group II states it on 0 of 38; outcome is null on all of them, on every one of the 14,132 rows. The machine-readable authority was right throughout and is unchanged — a World Group I match answers enrichment: {"serve": "stated", "outcome": "none"} and a World Group II match answers {"serve": "none", "outcome": "none"} — so this corrects the prose to match what the API already returns, in the direction of more coverage, not less. The serve description now names the Davis Cup tiers in its coverage list with the measurement; the outcome description now treats the two fields separately instead of excluding a competition from both. No field, endpoint or behaviour change.
\n
",
+ "date_published": "2026-09-21T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.34",
+ "date": "2026-09-21",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The reference said Davis Cup and the Grand Slams \"read null throughout\" for the per-point tags; serve is in fact stated on the top Davis Cup tiers. serve and outcome are stated separately, and the old sentence excluded both on the strength of one of them: team ties and Slams carry no per-point outcome, so outcome is genuinely null there — but the serve number is stated on them. Measured over completed matches with a per-point stream since 2026-09-12: Davis Cup World Group states serve on 22 of 22 matches and World Group I on 41 of 42, while World Group II states it on 0 of 38; outcome is null on all of them, on every one of the 14,132 rows. The machine-readable authority was right throughout and is unchanged — a World Group I match answers enrichment: {\"serve\": \"stated\", \"outcome\": \"none\"} and a World Group II match answers {\"serve\": \"none\", \"outcome\": \"none\"} — so this corrects the prose to match what the API already returns, in the direction of more coverage, not less. The serve description now names the Davis Cup tiers in its coverage list with the measurement; the outcome description now treats the two fields separately instead of excluding a competition from both. No field, endpoint or behaviour change."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.33",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.33",
+ "title": "1.13.33: serve and outcome said nothing about qualifying or doubles draws, and the internal reference called both blanket-null",
+ "summary": "Changed: serve and outcome said nothing about qualifying or doubles draws, and the internal reference called both blanket-null.",
+ "content_html": "
Changed
\n
\n
serve and outcome said nothing about qualifying or doubles draws, and the internal reference called both blanket-null. The per-point tags (added 2026-09-12) were documented tour by tour for main draws only. Measured over completed matches carrying a per-point stream since the source went live on 2026-09-12: WTA qualifying states serve on 48 of 48, Challenger (men) qualifying on 153 of 155, Challenger (women) on 49 of 51; ITF qualifying states outcome (aces and double faults) on 95 of 107 men's and 62 of 74 women's. Qualifying draws are covered like their own main draw, with exactly one real gap — outcome on Challenger (men) qualifying, which no source states at all. The doubles half of the old claim is confirmed and now carries its denominator: across 454 completed doubles matches in the same window, not one row carries either field. Both descriptions now state where each field is stated, including the qualifying and doubles positions; enrichment per match remains the authority and none of it is a promise about a match not yet played. Found by the every-sweep tag-coverage check, which reads 85 of 86 Challenger qualifying matches carrying serve against a documented null. No field, endpoint or behaviour change.
\n
",
+ "date_published": "2026-09-21T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.33",
+ "date": "2026-09-21",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "serve and outcome said nothing about qualifying or doubles draws, and the internal reference called both blanket-null. The per-point tags (added 2026-09-12) were documented tour by tour for main draws only. Measured over completed matches carrying a per-point stream since the source went live on 2026-09-12: WTA qualifying states serve on 48 of 48, Challenger (men) qualifying on 153 of 155, Challenger (women) on 49 of 51; ITF qualifying states outcome (aces and double faults) on 95 of 107 men's and 62 of 74 women's. Qualifying draws are covered like their own main draw, with exactly one real gap — outcome on Challenger (men) qualifying, which no source states at all. The doubles half of the old claim is confirmed and now carries its denominator: across 454 completed doubles matches in the same window, not one row carries either field. Both descriptions now state where each field is stated, including the qualifying and doubles positions; enrichment per match remains the authority and none of it is a promise about a match not yet played. Found by the every-sweep tag-coverage check, which reads 85 of 86 Challenger qualifying matches carrying serve against a documented null. No field, endpoint or behaviour change."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.32",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.32",
+ "title": "1.13.32: A match that ends in a tiebreak closes with the set roll-up row after the decisive tiebreak score",
+ "summary": "Changed: A match that ends in a tiebreak closes with the set roll-up row after the decisive tiebreak score.",
+ "content_html": "
Changed
\n
\n
A match that ends in a tiebreak closes with the set roll-up row after the decisive tiebreak score. The match-closing row on GET /matches/{matchId}/points was described as if the last point were always a game point one point short of the final, and a match decided in a tiebreak did not fit that shape. The docs now say what the stream actually holds: when a completed match ends in a tiebreak, the stream's last stored row is the decisive tiebreak score itself (7-3, or 8-6) and the closing row is the set roll-up after it — the next game number, number 0, tiebreak false, sets incremented for the tiebreak winner, the set banked 7-6 in games, score 0-0, server null, winner the tiebreak winner — the same row the stream stores after every other set-ending tiebreak, so ends_at_final reads true there. A 10-point match tiebreak still cannot be stated as one point and is refused as before (ends_at_final: false), as is a tape that never reached the final.
\n
",
+ "date_published": "2026-09-21T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.32",
+ "date": "2026-09-21",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "A match that ends in a tiebreak closes with the set roll-up row after the decisive tiebreak score. The match-closing row on GET /matches/{matchId}/points was described as if the last point were always a game point one point short of the final, and a match decided in a tiebreak did not fit that shape. The docs now say what the stream actually holds: when a completed match ends in a tiebreak, the stream's last stored row is the decisive tiebreak score itself (7-3, or 8-6) and the closing row is the set roll-up after it — the next game number, number 0, tiebreak false, sets incremented for the tiebreak winner, the set banked 7-6 in games, score 0-0, server null, winner the tiebreak winner — the same row the stream stores after every other set-ending tiebreak, so ends_at_final reads true there. A 10-point match tiebreak still cannot be stated as one point and is refused as before (ends_at_final: false), as is a tape that never reached the final."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.31",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.31",
+ "title": "1.13.31: GET /matches/{matchId}/points now closes a completed match with the match-closing point, and ends_at_final says whether it did, and 1 more change",
+ "summary": "Added: GET /matches/{matchId}/points now closes a completed match with the match-closing point, and ends_at_final says whether it did. tagged_at on every point row, and ?changed_since= on the same endpoint, so late serve/outcome tags are collectable without a socket.",
+ "content_html": "
Added
\n
\n
GET /matches/{matchId}/points now closes a completed match with the match-closing point, and ends_at_final says whether it did. Raised by a prospect evaluating per-point tapes, who asked where the point that wins the match is. Every row is the state *after* a point, so a game-winning point is carried by the next game's number: 0 opener — and the match-winning point had no next row to be carried by: the live stream never held it, and a serve statistic built off the stream was missing every match's last point. On a completed match served on the live basis the page now ends with one terminal row: seq = last + 1, number 0, sets / games the final score, score 0-0, tiebreak false, server null (nobody serves next), winner the match winner, ts the instant the final score was observed. It is built at read time from the stream's last row and the observed final score, and only when the two are one point apart; nothing is fabricated otherwise. ends_at_final (boolean) on the response says whether the sequence served ends on that row: false on a completed match whose stream stops short of it (a retirement or walkover, a capture that stopped two or more points short, a closer from deuce or from a match tiebreak that cannot be stated as one point), always false on a live match, and on the reconstruction basis judged from the projected sequence's last frame. serve / outcome on the terminal row are null: no source's tag for a match's last point is stored yet. WebSocket and push frames are unchanged.
\n
tagged_at on every point row, and ?changed_since= on the same endpoint, so late serve/outcome tags are collectable without a socket. The same prospect asked how to pick up tags that land after a row was read. after_seq is a cursor by seq, so it can never return a row already held — and a serve / outcome tag that lands late lands on exactly such a row; until now the only way to see it was the point_update frame on a WebSocket. Every row now carries tagged_at, the UTC instant its tags landed (null while none has, and on every reconstruction-basis frame), and ?changed_since=<ISO-8601 instant> (2026-09-20T00:35:18Z; Z or an offset, a naive value is read as UTC, a date alone is refused) returns only the rows whose tsortagged_at is later than that instant, in seq order, paged as usual and composable with after_seq. The post-match recipe: read the match, keep the instant, re-read with changed_since=<that instant> and replace held rows by seq. Anything that is not an ISO-8601 timestamp is a 400 bad_changed_since. On the reconstruction basis no row carries a clock or a tag, so a changed_since read of it is an empty page: that sequence is final at first read.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.31",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /matches/{matchId}/points now closes a completed match with the match-closing point, and ends_at_final says whether it did. Raised by a prospect evaluating per-point tapes, who asked where the point that wins the match is. Every row is the state *after* a point, so a game-winning point is carried by the next game's number: 0 opener — and the match-winning point had no next row to be carried by: the live stream never held it, and a serve statistic built off the stream was missing every match's last point. On a completed match served on the live basis the page now ends with one terminal row: seq = last + 1, number 0, sets / games the final score, score 0-0, tiebreak false, server null (nobody serves next), winner the match winner, ts the instant the final score was observed. It is built at read time from the stream's last row and the observed final score, and only when the two are one point apart; nothing is fabricated otherwise. ends_at_final (boolean) on the response says whether the sequence served ends on that row: false on a completed match whose stream stops short of it (a retirement or walkover, a capture that stopped two or more points short, a closer from deuce or from a match tiebreak that cannot be stated as one point), always false on a live match, and on the reconstruction basis judged from the projected sequence's last frame. serve / outcome on the terminal row are null: no source's tag for a match's last point is stored yet. WebSocket and push frames are unchanged.",
+ "tagged_at on every point row, and ?changed_since= on the same endpoint, so late serve/outcome tags are collectable without a socket. The same prospect asked how to pick up tags that land after a row was read. after_seq is a cursor by seq, so it can never return a row already held — and a serve / outcome tag that lands late lands on exactly such a row; until now the only way to see it was the point_update frame on a WebSocket. Every row now carries tagged_at, the UTC instant its tags landed (null while none has, and on every reconstruction-basis frame), and ?changed_since= (2026-09-20T00:35:18Z; Z or an offset, a naive value is read as UTC, a date alone is refused) returns only the rows whose ts or tagged_at is later than that instant, in seq order, paged as usual and composable with after_seq. The post-match recipe: read the match, keep the instant, re-read with changed_since= and replace held rows by seq. Anything that is not an ISO-8601 timestamp is a 400 bad_changed_since. On the reconstruction basis no row carries a clock or a tag, so a changed_since read of it is an empty page: that sequence is final at first read."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.30",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.30",
+ "title": "1.13.30: The ATP per-point outcome row said \"full\" as a projection, not a measurement",
+ "summary": "Changed: The ATP per-point outcome row said \"full\" as a projection, not a measurement.",
+ "content_html": "
Changed
\n
\n
The ATP per-point outcome row said "full" as a projection, not a measurement. In the 30 days since ATP per-point outcomes went live (2026-09-12), no ATP main-tour event was actually played — the only tour tennis in that window was a Grand Slam and Davis Cup, neither of which carries per-point outcomes — so every ATP main-tour match carried no outcome tags at all, while the docs asserted full (ace, double fault, winner, forced error, unforced error) unconditionally. The row now names the tours covered (ATP 250 / 500 / 1000), states plainly that Grand Slams and Davis Cup are not on it and read null throughout (check enrichment per match), and gives the first date the claim was actually measured against a played tour event: the week of 2026-09-22.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.30",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The ATP per-point outcome row said \"full\" as a projection, not a measurement. In the 30 days since ATP per-point outcomes went live (2026-09-12), no ATP main-tour event was actually played — the only tour tennis in that window was a Grand Slam and Davis Cup, neither of which carries per-point outcomes — so every ATP main-tour match carried no outcome tags at all, while the docs asserted full (ace, double fault, winner, forced error, unforced error) unconditionally. The row now names the tours covered (ATP 250 / 500 / 1000), states plainly that Grand Slams and Davis Cup are not on it and read null throughout (check enrichment per match), and gives the first date the claim was actually measured against a played tour event: the week of 2026-09-22."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.29",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.29",
+ "title": "1.13.29: The GET /history/matches listing's point-completeness ledger now judges the same raw sequence the per-match read serves, re-sent rows dropped — not the clean…, and 1 more change",
+ "summary": "Changed: The GET /history/matches listing's point-completeness ledger now judges the same raw sequence the per-match read serves, re-sent rows dropped — not the clean collapse. Fixed: tape.computed_at on the GET /history/matches listing carried the serving host's UTC offset (+03:00 on prod) instead of a trailing Z.",
+ "content_html": "
Changed
\n
\n
The GET /history/matches listing's point-completeness ledger now judges the same raw sequence the per-match read serves, re-sent rows dropped — not the clean collapse. The ledger's previous default basis, clean, collapses the tape to one row per distinct score state; that collapse can delete a real deuce point — two rows sitting at 40-40 with no advantage row between them is a legitimate rally, not a duplicate — and call the resulting, shorter tape point-complete when the tape actually served is not. The ledger now judges the raw sequence with provable re-sends dropped, the exact judgement the per-match read makes and publishes as meta.points.complete, so tape.points_complete_default on the listing and meta.points.complete on a fetched match agree by construction instead of disagreeing on a sequence the read never served. Measured on 7,717 completed matches scored since 1 Sep: the clean collapse called 70 of them point-complete whose served (raw) tape was not, and missed 810 whose served tape was. clean remains available only as a request option (?sequence=clean); rows scored before 2026-09-20 keep points_basis: clean until the nightly re-score reaches them. tape.points_basis, tape.points_complete_default and tape.points_complete descriptions updated to match.
\n
\n
Fixed
\n
\n
tape.computed_at on the GET /history/matches listing carried the serving host's UTC offset (+03:00 on prod) instead of a trailing Z. Same instant, valid ISO-8601, but every other timestamp in the payload — meta.generated_at, each row's timestamp — ends in Z, so a consumer diffing computed_at against either had to normalise it by hand first. Now projected to UTC with a trailing Z like the rest of the surface.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Changed",
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.29",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "The GET /history/matches listing's point-completeness ledger now judges the same raw sequence the per-match read serves, re-sent rows dropped — not the clean collapse. The ledger's previous default basis, clean, collapses the tape to one row per distinct score state; that collapse can delete a real deuce point — two rows sitting at 40-40 with no advantage row between them is a legitimate rally, not a duplicate — and call the resulting, shorter tape point-complete when the tape actually served is not. The ledger now judges the raw sequence with provable re-sends dropped, the exact judgement the per-match read makes and publishes as meta.points.complete, so tape.points_complete_default on the listing and meta.points.complete on a fetched match agree by construction instead of disagreeing on a sequence the read never served. Measured on 7,717 completed matches scored since 1 Sep: the clean collapse called 70 of them point-complete whose served (raw) tape was not, and missed 810 whose served tape was. clean remains available only as a request option (?sequence=clean); rows scored before 2026-09-20 keep points_basis: clean until the nightly re-score reaches them. tape.points_basis, tape.points_complete_default and tape.points_complete descriptions updated to match."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Fixed",
+ "items": [
+ "tape.computed_at on the GET /history/matches listing carried the serving host's UTC offset (+03:00 on prod) instead of a trailing Z. Same instant, valid ISO-8601, but every other timestamp in the payload — meta.generated_at, each row's timestamp — ends in Z, so a consumer diffing computed_at against either had to normalise it by hand first. Now projected to UTC with a trailing Z like the rest of the surface."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.28",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.28",
+ "title": "1.13.28: meta.points.games_withheld on GET /history/matches/{matchId}, and 3 more changes",
+ "summary": "Added: meta.points.games_withheld on GET /history/matches/{matchId}. tape.points_complete_default, tape.points_complete_recon and tape.points_basis on the GET /history/matches listing. Plus 2 more changes.",
+ "content_html": "
Added
\n
\n
meta.points.games_withheld on GET /history/matches/{matchId}. Present whenever reconstructed rows are in the response: the count of games whose recorded points were internally inconsistent — a recorded score state that repeats or moves backwards partway through the game — and whose in-game points are therefore withheld rather than published. The game's opening 0-0 row is still served, so the games spine stays intact; nothing is repaired, reordered or inferred — a game is served exactly as recorded, or not at all. > 0 forces meta.points.complete to false. null means not measured: either the reconstruction predates this check, or it was built by merging more than one point record rather than from a single one.
\n
tape.points_complete_default, tape.points_complete_recon and tape.points_basis on the GET /history/matches listing.points_complete on that listing is a best-basis OR: the ledger's default basis is the clean collapse of an observed tape, while the per-match read serves and measures the raw sequence, so points_complete: true can sit next to meta.points.complete: false on the same match with both honest about their own sequence. These three fields say which measurement produced the true and on which sequence — points_complete_recon: true means "fetch it with ?points=complete". All null when not yet measured.
\n
meta.points.resent_rows on GET /history/matches/{matchId}. A row identical to the row before it is the same state re-sent by a source on its timer, which the raw tape carries by design — no point separates the two rows, so they no longer count as a non-point transition. transitions_total is now rows − 1 − resent_rows. resent_rows is 0 on a clean or recon basis: the collapse already removes re-sends, and in a reconstruction one row is one point, so a repeat there is a tear in the recorded sequence and is never dropped.
\n
\n
Fixed
\n
\n
A finished-match point record that moved backwards inside a game was being reconstructed as if it were legal. Since July 2026 some recorded point sequences have not been monotone within a game (a real example: 40-30, 15-15, 15-30, 30-30, 15-30, 30-30, 30-30, 40-30 — 547 of 586 August ATP/WTA matches carried at least one such game). The tape builder kept the longest forward-moving chain through the inconsistency and published the result as an ordinary reconstructed row, so a state like 30-30 repeating three times in a row was served as fact. Those games are now withheld whole instead — counted in the new games_withheld field — rather than fitted into a plausible-looking sequence.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Added",
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.28",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "meta.points.games_withheld on GET /history/matches/{matchId}. Present whenever reconstructed rows are in the response: the count of games whose recorded points were internally inconsistent — a recorded score state that repeats or moves backwards partway through the game — and whose in-game points are therefore withheld rather than published. The game's opening 0-0 row is still served, so the games spine stays intact; nothing is repaired, reordered or inferred — a game is served exactly as recorded, or not at all. > 0 forces meta.points.complete to false. null means not measured: either the reconstruction predates this check, or it was built by merging more than one point record rather than from a single one.",
+ "tape.points_complete_default, tape.points_complete_recon and tape.points_basis on the GET /history/matches listing. points_complete on that listing is a best-basis OR: the ledger's default basis is the clean collapse of an observed tape, while the per-match read serves and measures the raw sequence, so points_complete: true can sit next to meta.points.complete: false on the same match with both honest about their own sequence. These three fields say which measurement produced the true and on which sequence — points_complete_recon: true means \"fetch it with ?points=complete\". All null when not yet measured.",
+ "meta.points.resent_rows on GET /history/matches/{matchId}. A row identical to the row before it is the same state re-sent by a source on its timer, which the raw tape carries by design — no point separates the two rows, so they no longer count as a non-point transition. transitions_total is now rows − 1 − resent_rows. resent_rows is 0 on a clean or recon basis: the collapse already removes re-sends, and in a reconstruction one row is one point, so a repeat there is a tear in the recorded sequence and is never dropped."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Fixed",
+ "items": [
+ "A finished-match point record that moved backwards inside a game was being reconstructed as if it were legal. Since July 2026 some recorded point sequences have not been monotone within a game (a real example: 40-30, 15-15, 15-30, 30-30, 15-30, 30-30, 30-30, 40-30 — 547 of 586 August ATP/WTA matches carried at least one such game). The tape builder kept the longest forward-moving chain through the inconsistency and published the result as an ordinary reconstructed row, so a state like 30-30 repeating three times in a row was served as fact. Those games are now withheld whole instead — counted in the new games_withheld field — rather than fitted into a plausible-looking sequence."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.27",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.27",
+ "title": "1.13.27: GET /history/matches/{matchId} tape rows now carry origin, and observed rows carry serve / outcome, and 2 more changes",
+ "summary": "Added: GET /history/matches/{matchId} tape rows now carry origin, and observed rows carry serve / outcome. meta.reconstructed_at, meta.observed_span and meta.enrichment on the same endpoint. reconstructed_at is the write time (UTC) of the newest reconstructed row actually served, null when none is; observed_span is {\"first\", \"last\"} — the timestamp of the first and last observed rows served, in served order, null when none is; enrichment is {\"serve\": \"stated\" | \"none\", \"outcome\": \"full\" | \"ace_double_fault\" | \"none\"}, the vocabulary that actually landed on this response's rows (none/none on a wholly reconstructed tape). Plus 1 more change.",
+ "content_html": "
Added
\n
\n
GET /history/matches/{matchId} tape rows now carry origin, and observed rows carry serve / outcome. Raised by a prospect evaluating reconstructed tapes as evidence for an independent model, who asked how to tell observed from reconstructed rows explicitly, when a reconstruction landed, and whether the per-point serve/outcome tags reach the history tape at all. origin (observed | reconstructed) is read from the row's stored provenance — the same fact that nulls the clock on a reconstructed row, never the null clock itself — so a caller no longer has to treat timestamp: null as an inferred marker. serve (1 | 2 | null) and outcome (ace | double_fault | winner | forced_error | unforced_error | null) are joined onto observed rows only from the per-point stream by EXACT score state: set count, every set's games, in-game points and the tiebreak flag all equal. A tape row stores no point ordinal, so a score state the stream asserts more than once (a deuce cycle revisiting deuce) is ambiguous and reads null, as does a row no stream point matches and every reconstructed row — nothing is inferred from the score. All additive on GET /history/matches/{matchId}; the pre-2023 ArchiveTape (every row reconstructed) does not carry origin, serve or outcome.
\n
meta.reconstructed_at, meta.observed_span and meta.enrichment on the same endpoint. reconstructed_at is the write time (UTC) of the newest reconstructed row actually served, null when none is; observed_span is {"first", "last"} — the timestamp of the first and last observed rows served, in served order, null when none is; enrichment is {"serve": "stated" | "none", "outcome": "full" | "ace_double_fault" | "none"}, the vocabulary that actually landed on this response's rows (none/none on a wholly reconstructed tape). All three are measured on the rows returned, after any sequence=clean collapse.
\n
\n
Changed
\n
\n
meta.points.available_complete is now read-through instead of nightly-only. It still answers from the nightly ledger when the ledger has an entry. When the ledger has none yet — a reconstruction that landed after the last nightly run — and this read served a reconstruction, it now answers from the live verdict just measured on that reconstruction instead of null. It reads null only when there is no ledger entry and nothing was reconstructed on this read to measure. No field or endpoint added; behaviour change on an existing field.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.27",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /history/matches/{matchId} tape rows now carry origin, and observed rows carry serve / outcome. Raised by a prospect evaluating reconstructed tapes as evidence for an independent model, who asked how to tell observed from reconstructed rows explicitly, when a reconstruction landed, and whether the per-point serve/outcome tags reach the history tape at all. origin (observed | reconstructed) is read from the row's stored provenance — the same fact that nulls the clock on a reconstructed row, never the null clock itself — so a caller no longer has to treat timestamp: null as an inferred marker. serve (1 | 2 | null) and outcome (ace | double_fault | winner | forced_error | unforced_error | null) are joined onto observed rows only from the per-point stream by EXACT score state: set count, every set's games, in-game points and the tiebreak flag all equal. A tape row stores no point ordinal, so a score state the stream asserts more than once (a deuce cycle revisiting deuce) is ambiguous and reads null, as does a row no stream point matches and every reconstructed row — nothing is inferred from the score. All additive on GET /history/matches/{matchId}; the pre-2023 ArchiveTape (every row reconstructed) does not carry origin, serve or outcome.",
+ "meta.reconstructed_at, meta.observed_span and meta.enrichment on the same endpoint. reconstructed_at is the write time (UTC) of the newest reconstructed row actually served, null when none is; observed_span is {\"first\", \"last\"} — the timestamp of the first and last observed rows served, in served order, null when none is; enrichment is {\"serve\": \"stated\" | \"none\", \"outcome\": \"full\" | \"ace_double_fault\" | \"none\"}, the vocabulary that actually landed on this response's rows (none/none on a wholly reconstructed tape). All three are measured on the rows returned, after any sequence=clean collapse."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "meta.points.available_complete is now read-through instead of nightly-only. It still answers from the nightly ledger when the ledger has an entry. When the ledger has none yet — a reconstruction that landed after the last nightly run — and this read served a reconstruction, it now answers from the live verdict just measured on that reconstruction instead of null. It reads null only when there is no ledger entry and nothing was reconstructed on this read to measure. No field or endpoint added; behaviour change on an existing field."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.26",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.26",
+ "title": "1.13.26: Team-competition matches now carry the surface of their tie, and 1 more change",
+ "summary": "Fixed: Team-competition matches now carry the surface of their tie. Changed: Match.surface and Tournament.surface descriptions now state the per-tie rule for team competitions instead of describing the null as permanent.",
+ "content_html": "
Fixed
\n
\n
Team-competition matches now carry the surface of their tie. Davis Cup, Billie Jean King Cup and Laver Cup matches published surface: null on the match and on tournament.surface — 268 of 268 team-event matches in the 60 days to 2026-09-20. 1.13.25 documented that null as structural; it was only structural for the data we held. Every surface we held was per tournament, and a team "tournament" (ATP Davis Cup - World Group I) is a tier with a draw stage, not a surface; a tie is played on whatever its host nation chose. Each tie's published record states the surface, venue and indoor/outdoor flag, so the API now holds one row per tie and resolves each match to its tie by competition, date window and the nations on court — answering only when a single tie remains, never from a nation, a name or a previous tie. Existing team-event matches (last 60 days and upcoming) were backfilled with one audit row per write; new ties are picked up as they are drawn. Ties whose venue the ITF has not announced stay null. tournament.surface for a team competition remains null by design: the surface belongs to the tie, and it is published on the match. Raised by a public report that Davis Cup match 191867 returned surface: null; that match now serves hard. Payload shape unchanged.
\n
\n
Changed
\n
\n
Match.surface and Tournament.surface descriptions now state the per-tie rule for team competitions instead of describing the null as permanent.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Fixed",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.26",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "Team-competition matches now carry the surface of their tie. Davis Cup, Billie Jean King Cup and Laver Cup matches published surface: null on the match and on tournament.surface — 268 of 268 team-event matches in the 60 days to 2026-09-20. 1.13.25 documented that null as structural; it was only structural for the data we held. Every surface we held was per tournament, and a team \"tournament\" (ATP Davis Cup - World Group I) is a tier with a draw stage, not a surface; a tie is played on whatever its host nation chose. Each tie's published record states the surface, venue and indoor/outdoor flag, so the API now holds one row per tie and resolves each match to its tie by competition, date window and the nations on court — answering only when a single tie remains, never from a nation, a name or a previous tie. Existing team-event matches (last 60 days and upcoming) were backfilled with one audit row per write; new ties are picked up as they are drawn. Ties whose venue the ITF has not announced stay null. tournament.surface for a team competition remains null by design: the surface belongs to the tie, and it is published on the match. Raised by a public report that Davis Cup match 191867 returned surface: null; that match now serves hard. Payload shape unchanged."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Match.surface and Tournament.surface descriptions now state the per-tie rule for team competitions instead of describing the null as permanent."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.25",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.25",
+ "title": "1.13.25: A match that had been given a tournament_id was never given the surface that id serves, and 1 more change",
+ "summary": "Fixed: A match that had been given a tournament_id was never given the surface that id serves. Changed: Match.surface and Tournament.surface now say when they are null and why.",
+ "content_html": "
Fixed
\n
\n
A match that had been given a tournament_id was never given the surface that id serves.matches.tournament_key was backfilled on 2026-09-04 for matches that had been publishing tournament_id: null; the pass gave those rows an id and stopped there. The tournament-surface fallback runs only in the ingest upsert, and every repaired row was already terminal, so it never ran again: 2,533 matches (2026-06-21..2026-09-04, all ITF) published surface: null while GET /tournaments/{id} — for the id that same row hands you — answered "surface": "hard". Two surfaces for one question, each written by a different writer. Measured and wire-confirmed before the repair on match 186637 (surface: null, tournament_id: "12158") against /tournaments/12158 ("surface": "hard"), and after it on the same row. All 2,533 now carry the surface their own tournament states; indoor was filled from the same tournament row where the match held none. Nothing after 2026-09-04 was affected — the same commit's ingest hook sets the key before the fallback reads it, so the forward path has been correct throughout and the weekly count is flat zero since. The repair resolves surface through the same chain the ingest uses (tournament-name map first, tournament row second), so a later upsert cannot answer one of these rows differently. Match-level surface null over the previous 90 days: 19.8% before, 11.8% after. No field or endpoint added.
\n
\n
Changed
\n
\n
Match.surface and Tournament.surface now say when they are null and why. Both were documented as nullable with no account of the population, while the tournament_id field beside them carries a measured one. The remaining match-level nulls are two structural groups and an 8-row residue: UTR events, which are not in our tournament list at all (and so publish tournament_id: null too), and TEAM COMPETITIONS — Davis Cup, Billie Jean King Cup, Laver Cup — where the surface is the host nation's choice per tie, so neither the season-long tournament row nor the match has one to state. Raised by a public report that Davis Cup matches return surface: null. For those 35 tournament rows a value outside hard/clay/grass is rejected rather than published. No behaviour change.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Fixed",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.25",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "A match that had been given a tournament_id was never given the surface that id serves. matches.tournament_key was backfilled on 2026-09-04 for matches that had been publishing tournament_id: null; the pass gave those rows an id and stopped there. The tournament-surface fallback runs only in the ingest upsert, and every repaired row was already terminal, so it never ran again: 2,533 matches (2026-06-21..2026-09-04, all ITF) published surface: null while GET /tournaments/{id} — for the id that same row hands you — answered \"surface\": \"hard\". Two surfaces for one question, each written by a different writer. Measured and wire-confirmed before the repair on match 186637 (surface: null, tournament_id: \"12158\") against /tournaments/12158 (\"surface\": \"hard\"), and after it on the same row. All 2,533 now carry the surface their own tournament states; indoor was filled from the same tournament row where the match held none. Nothing after 2026-09-04 was affected — the same commit's ingest hook sets the key before the fallback reads it, so the forward path has been correct throughout and the weekly count is flat zero since. The repair resolves surface through the same chain the ingest uses (tournament-name map first, tournament row second), so a later upsert cannot answer one of these rows differently. Match-level surface null over the previous 90 days: 19.8% before, 11.8% after. No field or endpoint added."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Match.surface and Tournament.surface now say when they are null and why. Both were documented as nullable with no account of the population, while the tournament_id field beside them carries a measured one. The remaining match-level nulls are two structural groups and an 8-row residue: UTR events, which are not in our tournament list at all (and so publish tournament_id: null too), and TEAM COMPETITIONS — Davis Cup, Billie Jean King Cup, Laver Cup — where the surface is the host nation's choice per tie, so neither the season-long tournament row nor the match has one to state. Raised by a public report that Davis Cup matches return surface: null. For those 35 tournament rows a value outside hard/clay/grass is rejected rather than published. No behaviour change."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.24",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.24",
+ "title": "1.13.24: GET /matches/{matchId}/score no longer answers 404 for a settled match whose live-score rows were retired by the 90-day retention sweep, or that only ever had…",
+ "summary": "Changed: GET /matches/{matchId}/score no longer answers 404 for a settled match whose live-score rows were retired by the 90-day retention sweep, or that only ever had an archived final.",
+ "content_html": "
Changed
\n
\n
GET /matches/{matchId}/score no longer answers 404 for a settled match whose live-score rows were retired by the 90-day retention sweep, or that only ever had an archived final. When there is no live row at all, a match with a non-null outcome (completed, retired, walkover, default, abandoned, unresolved) is answered from its archived final — the same read GET /matches?status=completed already embeds, through the same serializer — so the listing and the single-match read can no longer disagree about whether a score exists. Unchanged: a live match always reads its live tape (the archive never outranks it); an upcoming match with no row, a cancelled match that was never played, and a settled match with nothing recorded anywhere still answer 404. An archived final carries age_seconds, observed_age_seconds, sources_count and accepted_at as null — no clock is claimed for a state nobody watched — and timestamp null where the archived row has none; the Score object still carries no data-source label, GET /history/matches/{matchId}meta.point_source says whether the final was observed or reconstructed. Behaviour change on the API side 2026-09-20; measured at the change: 13,437 completed matches from the previous 180 days, 157,170 all time, now answer /score instead of 404. No field or endpoint added.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.24",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "GET /matches/{matchId}/score no longer answers 404 for a settled match whose live-score rows were retired by the 90-day retention sweep, or that only ever had an archived final. When there is no live row at all, a match with a non-null outcome (completed, retired, walkover, default, abandoned, unresolved) is answered from its archived final — the same read GET /matches?status=completed already embeds, through the same serializer — so the listing and the single-match read can no longer disagree about whether a score exists. Unchanged: a live match always reads its live tape (the archive never outranks it); an upcoming match with no row, a cancelled match that was never played, and a settled match with nothing recorded anywhere still answer 404. An archived final carries age_seconds, observed_age_seconds, sources_count and accepted_at as null — no clock is claimed for a state nobody watched — and timestamp null where the archived row has none; the Score object still carries no data-source label, GET /history/matches/{matchId} meta.point_source says whether the final was observed or reconstructed. Behaviour change on the API side 2026-09-20; measured at the change: 13,437 completed matches from the previous 180 days, 157,170 all time, now answer /score instead of 404. No field or endpoint added."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.23",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.23",
+ "title": "1.13.23: GET /rankings accepted ?surface=, ?min_matches=, ?activity_weeks= and a contradicting ?tour= on the official systems, and silently ignored all four",
+ "summary": "Fixed: GET /rankings accepted ?surface=, ?min_matches=, ?activity_weeks= and a contradicting ?tour= on the official systems, and silently ignored all four.",
+ "content_html": "
Fixed
\n
\n
GET /rankings accepted ?surface=, ?min_matches=, ?activity_weeks= and a contradicting ?tour= on the official systems, and silently ignored all four. Measured on production before the fix: ?system=atp&surface=clay, ?system=atp&min_matches=200, ?system=atp&activity_weeks=4 and ?system=atp&tour=wta each returned 200 and the plain overall ATP table, byte-identical to the same call with no parameter at all; ?system=atp&tour=zzz was accepted too. All four are documented "Elo only" / "Elo listing" and are read nowhere else, so a caller asking for a clay ATP leaderboard received the overall one with nothing in the response to say the filter had not run. The elo branch was correct throughout and is the control that makes this a defect rather than an empty population - ?system=elo&tour=atp&surface=clay returns a genuinely different top 10 from surface=hard, and surface=zzz is refused with the valid list. Unlike the has_market hole fixed in 1.13.21, this one cannot be closed by making the filter act: there is no clay ATP ranking and no WTA row in an ATP table. surface, min_matches and activity_weeks are now a 400 when no requested system is elo (a mixed ?system=atp&system=elo&surface=clay still passes); tour is a 400 only when it contradicts a system the caller NAMED, or when that system is not an ATP/WTA walk. ?system=atp&tour=atp is redundant rather than wrong and still answers, and the implicit default system set is unchanged.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.23",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "GET /rankings accepted ?surface=, ?min_matches=, ?activity_weeks= and a contradicting ?tour= on the official systems, and silently ignored all four. Measured on production before the fix: ?system=atp&surface=clay, ?system=atp&min_matches=200, ?system=atp&activity_weeks=4 and ?system=atp&tour=wta each returned 200 and the plain overall ATP table, byte-identical to the same call with no parameter at all; ?system=atp&tour=zzz was accepted too. All four are documented \"Elo only\" / \"Elo listing\" and are read nowhere else, so a caller asking for a clay ATP leaderboard received the overall one with nothing in the response to say the filter had not run. The elo branch was correct throughout and is the control that makes this a defect rather than an empty population - ?system=elo&tour=atp&surface=clay returns a genuinely different top 10 from surface=hard, and surface=zzz is refused with the valid list. Unlike the has_market hole fixed in 1.13.21, this one cannot be closed by making the filter act: there is no clay ATP ranking and no WTA row in an ATP table. surface, min_matches and activity_weeks are now a 400 when no requested system is elo (a mixed ?system=atp&system=elo&surface=clay still passes); tour is a 400 only when it contradicts a system the caller NAMED, or when that system is not an ATP/WTA walk. ?system=atp&tour=atp is redundant rather than wrong and still answers, and the implicit default system set is unchanged."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.22",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.22",
+ "title": "1.13.22: GET /history/archive/players now takes ?id= (alias ?player_id=), so the corpus person id we publish can finally be looked up, and 2 more changes",
+ "summary": "Added: GET /history/archive/players now takes ?id= (alias ?player_id=), so the corpus person id we publish can finally be looked up. Fixed: A corpus person id sent to GET /players/{playerId} returned a bare 404 that explained nothing. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
GET /history/archive/players now takes ?id= (alias ?player_id=), so the corpus person id we publish can finally be looked up. Archive match rows have published that id as winner.player_id / loser.player_id since 2026-08-03, and this endpoint has published it as id and described it in those words — but the only surface accepting the value was /rankings?archive_player= — ULTRA, system=elo only, and it answers with RATINGS, not the person. So the PERSON could not be reached by id at all: a reader holding a corpus id could only go back to the NAME. It is a filter on the list rather than a /history/archive/players/{id} detail route on purpose: the natural key is (tour, source_pid), the two tours number independently, and 14,627 corpus person ids are live in BOTH (measured 2026-09-20) — a detail route would have to pick a tour and would return the wrong human for those ids without saying so. The list returns every person wearing the id; add tour to narrow it to exactly one. Sending id and player_id with different values is a 400.
\n
\n
Fixed
\n
\n
A corpus person id sent to GET /players/{playerId} returned a bare 404 that explained nothing. The roster (/players, ids 1-37033) and the results-archive person registry (100001-270580) are separate, deliberately disjoint id spaces - 0 of 137,483 corpus ids resolve on /players/{playerId} - and the only id-shaped route with player in its name was the roster one. The 404 now carries detail and a see pointer at /history/archive/players?id=... when the id resolves in the archive registry; an id from neither space keeps the plain body. The same signpost is on /players/{playerId}/stoppages and /players/{playerId}/injuries. The status is deliberately still 404 and not 410: 410 asserts the id once existed HERE, and it never did. The person is never named in the 404 - the archive is History-gated and /players/{playerId} is not.
\n
GET /players/{playerId} was missing its 410 in this specification. The route has answered 410 PlayerMerged with a forwarding address for a merged or retired player id for as long as its two stoppage aliases have, and both aliases documented it while the main detail endpoint did not. Now documented, along with a PlayerNotFound schema that states exactly when detail and see are present.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Added",
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.22",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /history/archive/players now takes ?id= (alias ?player_id=), so the corpus person id we publish can finally be looked up. Archive match rows have published that id as winner.player_id / loser.player_id since 2026-08-03, and this endpoint has published it as id and described it in those words — but the only surface accepting the value was /rankings?archive_player= — ULTRA, system=elo only, and it answers with RATINGS, not the person. So the PERSON could not be reached by id at all: a reader holding a corpus id could only go back to the NAME. It is a filter on the list rather than a /history/archive/players/{id} detail route on purpose: the natural key is (tour, source_pid), the two tours number independently, and 14,627 corpus person ids are live in BOTH (measured 2026-09-20) — a detail route would have to pick a tour and would return the wrong human for those ids without saying so. The list returns every person wearing the id; add tour to narrow it to exactly one. Sending id and player_id with different values is a 400."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Fixed",
+ "items": [
+ "A corpus person id sent to GET /players/{playerId} returned a bare 404 that explained nothing. The roster (/players, ids 1-37033) and the results-archive person registry (100001-270580) are separate, deliberately disjoint id spaces - 0 of 137,483 corpus ids resolve on /players/{playerId} - and the only id-shaped route with player in its name was the roster one. The 404 now carries detail and a see pointer at /history/archive/players?id=... when the id resolves in the archive registry; an id from neither space keeps the plain body. The same signpost is on /players/{playerId}/stoppages and /players/{playerId}/injuries. The status is deliberately still 404 and not 410: 410 asserts the id once existed HERE, and it never did. The person is never named in the 404 - the archive is History-gated and /players/{playerId} is not.",
+ "GET /players/{playerId} was missing its 410 in this specification. The route has answered 410 PlayerMerged with a forwarding address for a merged or retired player id for as long as its two stoppage aliases have, and both aliases documented it while the main detail endpoint did not. Now documented, along with a PlayerNotFound schema that states exactly when detail and see are present."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.21",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.21",
+ "title": "1.13.21: GET /matches accepted ?has_market= and silently ignored it, and status=cancelled ignored ?has_analysis= as well, and 1 more change",
+ "summary": "Fixed: GET /matches accepted ?has_market= and silently ignored it, and status=cancelled ignored ?has_analysis= as well. Added: has_analysis and has_market are now DOCUMENTED as query parameters on GET /matches. has_analysis has worked on live, upcoming and completed since 2026-09-14 and has_market works everywhere from today, but neither appeared in this specification, so the only filter a reader could find was has_market on /history/matches.",
+ "content_html": "
Fixed
\n
\n
GET /matches accepted ?has_market= and silently ignored it, and status=cancelled ignored ?has_analysis= as well. The filter was implemented in the shared query builder on 2026-09-15 and exposed on /history/matches the same day, but it was never wired to /matches: every row published has_market, the query string was accepted, and the unfiltered page came back under a 200. Measured on production before the fix — status=live returned the same 14 rows (2 with a market) for has_market=true, has_market=false and no filter at all; status=completed the same 100 rows (38 with a market); /history/matches was correct throughout (40/40 and 0/40). Five of the eight status x filter combinations returned a wrong answer under a success code. Both filters now apply on every status, and an unparseable value is a 400 bad_has_market / bad_has_analysis as documented.
\n
\n
Added
\n
\n
has_analysis and has_market are now DOCUMENTED as query parameters on GET /matches. has_analysis has worked on live, upcoming and completed since 2026-09-14 and has_market works everywhere from today, but neither appeared in this specification, so the only filter a reader could find was has_market on /history/matches. The prose telling callers to "filter the slate first" was therefore advice with no documented instrument behind it.
\n
",
+ "date_published": "2026-09-20T00:00:00Z",
+ "tags": [
+ "Fixed",
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.21",
+ "date": "2026-09-20",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "GET /matches accepted ?has_market= and silently ignored it, and status=cancelled ignored ?has_analysis= as well. The filter was implemented in the shared query builder on 2026-09-15 and exposed on /history/matches the same day, but it was never wired to /matches: every row published has_market, the query string was accepted, and the unfiltered page came back under a 200. Measured on production before the fix — status=live returned the same 14 rows (2 with a market) for has_market=true, has_market=false and no filter at all; status=completed the same 100 rows (38 with a market); /history/matches was correct throughout (40/40 and 0/40). Five of the eight status x filter combinations returned a wrong answer under a success code. Both filters now apply on every status, and an unparseable value is a 400 bad_has_market / bad_has_analysis as documented."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Added",
+ "items": [
+ "has_analysis and has_market are now DOCUMENTED as query parameters on GET /matches. has_analysis has worked on live, upcoming and completed since 2026-09-14 and has_market works everywhere from today, but neither appeared in this specification, so the only filter a reader could find was has_market on /history/matches. The prose telling callers to \"filter the slate first\" was therefore advice with no documented instrument behind it."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.20",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.20",
+ "title": "1.13.20: The two win-probability regime boundaries published in 1.13.19 were three hours late, and are corrected here: 2026-08-18T08:29:28Z and 2026-09-16T07:59:54Z, and 2 more changes",
+ "summary": "Fixed: The two win-probability regime boundaries published in 1.13.19 were three hours late, and are corrected here: 2026-08-18T08:29:28Z and 2026-09-16T07:59:54Z. With it, the measured lag of the version string: the +itfdraw-2026-09-16 suffix first appears on rows generated 2026-09-17T05:04:25Z, which is 21 h 04 m after the rule landed, not 18 h 02 m. Plus 1 more change.",
+ "content_html": "
Fixed
\n
\n
The two win-probability regime boundaries published in 1.13.19 were three hours late, and are corrected here: 2026-08-18T08:29:28Z and 2026-09-16T07:59:54Z. Both switch instants were read off config.updated_at, a column this application stores in naive local wall time (UTC+3), and were published as if they were UTC. Verified six independent ways on 2026-09-19: six config rows whose own value is an epoch each read exactly +3 h against their updated_at (e.g. monitoring.worker_heartbeat_ts = 2026-09-19T20:22:46.465Z stored as 23:22:46.467). The corrected instants are the config switches themselves — win_probability.match_tiebreak_detection_enabled at 2026-08-18 08:29:28.93Z and win_probability.itf_pricing_draw_rule_enabled at 2026-09-16 07:59:54.76Z — which is the earliest instant from which a state can carry the new rule, and therefore the safe place to cut.
\n
With it, the measured lag of the version string: the +itfdraw-2026-09-16 suffix first appears on rows generated 2026-09-17T05:04:25Z, which is 21 h 04 m after the rule landed, not 18 h 02 m. That figure had been derived from the same three-hour-late instant. 2026-09-17T05:04:25Z and 2026-09-12T12:08:17Z were both taken from win_probability_meta.generated_at, a timezone-aware column, and are unchanged and correct.
\n
Anyone who segregated a corpus on 2026-08-18T11:29:28Z or 2026-09-16T11:02:00Z should re-cut: the three hours before each corrected instant are on the wrong side of the split.
\n
",
+ "date_published": "2026-09-19T00:00:00Z",
+ "tags": [
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.20",
+ "date": "2026-09-19",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "The two win-probability regime boundaries published in 1.13.19 were three hours late, and are corrected here: 2026-08-18T08:29:28Z and 2026-09-16T07:59:54Z. Both switch instants were read off config.updated_at, a column this application stores in naive local wall time (UTC+3), and were published as if they were UTC. Verified six independent ways on 2026-09-19: six config rows whose own value is an epoch each read exactly +3 h against their updated_at (e.g. monitoring.worker_heartbeat_ts = 2026-09-19T20:22:46.465Z stored as 23:22:46.467). The corrected instants are the config switches themselves — win_probability.match_tiebreak_detection_enabled at 2026-08-18 08:29:28.93Z and win_probability.itf_pricing_draw_rule_enabled at 2026-09-16 07:59:54.76Z — which is the earliest instant from which a state can carry the new rule, and therefore the safe place to cut.",
+ "With it, the measured lag of the version string: the +itfdraw-2026-09-16 suffix first appears on rows generated 2026-09-17T05:04:25Z, which is 21 h 04 m after the rule landed, not 18 h 02 m. That figure had been derived from the same three-hour-late instant. 2026-09-17T05:04:25Z and 2026-09-12T12:08:17Z were both taken from win_probability_meta.generated_at, a timezone-aware column, and are unchanged and correct.",
+ "Anyone who segregated a corpus on 2026-08-18T11:29:28Z or 2026-09-16T11:02:00Z should re-cut: the three hours before each corrected instant are on the wrong side of the split."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.19",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.19",
+ "title": "1.13.19: Score.win_probability_meta: the two win-probability regime boundaries are published here for the first time, so a customer can segregate a recorded corpus…, and 3 more changes",
+ "summary": "Changed: Score.win_probability_meta: the two win-probability regime boundaries are published here for the first time, so a customer can segregate a recorded corpus without asking us. The same description states, newly measured, that model_version is not a safe discriminator across the second boundary: the +itfdraw-2026-09-16 suffix first appears on rows generated 2026-09-17T05:04:25Z, 18 h 02 m after the rule landed (corrected to 21 h 04 m in 1.13.20). Plus 2 more changes.",
+ "content_html": "
Changed
\n
\n
Score.win_probability_meta: the two win-probability regime boundaries are published here for the first time, so a customer can segregate a recorded corpus without asking us. Two changes moved published live probabilities before model_version reflected them — 2026-08-18T11:29:28Z, when the over-inclusive deciding-set rule for lower-tier singles began, and 2026-09-16T11:02:00Z, when the draw-based rule replaced it (BOTH INSTANTS WERE THREE HOURS LATE AND ARE CORRECTED IN 1.13.20 — use 08:29:28Z and 07:59:54Z) (up to 0.20 on affected deciding-set states) — so the cut is on generated_at/timestamp, never on the version string. Published in the application's internal reference on 2026-09-17 and not ported here until now; docs.livetennisapi.com, which is the artifact a customer opens, carried none of it.
\n
The same description states, newly measured, that model_version is not a safe discriminator across the second boundary: the +itfdraw-2026-09-16 suffix first appears on rows generated 2026-09-17T05:04:25Z, 18 h 02 m after the rule landed (corrected to 21 h 04 m in 1.13.20). Measured against production on 2026-09-19: in that gap 33,510 states across 351 matches — 9,719 states over 113 matches at the ITF M15/W15/W35 levels the rule governs — were computed under the corrected rule while still carrying markov-population-2026-08-23. A cut on a whole-day boundary mislabels exactly those rows; a cut at the 11:02:00Z instant does not. The previous internal wording ("timestamp only up to 2026-09-17") was right to the day and vague by eighteen hours.
\n
Score.win_probability_meta.model_version now carries a description of its own, and the pre-stamp caveat is stated with its instant: no row generated before 2026-09-12T12:08:17Z carries model_version or generated_at at all.
\n
HistoryTapeRow.win_probability_p1: documented that the tape carries no model-regime stamp — no model_version or generated_at column here or in the bulk tape packages — so a recorded tape corpus is segregated by the row's own timestamp against those two instants.
\n
\n
Documentation only. No field, endpoint or wire value changed.
",
+ "date_published": "2026-09-19T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.19",
+ "date": "2026-09-19",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "Score.win_probability_meta: the two win-probability regime boundaries are published here for the first time, so a customer can segregate a recorded corpus without asking us. Two changes moved published live probabilities before model_version reflected them — 2026-08-18T11:29:28Z, when the over-inclusive deciding-set rule for lower-tier singles began, and 2026-09-16T11:02:00Z, when the draw-based rule replaced it (BOTH INSTANTS WERE THREE HOURS LATE AND ARE CORRECTED IN 1.13.20 — use 08:29:28Z and 07:59:54Z) (up to 0.20 on affected deciding-set states) — so the cut is on generated_at/timestamp, never on the version string. Published in the application's internal reference on 2026-09-17 and not ported here until now; docs.livetennisapi.com, which is the artifact a customer opens, carried none of it.",
+ "The same description states, newly measured, that model_version is not a safe discriminator across the second boundary: the +itfdraw-2026-09-16 suffix first appears on rows generated 2026-09-17T05:04:25Z, 18 h 02 m after the rule landed (corrected to 21 h 04 m in 1.13.20). Measured against production on 2026-09-19: in that gap 33,510 states across 351 matches — 9,719 states over 113 matches at the ITF M15/W15/W35 levels the rule governs — were computed under the corrected rule while still carrying markov-population-2026-08-23. A cut on a whole-day boundary mislabels exactly those rows; a cut at the 11:02:00Z instant does not. The previous internal wording (\"timestamp only up to 2026-09-17\") was right to the day and vague by eighteen hours.",
+ "Score.win_probability_meta.model_version now carries a description of its own, and the pre-stamp caveat is stated with its instant: no row generated before 2026-09-12T12:08:17Z carries model_version or generated_at at all.",
+ "HistoryTapeRow.win_probability_p1: documented that the tape carries no model-regime stamp — no model_version or generated_at column here or in the bulk tape packages — so a recorded tape corpus is segregated by the row's own timestamp against those two instants."
+ ],
+ "notes": [
+ "Documentation only. No field, endpoint or wire value changed."
+ ]
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.18",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.18",
+ "title": "1.13.18: LivePoint.number (GET /matches/{matchId}/points, the point frame, the webhook point event): the field is documented as null when we joined the game already in…, and 1 more change",
+ "summary": "Fixed: LivePoint.number (GET /matches/{matchId}/points, the point frame, the webhook point event): the field is documented as null when we joined the game already in progress, and the API now writes null there instead of 0. Changed: Documented, in all three places the replay guidance appears, that (set, game, number) orders points and does not identify them: the tuple may repeat, and number may be null.",
+ "content_html": "
Fixed
\n
\n
LivePoint.number (GET /matches/{matchId}/points, the point frame, the webhook point event): the field is documented as null when we joined the game already in progress, and the API now writes null there instead of 0. 0 asserts a game's opening state, and on a game we picked up mid-way that is false — how many points it already holds is not derivable from the score, because the deuce zone maps many ordinals onto one score, and it is never guessed. Measured on 2026-09-19, 1,341 rows over three days carried number: 0 at a non-love score. A second defect in the same computation is fixed with it: a late-arriving row for a *different* game committed between two points of the game in progress reset the ordinal, so on match 191801 seq 65 and 67 were both (set 2, game 1, number 0) — at 15-15 and 15-40 — with seq 66, a point of set 1 game 9, between them. The ordinal now continues within its own game regardless of what arrived in between. The schema already permitted null; only the description and the written values change.
\n
\n
Changed
\n
\n
Documented, in all three places the replay guidance appears, that (set, game, number)orders points and does not identify them: the tuple may repeat, and number may be null. A repeat on the live basis is a re-statement of a game by a second source, not a correction — there is no revision id, superseded-seq, version or correction flag, because rows are append-only and never rewritten. after_seq therefore never needs a full refetch, and the page-level quality field already reads revised when a page contains such a re-statement. seq remains the only unique, stable per-row key.
\n
",
+ "date_published": "2026-09-19T00:00:00Z",
+ "tags": [
+ "Fixed",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.18",
+ "date": "2026-09-19",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "LivePoint.number (GET /matches/{matchId}/points, the point frame, the webhook point event): the field is documented as null when we joined the game already in progress, and the API now writes null there instead of 0. 0 asserts a game's opening state, and on a game we picked up mid-way that is false — how many points it already holds is not derivable from the score, because the deuce zone maps many ordinals onto one score, and it is never guessed. Measured on 2026-09-19, 1,341 rows over three days carried number: 0 at a non-love score. A second defect in the same computation is fixed with it: a late-arriving row for a *different* game committed between two points of the game in progress reset the ordinal, so on match 191801 seq 65 and 67 were both (set 2, game 1, number 0) — at 15-15 and 15-40 — with seq 66, a point of set 1 game 9, between them. The ordinal now continues within its own game regardless of what arrived in between. The schema already permitted null; only the description and the written values change."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Documented, in all three places the replay guidance appears, that (set, game, number) orders points and does not identify them: the tuple may repeat, and number may be null. A repeat on the live basis is a re-statement of a game by a second source, not a correction — there is no revision id, superseded-seq, version or correction flag, because rows are append-only and never rewritten. after_seq therefore never needs a full refetch, and the page-level quality field already reads revised when a page contains such a re-statement. seq remains the only unique, stable per-row key."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.17",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.17",
+ "title": "1.13.17: Native WebSocket close codes, documented for the first time, and 1 more change",
+ "summary": "Added: Native WebSocket close codes, documented for the first time. Changed: GET /matches/{matchId}/points, the point frame and the points signal: seq is arrival order, not match order, on the live basis.",
+ "content_html": "
Added
\n
\n
Native WebSocket close codes, documented for the first time. Every refusal sends its error frame and then closes with a code that says what to do next — 1013 Try Again Later for transient refusals (connection_limit:per_key, connection_limit:server, service_unavailable), 1008 Policy Violation for anything a retry cannot fix (unauthorized, upgrade_required, email_unverified, client_blocked, bad_json, no_topics, and any mid-stream loss of access), 1012 on a restart, 1000 on a normal close. The close reason repeats the frame's error string, so (code, reason) is a complete diagnosis even if the frame was missed. Behaviour change the same day: refusals raised during the *handshake* previously closed 1000 with an empty reason, indistinguishable from an orderly shutdown, so a client awaiting its subscribed ack saw only a normal close. The error frame was, and still is, delivered before the close; only the close code and reason changed.
\n
\n
Changed
\n
\n
GET /matches/{matchId}/points, the point frame and the points signal: seq is arrival order, not match order, on the live basis. It is assigned in commit order, and points of a live match arrive by more than one path at different speeds, so a point from a set that has just ended can be committed after points from the set that follows it and carry the higher seq. Each row is self-consistent; reading the tape in seq order can show the set or game counter step backwards. Measured over a recent seven-day window this affected a minority of live matches, and never the reconstruction basis. seq remains the right key for paging, dedup and resume, and is the wrong key for chronology — sort by (set, game, number) to replay in playing order. The spec previously said "ordered per match by seq", which reads as chronological. Documentation only; nothing on the wire changed.
\n
",
+ "date_published": "2026-09-18T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.17",
+ "date": "2026-09-18",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "Native WebSocket close codes, documented for the first time. Every refusal sends its error frame and then closes with a code that says what to do next — 1013 Try Again Later for transient refusals (connection_limit:per_key, connection_limit:server, service_unavailable), 1008 Policy Violation for anything a retry cannot fix (unauthorized, upgrade_required, email_unverified, client_blocked, bad_json, no_topics, and any mid-stream loss of access), 1012 on a restart, 1000 on a normal close. The close reason repeats the frame's error string, so (code, reason) is a complete diagnosis even if the frame was missed. Behaviour change the same day: refusals raised during the *handshake* previously closed 1000 with an empty reason, indistinguishable from an orderly shutdown, so a client awaiting its subscribed ack saw only a normal close. The error frame was, and still is, delivered before the close; only the close code and reason changed."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "GET /matches/{matchId}/points, the point frame and the points signal: seq is arrival order, not match order, on the live basis. It is assigned in commit order, and points of a live match arrive by more than one path at different speeds, so a point from a set that has just ended can be committed after points from the set that follows it and carry the higher seq. Each row is self-consistent; reading the tape in seq order can show the set or game counter step backwards. Measured over a recent seven-day window this affected a minority of live matches, and never the reconstruction basis. seq remains the right key for paging, dedup and resume, and is the wrong key for chronology — sort by (set, game, number) to replay in playing order. The spec previously said \"ordered per match by seq\", which reads as chronological. Documentation only; nothing on the wire changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.16",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.16",
+ "title": "1.13.16: GET /usage → today.remaining_day: documented as the day allowance less SERVED calls (calls - errors), and the API now computes it that way",
+ "summary": "Changed: GET /usage → today.remaining_day: documented as the day allowance less SERVED calls (calls - errors), and the API now computes it that way.",
+ "content_html": "
Changed
\n
\n
GET /usage → today.remaining_day: documented as the day allowance less SERVED calls (calls - errors), and the API now computes it that way. Refused requests have never spent the daily allowance — the 429 guidance in this spec already said so — but the endpoint subtracted gross calls, so it under-reported what a key could still send by exactly its error count. Measured on 2026-09-18, one ULTRA key was shown 14,360 fewer requests remaining than the API would have served it. today.calls and today.errors are unchanged: raw counters of everything the key sent.
\n
",
+ "date_published": "2026-09-18T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.16",
+ "date": "2026-09-18",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "GET /usage → today.remaining_day: documented as the day allowance less SERVED calls (calls - errors), and the API now computes it that way. Refused requests have never spent the daily allowance — the 429 guidance in this spec already said so — but the endpoint subtracted gross calls, so it under-reported what a key could still send by exactly its error count. Measured on 2026-09-18, one ULTRA key was shown 14,360 fewer requests remaining than the API would have served it. today.calls and today.errors are unchanged: raw counters of everything the key sent."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.15",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.15",
+ "title": "1.13.15: event_status (match object): the enum omitted Finished and Unresolved, and 2 more changes",
+ "summary": "Fixed: event_status (match object): the enum omitted Finished and Unresolved. Changed: event_status description: \"NULL means the match completed normally\" was wrong in the direction that matters — null means nothing was ever stated for the match, which covers matches that ran their course and matches nothing was ever said about alike. Plus 1 more change.",
+ "content_html": "
Fixed
\n
\n
event_status (match object): the enum omitted Finished and Unresolved. Finished is the value the field carries on a normally-completed match and is by far its most common — 144,266 of the 150,678 rows carrying one, measured on 2026-09-18 — so a client generated from this spec with strict enum validation rejected the majority of completed matches. Both values are listed now. The API has always published them; only the spec was wrong, so nothing on the wire changed.
\n
\n
Changed
\n
\n
event_status description: "NULL means the match completed normally" was wrong in the direction that matters — null means nothing was ever stated for the match, which covers matches that ran their course and matches nothing was ever said about alike. Branch on outcome, not on the absence of a badge.
\n
event_status description: event_status: Finished while status is still live is documented as the pending-final state (one source has called the match over, the final is not confirmed). The score stands still through it — stale true, age_seconds climbing. Measured over the seven days to 2026-09-18, across 649 matches, that gap closed in 111 s at the median and 911 s at the ninetieth percentile; 85 ran past ten minutes.
\n
",
+ "date_published": "2026-09-18T00:00:00Z",
+ "tags": [
+ "Fixed",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.15",
+ "date": "2026-09-18",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "event_status (match object): the enum omitted Finished and Unresolved. Finished is the value the field carries on a normally-completed match and is by far its most common — 144,266 of the 150,678 rows carrying one, measured on 2026-09-18 — so a client generated from this spec with strict enum validation rejected the majority of completed matches. Both values are listed now. The API has always published them; only the spec was wrong, so nothing on the wire changed."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "event_status description: \"NULL means the match completed normally\" was wrong in the direction that matters — null means nothing was ever stated for the match, which covers matches that ran their course and matches nothing was ever said about alike. Branch on outcome, not on the absence of a badge.",
+ "event_status description: event_status: Finished while status is still live is documented as the pending-final state (one source has called the match over, the final is not confirmed). The score stands still through it — stale true, age_seconds climbing. Measured over the seven days to 2026-09-18, across 649 matches, that gap closed in 111 s at the median and 911 s at the ninetieth percentile; 85 ran past ten minutes."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.14",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.14",
+ "title": "1.13.14: Plan end: when a paid plan's period ends (or a renewal goes unpaid) the same key now drops to the FREE tier automatically instead of being switched off —…",
+ "summary": "Changed: Plan end: when a paid plan's period ends (or a renewal goes unpaid) the same key now drops to the FREE tier automatically instead of being switched off — nothing is re-issued, free endpoints keep answering at the free limits, paid endpoints answer 403 upgrade_required from that moment, and subscribing again lifts the same key.",
+ "content_html": "
Changed
\n
\n
Plan end: when a paid plan's period ends (or a renewal goes unpaid) the same key now drops to the FREE tier automatically instead of being switched off — nothing is re-issued, free endpoints keep answering at the free limits, paid endpoints answer 403 upgrade_required from that moment, and subscribing again lifts the same key. Behaviour change on the billing side effective 2026-09-17 19:59 UTC; keys of plans that ended in the previous 30 days were moved to FREE the same evening. No field or endpoint changed.
\n
",
+ "date_published": "2026-09-17T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.14",
+ "date": "2026-09-17",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "Plan end: when a paid plan's period ends (or a renewal goes unpaid) the same key now drops to the FREE tier automatically instead of being switched off — nothing is re-issued, free endpoints keep answering at the free limits, paid endpoints answer 403 upgrade_required from that moment, and subscribing again lifts the same key. Behaviour change on the billing side effective 2026-09-17 19:59 UTC; keys of plans that ended in the previous 30 days were moved to FREE the same evening. No field or endpoint changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.13",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.13",
+ "title": "1.13.13: GET /history/matches/{matchId} tiebreaks / meta.tiebreaks_source: the timing was stated wrongly",
+ "summary": "Changed: GET /history/matches/{matchId} tiebreaks / meta.tiebreaks_source: the timing was stated wrongly.",
+ "content_html": "
Changed
\n
\n
GET /history/matches/{matchId}tiebreaks / meta.tiebreaks_source: the timing was stated wrongly. The summary and reconstruction kinds were described as "recorded at completion"; they are written by a finals pass that runs once a day, so a match that finished earlier the same day commonly reads null for its 7-6 sets and carries them from the next pass onward. Only the tape kind is available the instant a match ends. Measured 2026-09-17: 72-100% of 7-6 sets populated on matches completed over the five previous days, 8% on matches completed the same day. Documentation only — no behaviour or field changed.
\n
",
+ "date_published": "2026-09-17T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.13",
+ "date": "2026-09-17",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "GET /history/matches/{matchId} tiebreaks / meta.tiebreaks_source: the timing was stated wrongly. The summary and reconstruction kinds were described as \"recorded at completion\"; they are written by a finals pass that runs once a day, so a match that finished earlier the same day commonly reads null for its 7-6 sets and carries them from the next pass onward. Only the tape kind is available the instant a match ends. Measured 2026-09-17: 72-100% of 7-6 sets populated on matches completed over the five previous days, 8% on matches completed the same day. Documentation only — no behaviour or field changed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.12",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.12",
+ "title": "1.13.12: GET /history/matches/{matchId} tiebreaks: per-set breaker finals are now filled from the point-by-point reconstruction and from the sources' set summaries when…",
+ "summary": "Changed: GET /history/matches/{matchId} tiebreaks: per-set breaker finals are now filled from the point-by-point reconstruction and from the sources' set summaries when the live tape stopped one point short (measured before this change: null on about 99% of 7-6 sets — 297 measured, 3 populated).",
+ "content_html": "
Changed
\n
\n
GET /history/matches/{matchId}tiebreaks: per-set breaker finals are now filled from the point-by-point reconstruction and from the sources' set summaries when the live tape stopped one point short (measured before this change: null on about 99% of 7-6 sets — 297 measured, 3 populated). New meta.tiebreaks_source says, per set, whether the final came from tape, summary or reconstruction. The response shape is unchanged.
\n
",
+ "date_published": "2026-09-16T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.12",
+ "date": "2026-09-16",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "GET /history/matches/{matchId} tiebreaks: per-set breaker finals are now filled from the point-by-point reconstruction and from the sources' set summaries when the live tape stopped one point short (measured before this change: null on about 99% of 7-6 sets — 297 measured, 3 populated). New meta.tiebreaks_source says, per set, whether the final came from tape, summary or reconstruction. The response shape is unchanged."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.11",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.11",
+ "title": "1.13.11: GET /players/{playerId}/stoppages (PRO) and its alias GET /players/{playerId}/injuries: one player's in-match stoppages (medical timeouts, trainer calls…",
+ "summary": "Added: GET /players/{playerId}/stoppages (PRO) and its alias GET /players/{playerId}/injuries: one player's in-match stoppages (medical timeouts, trainer calls; toilet breaks, pauses and other stoppages on request) plus the matches the player retired from or gave a walkover, newest first, over a window that defaults to the last 180 days. meta.latest_medical_timeout / previous_medical_timeout answer \"the latest and the one before\"; meta.record_starts states where each record begins (stoppage events from 2026-09-12; outcomes from 2026-08-18).",
+ "content_html": "
Added
\n
\n
GET /players/{playerId}/stoppages (PRO) and its alias GET /players/{playerId}/injuries: one player's in-match stoppages (medical timeouts, trainer calls; toilet breaks, pauses and other stoppages on request) plus the matches the player retired from or gave a walkover, newest first, over a window that defaults to the last 180 days. meta.latest_medical_timeout / previous_medical_timeout answer "the latest and the one before"; meta.record_starts states where each record begins (stoppage events from 2026-09-12; outcomes from 2026-08-18). In-match stoppages and match outcomes only; there is no off-court injury record.
\n
",
+ "date_published": "2026-09-16T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.11",
+ "date": "2026-09-16",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /players/{playerId}/stoppages (PRO) and its alias GET /players/{playerId}/injuries: one player's in-match stoppages (medical timeouts, trainer calls; toilet breaks, pauses and other stoppages on request) plus the matches the player retired from or gave a walkover, newest first, over a window that defaults to the last 180 days. meta.latest_medical_timeout / previous_medical_timeout answer \"the latest and the one before\"; meta.record_starts states where each record begins (stoppage events from 2026-09-12; outcomes from 2026-08-18). In-match stoppages and match outcomes only; there is no off-court injury record."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.10",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.10",
+ "title": "1.13.10: Score.sequence: a backwards move on GET /matches/{matchId}/score has two documented causes, not one — a withdrawn state, or the read deferring to a strictly…",
+ "summary": "Changed: Score.sequence: a backwards move on GET /matches/{matchId}/score has two documented causes, not one — a withdrawn state, or the read deferring to a strictly higher-trust source's fresh state (about 1% of live reads).",
+ "content_html": "
Changed
\n
\n
Score.sequence: a backwards move on GET /matches/{matchId}/score has two documented causes, not one — a withdrawn state, or the read deferring to a strictly higher-trust source's fresh state (about 1% of live reads). On the push feed and the native WebSocket the sequence only ever rises.
\n
",
+ "date_published": "2026-09-15T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.10",
+ "date": "2026-09-15",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "Score.sequence: a backwards move on GET /matches/{matchId}/score has two documented causes, not one — a withdrawn state, or the read deferring to a strictly higher-trust source's fresh state (about 1% of live reads). On the push feed and the native WebSocket the sequence only ever rises."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.9",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.9",
+ "title": "1.13.9: Score.accepted_at (string|null, live score reads and push frames): the instant we accepted this state, stamped once and never refreshed — the clock to…, and 2 more changes",
+ "summary": "Added: Score.accepted_at (string|null, live score reads and push frames): the instant we accepted this state, stamped once and never refreshed — the clock to difference a latency study against. Changed: Score.timestamp documented for what it is: our clock, stamped at acceptance and then refreshed (at most every 8 s) while the owning source keeps re-asserting an unchanged state, so on a live read it is usually the last-assertion instant. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
Score.accepted_at (string|null, live score reads and push frames): the instant we accepted this state, stamped once and never refreshed — the clock to difference a latency study against.
\n
\n
Changed
\n
\n
Score.timestamp documented for what it is: our clock, stamped at acceptance and then refreshed (at most every 8 s) while the owning source keeps re-asserting an unchanged state, so on a live read it is usually the last-assertion instant. It was described as "when this state was true"; it never was. Never an observation time from outside our process.
\n
The Score schema on this site now carries the full field set and descriptions: sequence, age_seconds, stale, observed_age_seconds, sources_count and detail were missing from the published table, and sets/server/is_tiebreak/win_probability_p1/danger/timestamp had no description.
\n
",
+ "date_published": "2026-09-15T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.9",
+ "date": "2026-09-15",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "Score.accepted_at (string|null, live score reads and push frames): the instant we accepted this state, stamped once and never refreshed — the clock to difference a latency study against."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Score.timestamp documented for what it is: our clock, stamped at acceptance and then refreshed (at most every 8 s) while the owning source keeps re-asserting an unchanged state, so on a live read it is usually the last-assertion instant. It was described as \"when this state was true\"; it never was. Never an observation time from outside our process.",
+ "The Score schema on this site now carries the full field set and descriptions: sequence, age_seconds, stale, observed_age_seconds, sources_count and detail were missing from the published table, and sets/server/is_tiebreak/win_probability_p1/danger/timestamp had no description."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.8",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.8",
+ "title": "1.13.8: Wording correction to 1.13.7: the pre-match hold rates enter the engine snapped to a 0.01 grid, not \"rounded to three decimals\"",
+ "summary": "Changed: Wording correction to 1.13.7: the pre-match hold rates enter the engine snapped to a 0.01 grid, not \"rounded to three decimals\".",
+ "content_html": "
Changed
\n
\n
Wording correction to 1.13.7: the pre-match hold rates enter the engine snapped to a 0.01 grid, not "rounded to three decimals". The mechanism and every field are unchanged; only the stated granularity was wrong.
\n
",
+ "date_published": "2026-09-15T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.8",
+ "date": "2026-09-15",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "Wording correction to 1.13.7: the pre-match hold rates enter the engine snapped to a 0.01 grid, not \"rounded to three decimals\". The mechanism and every field are unchanged; only the stated granularity was wrong."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.7",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.7",
+ "title": "1.13.7: win_probability_meta.market_anchored is documented for what it always was: the market-prior anchor applies to this match (a two-sided pre-match price and a…, and 1 more change",
+ "summary": "Changed: win_probability_meta.market_anchored is documented for what it always was: the market-prior anchor applies to this match (a two-sided pre-match price and a pre-play first score). Added: win_probability_meta.anchor_effective (boolean|null): this row's win_probability_p1 differs from win_probability_p1_model.",
+ "content_html": "
Changed
\n
\n
win_probability_meta.market_anchored is documented for what it always was: the market-prior anchor applies to this match (a two-sided pre-match price and a pre-play first score). It was described as "whether the anchor moved win_probability_p1", which is a per-row claim the flag never made — a small anchor shift can solve to a probability identical to win_probability_p1_model. Recorded values keep their meaning.
\n
\n
Added
\n
\n
win_probability_meta.anchor_effective (boolean|null): this row's win_probability_p1 differs from win_probability_p1_model. The per-row test win_probability_p1 != win_probability_p1_model works on frames recorded before this field existed.
\n
",
+ "date_published": "2026-09-15T00:00:00Z",
+ "tags": [
+ "Changed",
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.7",
+ "date": "2026-09-15",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "win_probability_meta.market_anchored is documented for what it always was: the market-prior anchor applies to this match (a two-sided pre-match price and a pre-play first score). It was described as \"whether the anchor moved win_probability_p1\", which is a per-row claim the flag never made — a small anchor shift can solve to a probability identical to win_probability_p1_model. Recorded values keep their meaning."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Added",
+ "items": [
+ "win_probability_meta.anchor_effective (boolean|null): this row's win_probability_p1 differs from win_probability_p1_model. The per-row test win_probability_p1 != win_probability_p1_model works on frames recorded before this field existed."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.6",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.6",
+ "title": "1.13.6: GET /history/matches/{id}/prices (PRO): the per-point price history — one row per played point with the score state the BASIC tape publishes and, per neutral…, and 3 more changes",
+ "summary": "Added: GET /history/matches/{id}/prices (PRO): the per-point price history — one row per played point with the score state the BASIC tape publishes and, per neutral side, the match-winner quote in force when the point was captured (bid, ask, mid, spread), with lag_seconds and an honest resolution label (tick, minute, coarse). GET /history/matches?has_market=true|false and a has_market flag on every history row, so a match with a price tape can be enumerated before it is pulled. Plus 2 more changes.",
+ "content_html": "
Added
\n
\n
GET /history/matches/{id}/prices (PRO): the per-point price history — one row per played point with the score state the BASIC tape publishes and, per neutral side, the match-winner quote in force when the point was captured (bid, ask, mid, spread), with lag_seconds and an honest resolution label (tick, minute, coarse). Works on a live match too; 404 no_market when no market is mapped.
\n
GET /history/matches?has_market=true|false and a has_market flag on every history row, so a match with a price tape can be enumerated before it is pulled.
\n
\n
Changed
\n
\n
Price retention: from 2026-09-15 the in-play ticks of a match with a mapped market are kept at full resolution and are not deleted; pre-match and idle ticks keep the existing tiers. Earlier ticks were already thinned.
\n
GET /matches/{id}/prices answers 404 no_market for a known match without a market (it said not_found).
\n
",
+ "date_published": "2026-09-15T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.6",
+ "date": "2026-09-15",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /history/matches/{id}/prices (PRO): the per-point price history — one row per played point with the score state the BASIC tape publishes and, per neutral side, the match-winner quote in force when the point was captured (bid, ask, mid, spread), with lag_seconds and an honest resolution label (tick, minute, coarse). Works on a live match too; 404 no_market when no market is mapped.",
+ "GET /history/matches?has_market=true|false and a has_market flag on every history row, so a match with a price tape can be enumerated before it is pulled."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Price retention: from 2026-09-15 the in-play ticks of a match with a mapped market are kept at full resolution and are not deleted; pre-match and idle ticks keep the existing tiers. Earlier ticks were already thinned.",
+ "GET /matches/{id}/prices answers 404 no_market for a known match without a market (it said not_found)."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.5",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.5",
+ "title": "1.13.5: pbp_coverage: \"point\" now means a per-point stream has delivered for the match — at least one played point past the seq 1 opener",
+ "summary": "Changed: pbp_coverage: \"point\" now means a per-point stream has delivered for the match — at least one played point past the seq 1 opener.",
+ "content_html": "
Changed
\n
\n
pbp_coverage: "point" now means a per-point stream has delivered for the match — at least one played point past the seq 1 opener. Until then it reads "game", including for a listed match whose stream holds only its opener because play has not started. Previously any point row, including the opener written before the first ball, was enough for "point", so a client told to gate on the field could subscribe to a match that never advanced. The admission gate we recommend is sequence > 1 together with stale: false. On the server, a match flagged live whose entire tape is still opener rows is demoted back to upcoming after twenty minutes, so such matches no longer linger in the live list.
\n
",
+ "date_published": "2026-09-14T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.5",
+ "date": "2026-09-14",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "pbp_coverage: \"point\" now means a per-point stream has delivered for the match — at least one played point past the seq 1 opener. Until then it reads \"game\", including for a listed match whose stream holds only its opener because play has not started. Previously any point row, including the opener written before the first ball, was enough for \"point\", so a client told to gate on the field could subscribe to a match that never advanced. The admission gate we recommend is sequence > 1 together with stale: false. On the server, a match flagged live whose entire tape is still opener rows is demoted back to upcoming after twenty minutes, so such matches no longer linger in the live list."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.4",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.4",
+ "title": "1.13.4: ITF World Tennis Tour qualifying rounds (singles) are now listed and live-scored: matches appear in GET /matches with round, tournament and scheduled_time…",
+ "summary": "Added: ITF World Tennis Tour qualifying rounds (singles) are now listed and live-scored: matches appear in GET /matches with round, tournament and scheduled_time, receive live scores while in play and complete on the official result.",
+ "content_html": "
Added
\n
\n
ITF World Tennis Tour qualifying rounds (singles) are now listed and live-scored: matches appear in GET /matches with round, tournament and scheduled_time, receive live scores while in play and complete on the official result. Sundays and Mondays, the qualifying days, were previously almost empty because the published schedule carries main draws only. GET /fixtures still follows that schedule and does not list qualifying.
\n
",
+ "date_published": "2026-09-14T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.4",
+ "date": "2026-09-14",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "ITF World Tennis Tour qualifying rounds (singles) are now listed and live-scored: matches appear in GET /matches with round, tournament and scheduled_time, receive live scores while in play and complete on the official result. Sundays and Mondays, the qualifying days, were previously almost empty because the published schedule carries main draws only. GET /fixtures still follows that schedule and does not list qualifying."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.3",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.3",
+ "title": "1.13.3: WTA / WTA 125 stoppages now arrive in two layers",
+ "summary": "Changed: WTA / WTA 125 stoppages now arrive in two layers.",
+ "content_html": "
Changed
\n
\n
WTA / WTA 125 stoppages now arrive in two layers. The match state is read within about 20 seconds and yields trainer_called, toilet_break_* and stoppage_* rows (player null). A later event layer with a variable delay — measured on 2026-09-13 from about one minute to an hour behind play — adds medical_timeout_start/end with the player and the exact instants. The 1.13.1 note implied the event layer was live; it is not.
\n
",
+ "date_published": "2026-09-13T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.3",
+ "date": "2026-09-13",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "WTA / WTA 125 stoppages now arrive in two layers. The match state is read within about 20 seconds and yields trainer_called, toilet_break_* and stoppage_* rows (player null). A later event layer with a variable delay — measured on 2026-09-13 from about one minute to an hour behind play — adds medical_timeout_start/end with the player and the exact instants. The 1.13.1 note implied the event layer was live; it is not."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.2",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.2",
+ "title": "1.13.2: ATP main-tour stoppages: the live match stage is now watched for its medical-timeout and interruption stages, and 1 more change",
+ "summary": "Added: ATP main-tour stoppages: the live match stage is now watched for its medical-timeout and interruption stages. Changed: ITF coverage note: the first medical timeout was observed and published on 2026-09-13.",
+ "content_html": "
Added
\n
\n
ATP main-tour stoppages: the live match stage is now watched for its medical-timeout and interruption stages. Those stages exist in the vocabulary but have not yet been observed on a tennis match, so ATP rows are documented as mapped, not yet proven; they carry player: null and an at equal to the instant the stage was observed.
\n
\n
Changed
\n
\n
ITF coverage note: the first medical timeout was observed and published on 2026-09-13.
\n
",
+ "date_published": "2026-09-13T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.2",
+ "date": "2026-09-13",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "ATP main-tour stoppages: the live match stage is now watched for its medical-timeout and interruption stages. Those stages exist in the vocabulary but have not yet been observed on a tennis match, so ATP rows are documented as mapped, not yet proven; they carry player: null and an at equal to the instant the stage was observed."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "ITF coverage note: the first medical timeout was observed and published on 2026-09-13."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.1",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.1",
+ "title": "1.13.1: Scorer-stated stoppages for WTA and WTA 125 singles: medical_timeout_*, trainer_called*, toilet_break_* and stoppage_* rows are now published for those tours…, and 2 more changes",
+ "summary": "Added: Scorer-stated stoppages for WTA and WTA 125 singles: medical_timeout_*, trainer_called*, toilet_break_* and stoppage_* rows are now published for those tours (measured over 138 finished matches: 27 physio calls, 65 treatment records, 22 suspensions). Scorer-stated stoppages for ITF World Tennis Tour singles from the court's live-scoring state: toilet_break_*, medical_timeout_*, trainer_called* and stoppage_* rows on entering and leaving the state, with duration_seconds; player is null on these rows because the live-scoring state names the state, not the player. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
Scorer-stated stoppages for WTA and WTA 125 singles: medical_timeout_*, trainer_called*, toilet_break_* and stoppage_* rows are now published for those tours (measured over 138 finished matches: 27 physio calls, 65 treatment records, 22 suspensions). Same fields, same endpoints, same WebSocket signal. stoppage_start.reason gains heat, darkness and injury; every stoppage_end carries reason: "resumed".
\n
\n
\n
Scorer-stated stoppages for ITF World Tennis Tour singles from the court's live-scoring state: toilet_break_*, medical_timeout_*, trainer_called* and stoppage_* rows on entering and leaving the state, with duration_seconds; player is null on these rows because the live-scoring state names the state, not the player.
\n
\n
Changed
\n
\n
Coverage note: WTA, WTA 125, Challenger, UTR and ITF singles are covered; ATP main tour is not covered yet and is stated as not covered.
\n
",
+ "date_published": "2026-09-13T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.1",
+ "date": "2026-09-13",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "Scorer-stated stoppages for WTA and WTA 125 singles: medical_timeout_*, trainer_called*, toilet_break_* and stoppage_* rows are now published for those tours (measured over 138 finished matches: 27 physio calls, 65 treatment records, 22 suspensions). Same fields, same endpoints, same WebSocket signal. stoppage_start.reason gains heat, darkness and injury; every stoppage_end carries reason: \"resumed\".",
+ "Scorer-stated stoppages for ITF World Tennis Tour singles from the court's live-scoring state: toilet_break_*, medical_timeout_*, trainer_called* and stoppage_* rows on entering and leaving the state, with duration_seconds; player is null on these rows because the live-scoring state names the state, not the player."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Coverage note: WTA, WTA 125, Challenger, UTR and ITF singles are covered; ATP main tour is not covered yet and is stated as not covered."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.13.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.13.0",
+ "title": "1.13.0: GET /events (PRO): the slate-wide events feed — the rows of GET /matches/{id}/events for every match in one call, oldest first, cursor by after_id…, and 2 more changes",
+ "summary": "Added: GET /events (PRO): the slate-wide events feed — the rows of GET /matches/{id}/events for every match in one call, oldest first, cursor by after_id (meta.next_cursor, null = caught up), since as the first-call UTC lower bound, type as a comma-separated list of event types or the family name stoppages. SlateEvent schema; Event.type documents the full vocabulary. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
GET /events (PRO): the slate-wide events feed — the rows of GET /matches/{id}/events for every match in one call, oldest first, cursor by after_id (meta.next_cursor, null = caught up), since as the first-call UTC lower bound, type as a comma-separated list of event types or the family name stoppages. Rows carry id and match_id. One request per tick covers the whole live slate: medical timeouts across every live match on PRO by polling.
\n
SlateEvent schema; Event.type documents the full vocabulary.
\n
\n
Changed
\n
\n
Stoppage coverage, corrected: scorer-stated medical_timeout_*, trainer_called* and toilet_break_* rows exist for Challenger and UTR singles. The 1.11.0/1.12.0 text said ATP, WTA and WTA 125 as well; measured over 11 days those tours' scorer data carries none (0 of 70 finished matches each, against 22 of 70 for Challenger). Main-tour and ITF medical timeouts are not covered yet; the note changes the day they are live.
\n
",
+ "date_published": "2026-09-13T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.13.0",
+ "date": "2026-09-13",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /events (PRO): the slate-wide events feed — the rows of GET /matches/{id}/events for every match in one call, oldest first, cursor by after_id (meta.next_cursor, null = caught up), since as the first-call UTC lower bound, type as a comma-separated list of event types or the family name stoppages. Rows carry id and match_id. One request per tick covers the whole live slate: medical timeouts across every live match on PRO by polling.",
+ "SlateEvent schema; Event.type documents the full vocabulary."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Stoppage coverage, corrected: scorer-stated medical_timeout_*, trainer_called* and toilet_break_* rows exist for Challenger and UTR singles. The 1.11.0/1.12.0 text said ATP, WTA and WTA 125 as well; measured over 11 days those tours' scorer data carries none (0 of 70 finished matches each, against 22 of 70 for Challenger). Main-tour and ITF medical timeouts are not covered yet; the note changes the day they are live."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.12.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.12.0",
+ "title": "1.12.0: serve and outcome on live point rows (GET /matches/{matchId}/points, the point frames): the serve the point was played on (1 | 2) and how it ended (ace |…, and 2 more changes",
+ "summary": "Added: serve and outcome on live point rows (GET /matches/{matchId}/points, the point frames): the serve the point was played on (1 | 2) and how it ended (ace | double_fault | winner | forced_error | unforced_error), as stated for the point, joined onto our rows by exact state; null when not stated. point_update frame on the point opt-in — a point's serve/outcome landing after its point frame; apply by seq. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
serve and outcome on live point rows (GET /matches/{matchId}/points, the point frames): the serve the point was played on (1 | 2) and how it ended (ace | double_fault | winner | forced_error | unforced_error), as stated for the point, joined onto our rows by exact state; null when not stated. Coverage per match in the new enrichment object; per tour in the reference.
\n
point_update frame on the point opt-in — a point's serve/outcome landing after its point frame; apply by seq. Asked by an ULTRA customer trading on serve outcome.
\n
published_at on every data frame of the native WebSocket and the push feed: the UTC instant (ms) the frame left our process — the third clock next to the state's timestamp and a point's ts, so processing and transport latency can be separated. Asked by a prospect running a latency benchmark.
\n
",
+ "date_published": "2026-09-12T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.12.0",
+ "date": "2026-09-12",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "serve and outcome on live point rows (GET /matches/{matchId}/points, the point frames): the serve the point was played on (1 | 2) and how it ended (ace | double_fault | winner | forced_error | unforced_error), as stated for the point, joined onto our rows by exact state; null when not stated. Coverage per match in the new enrichment object; per tour in the reference.",
+ "point_update frame on the point opt-in — a point's serve/outcome landing after its point frame; apply by seq. Asked by an ULTRA customer trading on serve outcome.",
+ "published_at on every data frame of the native WebSocket and the push feed: the UTC instant (ms) the frame left our process — the third clock next to the state's timestamp and a point's ts, so processing and transport latency can be separated. Asked by a prospect running a latency benchmark."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.11.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.11.0",
+ "title": "1.11.0: Medical timeouts, trainer calls and toilet breaks as events on GET /matches/{matchId}/events: medical_timeout_start/end, trainer_called/_end…, and 2 more changes",
+ "summary": "Added: Medical timeouts, trainer calls and toilet breaks as events on GET /matches/{matchId}/events: medical_timeout_start/end, trainer_called/_end, toilet_break_start/end, each with player (the player concerned), at (UTC), the score at that instant, reason and duration_seconds on the end row — as the match scorer states them (ATP, WTA, Challenger, WTA 125, UTR singles). stoppage_start/end now carry a stated reason (weather | other) when one exists. pause_start / pause_end (basis: inferred): interruptions of play measured from our own point clocks, on every live match, never labelled medical. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
Medical timeouts, trainer calls and toilet breaks as events on GET /matches/{matchId}/events: medical_timeout_start/end, trainer_called/_end, toilet_break_start/end, each with player (the player concerned), at (UTC), the score at that instant, reason and duration_seconds on the end row — as the match scorer states them (ATP, WTA, Challenger, WTA 125, UTR singles). stoppage_start/end now carry a stated reason (weather | other) when one exists. Asked by a prospect building on medical timeouts.
\n
pause_start / pause_end (basis: inferred): interruptions of play measured from our own point clocks, on every live match, never labelled medical.
\n
signals:["stoppages"] on the native WebSocket and the stoppage family on the signal:* push channels — every row above pushed the moment it is recorded.
\n
",
+ "date_published": "2026-09-12T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.11.0",
+ "date": "2026-09-12",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "Medical timeouts, trainer calls and toilet breaks as events on GET /matches/{matchId}/events: medical_timeout_start/end, trainer_called/_end, toilet_break_start/end, each with player (the player concerned), at (UTC), the score at that instant, reason and duration_seconds on the end row — as the match scorer states them (ATP, WTA, Challenger, WTA 125, UTR singles). stoppage_start/end now carry a stated reason (weather | other) when one exists. Asked by a prospect building on medical timeouts.",
+ "pause_start / pause_end (basis: inferred): interruptions of play measured from our own point clocks, on every live match, never labelled medical.",
+ "signals:[\"stoppages\"] on the native WebSocket and the stoppage family on the signal:* push channels — every row above pushed the moment it is recorded."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.10.1",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.10.1",
+ "title": "1.10.1: A ranking tie is two rows with the same rank (doubles partners always tie)",
+ "summary": "Clarified: A ranking tie is two rows with the same rank (doubles partners always tie).",
+ "content_html": "
Clarified
\n
\n
A ranking tie is two rows with the same rank (doubles partners always tie).
\n
",
+ "date_published": "2026-09-12T00:00:00Z",
+ "tags": [
+ "Clarified"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.10.1",
+ "date": "2026-09-12",
+ "sections": [
+ {
+ "name": "Clarified",
+ "items": [
+ "A ranking tie is two rows with the same rank (doubles partners always tie)."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.10.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.10.0",
+ "title": "1.10.0: win_probability_p1_model and win_probability_meta on ULTRA live score objects: the same model read computed without the market-prior anchor (a probability no…, and 3 more changes",
+ "summary": "Added: win_probability_p1_model and win_probability_meta on ULTRA live score objects: the same model read computed without the market-prior anchor (a probability no market price touched), plus model_version, generated_at and market_anchored for the pair. win_probability_p1 is unchanged. basis on status-ledger rows (observed | backfill); the ledger now reaches back to the per-match stamps held before it existed (completions from 2026-08-21, promotions to live from 2026-09-05), one reconstructed row per stamp, labelled. Plus 2 more changes.",
+ "content_html": "
Added
\n
\n
win_probability_p1_model and win_probability_meta on ULTRA live score objects: the same model read computed without the market-prior anchor (a probability no market price touched), plus model_version, generated_at and market_anchored for the pair. win_probability_p1 is unchanged. Asked by an ULTRA customer who needed to know whether the live probability is independent of market prices — on anchored matches it is not, and now the row says so.
\n
basis on status-ledger rows (observed | backfill); the ledger now reaches back to the per-match stamps held before it existed (completions from 2026-08-21, promotions to live from 2026-09-05), one reconstructed row per stamp, labelled.
\n
\n
\n
system=atp_doubles / system=wta_doubles on /rankings: the official weekly doubles tables (individual players), in both modes at the same gates as atp/wta, never included implicitly. Loaded from 2023 forward.
\n
\n
Clarified
\n
\n
Status-ledger rows are ordered by their instant (at), not insertion order.
\n
",
+ "date_published": "2026-09-12T00:00:00Z",
+ "tags": [
+ "Added",
+ "Clarified"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.10.0",
+ "date": "2026-09-12",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "win_probability_p1_model and win_probability_meta on ULTRA live score objects: the same model read computed without the market-prior anchor (a probability no market price touched), plus model_version, generated_at and market_anchored for the pair. win_probability_p1 is unchanged. Asked by an ULTRA customer who needed to know whether the live probability is independent of market prices — on anchored matches it is not, and now the row says so.",
+ "basis on status-ledger rows (observed | backfill); the ledger now reaches back to the per-match stamps held before it existed (completions from 2026-08-21, promotions to live from 2026-09-05), one reconstructed row per stamp, labelled.",
+ "system=atp_doubles / system=wta_doubles on /rankings: the official weekly doubles tables (individual players), in both modes at the same gates as atp/wta, never included implicitly. Loaded from 2023 forward."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Clarified",
+ "items": [
+ "Status-ledger rows are ordered by their instant (at), not insertion order."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.9.9",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.9.9",
+ "title": "1.9.9: The status ledger records from 2026-09-11T22:45:48Z (its first row), not the calendar date of the deploy",
+ "summary": "Clarified: The status ledger records from 2026-09-11T22:45:48Z (its first row), not the calendar date of the deploy.",
+ "content_html": "
Clarified
\n
\n
The status ledger records from 2026-09-11T22:45:48Z (its first row), not the calendar date of the deploy. Reported by an ULTRA customer reading match 188711.
\n
",
+ "date_published": "2026-09-12T00:00:00Z",
+ "tags": [
+ "Clarified"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.9.9",
+ "date": "2026-09-12",
+ "sections": [
+ {
+ "name": "Clarified",
+ "items": [
+ "The status ledger records from 2026-09-11T22:45:48Z (its first row), not the calendar date of the deploy. Reported by an ULTRA customer reading match 188711."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.9.8",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.9.8",
+ "title": "1.9.8: GET /matches/{matchId}/status-history — the per-match status ledger: every status / event_status transition with the UTC instant we published it, the value…, and 1 more change",
+ "summary": "Added: GET /matches/{matchId}/status-history — the per-match status ledger: every status / event_status transition with the UTC instant we published it, the value before and after, the derived outcome and the score at that instant. stoppage_start / stoppage_end events on GET /matches/{matchId}/events: an in-play suspension with the score at the moment, reason (unknown until a source states one — never inferred) and duration_seconds on the end row.",
+ "content_html": "
Added
\n
\n
GET /matches/{matchId}/status-history — the per-match status ledger: every status / event_status transition with the UTC instant we published it, the value before and after, the derived outcome and the score at that instant. Append-only, oldest first; rows exist from 2026-09-12. History capability (BASIC and the Historical Data plans). Asked for by a researcher reconciling corrections against the moment they were published.
\n
stoppage_start / stoppage_end events on GET /matches/{matchId}/events: an in-play suspension with the score at the moment, reason (unknown until a source states one — never inferred) and duration_seconds on the end row. Asked for by a product builder wanting medical timeouts as an event.
\n
",
+ "date_published": "2026-09-12T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.9.8",
+ "date": "2026-09-12",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /matches/{matchId}/status-history — the per-match status ledger: every status / event_status transition with the UTC instant we published it, the value before and after, the derived outcome and the score at that instant. Append-only, oldest first; rows exist from 2026-09-12. History capability (BASIC and the Historical Data plans). Asked for by a researcher reconciling corrections against the moment they were published.",
+ "stoppage_start / stoppage_end events on GET /matches/{matchId}/events: an in-play suspension with the score at the moment, reason (unknown until a source states one — never inferred) and duration_seconds on the end row. Asked for by a product builder wanting medical timeouts as an event."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.9.7",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.9.7",
+ "title": "1.9.7: Match.outcome is now documented on the Match schema (it has been on every match since 2026-08-18): completed | retired | walkover | default | abandoned |…, and 1 more change",
+ "summary": "Added: Match.outcome is now documented on the Match schema (it has been on every match since 2026-08-18): completed | retired | walkover | default | abandoned | unresolved | null, derived from status + event_status. outcome: unresolved — a match every source lost before a result is closed unfinished and says so: score is the last state observed, winner is null, no result is asserted; it flips to completed with the proven final when an authority confirms the result.",
+ "content_html": "
Added
\n
\n
Match.outcome is now documented on the Match schema (it has been on every match since 2026-08-18): completed | retired | walkover | default | abandoned | unresolved | null, derived from status + event_status.
\n
outcome: unresolved — a match every source lost before a result is closed unfinished and says so: score is the last state observed, winner is null, no result is asserted; it flips to completed with the proven final when an authority confirms the result. Until 2026-09-10 such closes were published as completed with a mid-match score, which an ULTRA customer read as a wrong result.
\n
",
+ "date_published": "2026-09-10T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.9.7",
+ "date": "2026-09-10",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "Match.outcome is now documented on the Match schema (it has been on every match since 2026-08-18): completed | retired | walkover | default | abandoned | unresolved | null, derived from status + event_status.",
+ "outcome: unresolved — a match every source lost before a result is closed unfinished and says so: score is the last state observed, winner is null, no result is asserted; it flips to completed with the proven final when an authority confirms the result. Until 2026-09-10 such closes were published as completed with a mid-match score, which an ULTRA customer read as a wrong result."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.9.6",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.9.6",
+ "title": "1.9.6: stats.ratings_as_of on GET /players/{id} — the date (UTC) the ratings block was last refreshed, and 2 more changes",
+ "summary": "Added: stats.ratings_as_of on GET /players/{id} — the date (UTC) the ratings block was last refreshed. GET /history/packages/{period}?format=corrections — a CSV keyed by match_id (match_id, field, before, after, tournament_key, source, corrected_at) on every package whose stored data was repaired after publication, listed in the manifest as format: corrections; 404 when a package has none. Plus 1 more change.",
+ "content_html": "
Added
\n
\n
stats.ratings_as_of on GET /players/{id} — the date (UTC) the ratings block was last refreshed. The current Elo is now refreshed every week for every rated player from the published rating tables (Tuesdays); before this the rating carried no date and a May capture read as this week's. Reported by a PRO customer building pre-match research.
\n
GET /history/packages/{period}?format=corrections — a CSV keyed by match_id (match_id, field, before, after, tournament_key, source, corrected_at) on every package whose stored data was repaired after publication, listed in the manifest as format: corrections; 404 when a package has none. First use: surface corrections on the 2023-02 → 2024-12 tape packages (19,439 Challenger matches whose court surface a tournament-name rule had set to grass; repaired and rebuilt 2026-09-10). Reported by a History Pro customer.
\n
\n
Fixed
\n
\n
Merged player ids: a merge now repoints the weekly ranking tables too, so /rankings rows carry the same player_id the match records carry, and the retired id answers 410 with merged_into (see Merged and retired player ids).
\n
",
+ "date_published": "2026-09-10T00:00:00Z",
+ "tags": [
+ "Added",
+ "Fixed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.9.6",
+ "date": "2026-09-10",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "stats.ratings_as_of on GET /players/{id} — the date (UTC) the ratings block was last refreshed. The current Elo is now refreshed every week for every rated player from the published rating tables (Tuesdays); before this the rating carried no date and a May capture read as this week's. Reported by a PRO customer building pre-match research.",
+ "GET /history/packages/{period}?format=corrections — a CSV keyed by match_id (match_id, field, before, after, tournament_key, source, corrected_at) on every package whose stored data was repaired after publication, listed in the manifest as format: corrections; 404 when a package has none. First use: surface corrections on the 2023-02 → 2024-12 tape packages (19,439 Challenger matches whose court surface a tournament-name rule had set to grass; repaired and rebuilt 2026-09-10). Reported by a History Pro customer."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Fixed",
+ "items": [
+ "Merged player ids: a merge now repoints the weekly ranking tables too, so /rankings rows carry the same player_id the match records carry, and the retired id answers 410 with merged_into (see Merged and retired player ids)."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.9.5",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.9.5",
+ "title": "1.9.5: GET /matches/{matchId}/prices?cursor= — keyset paging past 500 ticks: while meta.has_more is true, pass meta.next_cursor back as ?cursor= for the next (older)…, and 4 more changes",
+ "summary": "Added: GET /matches/{matchId}/prices?cursor= — keyset paging past 500 ticks: while meta.has_more is true, pass meta.next_cursor back as ?cursor= for the next (older) page; pages never overlap or skip a tick; anything else is 400 bad_cursor. Match.live_at — the instant our feed last reported the match in play (UTC), the closest thing to an actual start time. Plus 3 more changes.",
+ "content_html": "
Added
\n
\n
GET /matches/{matchId}/prices?cursor= — keyset paging past 500 ticks: while meta.has_more is true, pass meta.next_cursor back as ?cursor= for the next (older) page; pages never overlap or skip a tick; anything else is 400 bad_cursor. Mirrors tennis 3123b27f (live 2026-09-09 06:01Z).
\n
Match.live_at — the instant our feed last reported the match in play (UTC), the closest thing to an actual start time. Null for matches that went live before 2026-09-05 or were never observed live. Same commit.
\n
410 Gone with a MatchMerged body on the ten per-match routes when a match id was merged into another record: error: merged, merged_into (the end of the chain, or null), merged_at, detail. Live since 2026-09-05 (tennis d131a12f).
\n
Tape point is_unreturned / unreturned_kind (rally tapes, tennis 6e7b82f6, 2026-09-05).
\n
\n
Clarified
\n
\n
Price ticks are kept for 30 days and then deleted: an older match answers an empty data / prices array while its market stays mapped — retention, not a fault. Stated on both prices endpoints.
\n
",
+ "date_published": "2026-09-09T00:00:00Z",
+ "tags": [
+ "Added",
+ "Clarified"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.9.5",
+ "date": "2026-09-09",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /matches/{matchId}/prices?cursor= — keyset paging past 500 ticks: while meta.has_more is true, pass meta.next_cursor back as ?cursor= for the next (older) page; pages never overlap or skip a tick; anything else is 400 bad_cursor. Mirrors tennis 3123b27f (live 2026-09-09 06:01Z).",
+ "Match.live_at — the instant our feed last reported the match in play (UTC), the closest thing to an actual start time. Null for matches that went live before 2026-09-05 or were never observed live. Same commit.",
+ "410 Gone with a MatchMerged body on the ten per-match routes when a match id was merged into another record: error: merged, merged_into (the end of the chain, or null), merged_at, detail. Live since 2026-09-05 (tennis d131a12f).",
+ "Tape point is_unreturned / unreturned_kind (rally tapes, tennis 6e7b82f6, 2026-09-05)."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Clarified",
+ "items": [
+ "Price ticks are kept for 30 days and then deleted: an older match answers an empty data / prices array while its market stays mapped — retention, not a fault. Stated on both prices endpoints."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.9.4",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.9.4",
+ "title": "1.9.4: Price.side (per-match and per-market prices endpoints, and the match embed) now follows players.p1 / players.p2 through our name-verified match-market mapping, and 1 more change",
+ "summary": "Fixed: Price.side (per-match and per-market prices endpoints, and the match embed) now follows players.p1 / players.p2 through our name-verified match-market mapping. Clarified: Price.mid is the venue's observed midpoint, never a model estimate; synthetic describes only bid/ask (mid +/- 0.005 when true); pre-match ticks are synthetic: true by design because the live book attaches at match start; timestamp is our observation clock (UTC), not a venue publication time.",
+ "content_html": "
Fixed
\n
\n
Price.side (per-match and per-market prices endpoints, and the match embed) now follows players.p1 / players.p2 through our name-verified match-market mapping. Until 2026-09-09 it followed the venue's display order, which lists roughly half of all pairings the other way round, so on those markets side: 1 was in fact players.p2. The mapping is stored, so re-reading any tick - historical included - returns the correct side. Reported by a PRO customer cross-checking pre-match prices against rankings; mirrors tennis cbd3f647 (live 2026-09-09 04:59Z).
\n
\n
Clarified
\n
\n
Price.mid is the venue's observed midpoint, never a model estimate; synthetic describes only bid/ask (mid +/- 0.005 when true); pre-match ticks are synthetic: true by design because the live book attaches at match start; timestamp is our observation clock (UTC), not a venue publication time.
\n
",
+ "date_published": "2026-09-09T00:00:00Z",
+ "tags": [
+ "Fixed",
+ "Clarified"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.9.4",
+ "date": "2026-09-09",
+ "sections": [
+ {
+ "name": "Fixed",
+ "items": [
+ "Price.side (per-match and per-market prices endpoints, and the match embed) now follows players.p1 / players.p2 through our name-verified match-market mapping. Until 2026-09-09 it followed the venue's display order, which lists roughly half of all pairings the other way round, so on those markets side: 1 was in fact players.p2. The mapping is stored, so re-reading any tick - historical included - returns the correct side. Reported by a PRO customer cross-checking pre-match prices against rankings; mirrors tennis cbd3f647 (live 2026-09-09 04:59Z)."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Clarified",
+ "items": [
+ "Price.mid is the venue's observed midpoint, never a model estimate; synthetic describes only bid/ask (mid +/- 0.005 when true); pre-match ticks are synthetic: true by design because the live book attaches at match start; timestamp is our observation clock (UTC), not a venue publication time."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.9.3",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.9.3",
+ "title": "1.9.3: Match.tournament_id null cases restated with the real causes and a measured rate: not in our tournament list (UTR), or a match whose name matched no single…",
+ "summary": "Clarified: Match.tournament_id null cases restated with the real causes and a measured rate: not in our tournament list (UTR), or a match whose name matched no single edition in that list.",
+ "content_html": "
Clarified
\n
\n
Match.tournament_id null cases restated with the real causes and a measured rate: not in our tournament list (UTR), or a match whose name matched no single edition in that list. The API now resolves the latter at ingest by name + event type + date against the published fixtures, and the 2026 backlog was backfilled the same day (2,547 rows); ITF null rate went from about 25% to about 2.5%. No wire change.
\n
",
+ "date_published": "2026-09-04T00:00:00Z",
+ "tags": [
+ "Clarified"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.9.3",
+ "date": "2026-09-04",
+ "sections": [
+ {
+ "name": "Clarified",
+ "items": [
+ "Match.tournament_id null cases restated with the real causes and a measured rate: not in our tournament list (UTR), or a match whose name matched no single edition in that list. The API now resolves the latter at ingest by name + event type + date against the published fixtures, and the 2026 backlog was backfilled the same day (2,547 rows); ITF null rate went from about 25% to about 2.5%. No wire change."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.9.2",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.9.2",
+ "title": "1.9.2: GET /matches?status=cancelled is now in the status enum, with what it covers (feed-cancelled, walkover with no stated winner, postponed and never played), its…, and 3 more changes",
+ "summary": "Added: GET /matches?status=cancelled is now in the status enum, with what it covers (feed-cancelled, walkover with no stated winner, postponed and never played), its tier (BASIC or any History plan, like completed), and the one restriction: it pages with limit/offset and refuses updated_since. tournament_id= filter on /matches (every status) and /history/matches — the stable id each match row carries and /tournaments publishes. Plus 2 more changes.",
+ "content_html": "
Added (spec catch-up — every one of these has been live on the API; the document lagged)
\n
\n
GET /matches?status=cancelled is now in the status enum, with what it covers (feed-cancelled, walkover with no stated winner, postponed and never played), its tier (BASIC or any History plan, like completed), and the one restriction: it pages with limit/offset and refuses updated_since.
\n
tournament_id= filter on /matches (every status) and /history/matches — the stable id each match row carries and /tournaments publishes. The description states there is no separate edition/occurrence id: one edition is tournament_id plus a from/to window, and matches with a null tournament_id (tournament not yet catalogued) fall outside the filter.
\n
Change feed on /matches: updated_since= / cursor= parameters and the meta.next_cursor / meta.watermark fields, with the at-least-once and no-from/to rules.
\n
\n
Clarified
\n
\n
meta.has_more says how to enumerate a filtered set completely (page offset by limit until false) and that total is null on the terminal listings, so has_more is the only end-of-data signal there. Asked by a Basic customer on Discord; no wire change.
\n
",
+ "date_published": "2026-09-04T00:00:00Z",
+ "tags": [
+ "Added",
+ "Clarified"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.9.2",
+ "date": "2026-09-04",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /matches?status=cancelled is now in the status enum, with what it covers (feed-cancelled, walkover with no stated winner, postponed and never played), its tier (BASIC or any History plan, like completed), and the one restriction: it pages with limit/offset and refuses updated_since.",
+ "tournament_id= filter on /matches (every status) and /history/matches — the stable id each match row carries and /tournaments publishes. The description states there is no separate edition/occurrence id: one edition is tournament_id plus a from/to window, and matches with a null tournament_id (tournament not yet catalogued) fall outside the filter.",
+ "Change feed on /matches: updated_since= / cursor= parameters and the meta.next_cursor / meta.watermark fields, with the at-least-once and no-from/to rules."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Clarified",
+ "items": [
+ "meta.has_more says how to enumerate a filtered set completely (page offset by limit until false) and that total is null on the terminal listings, so has_more is the only end-of-data signal there. Asked by a Basic customer on Discord; no wire change."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.9.1",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.9.1",
+ "title": "1.9.1: Player.ranking / Player.ranking_points now say what they are: the official singles ranking POSITION (the ordinal), ATP table for men and WTA table for women…",
+ "summary": "Clarified: Player.ranking / Player.ranking_points now say what they are: the official singles ranking POSITION (the ordinal), ATP table for men and WTA table for women chosen by the player, refreshed ahead of each match the player has with us, null when no ranking is held, and always the CURRENT record even on historical matches (/rankings?as_of= is the as-of surface).",
+ "content_html": "
Clarified
\n
\n
Player.ranking / Player.ranking_points now say what they are: the official singles ranking POSITION (the ordinal), ATP table for men and WTA table for women chosen by the player, refreshed ahead of each match the player has with us, null when no ranking is held, and always the CURRENT record even on historical matches (/rankings?as_of= is the as-of surface). Asked by a Basic customer; no wire change.
\n
",
+ "date_published": "2026-09-02T00:00:00Z",
+ "tags": [
+ "Clarified"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.9.1",
+ "date": "2026-09-02",
+ "sections": [
+ {
+ "name": "Clarified",
+ "items": [
+ "Player.ranking / Player.ranking_points now say what they are: the official singles ranking POSITION (the ordinal), ATP table for men and WTA table for women chosen by the player, refreshed ahead of each match the player has with us, null when no ranking is held, and always the CURRENT record even on historical matches (/rankings?as_of= is the as-of surface). Asked by a Basic customer; no wire change."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.9.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.9.0",
+ "title": "1.9.0: has_analysis and has_market on Match — two booleans on every row of GET /matches and on the detail, every tier, and 1 more change",
+ "summary": "Added: has_analysis and has_market on Match — two booleans on every row of GET /matches and on the detail, every tier. Distinguishable absence on GET /matches/{matchId}/analysis and GET /markets/{matchId}/prices.",
+ "content_html": "
Added
\n
\n
has_analysis and has_market on Match — two booleans on every row of GET /matches and on the detail, every tier. They carry the same fact the per-match endpoints answer 404 about, so a slate is filtered in one call instead of one 404 per match. Shipped to production 2026-09-02.
\n
Distinguishable absence on GET /matches/{matchId}/analysis and GET /markets/{matchId}/prices. The status stays 404 (unchanged, and shipped clients branch on it), but the body now says which absence it is: {"error":"not_found"} for an id that does not exist, and {"error":"no_analysis"|"no_market","match_id":…,"coverage":"none","detail":…} for a real match we hold nothing for — the same coverage: "none" vocabulary /matches/{matchId}/statistics already uses in its 200. New error codes no_analysis, no_market documented on the Error schema.
\n
",
+ "date_published": "2026-09-02T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.9.0",
+ "date": "2026-09-02",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "has_analysis and has_market on Match — two booleans on every row of GET /matches and on the detail, every tier. They carry the same fact the per-match endpoints answer 404 about, so a slate is filtered in one call instead of one 404 per match. Shipped to production 2026-09-02.",
+ "Distinguishable absence on GET /matches/{matchId}/analysis and GET /markets/{matchId}/prices. The status stays 404 (unchanged, and shipped clients branch on it), but the body now says which absence it is: {\"error\":\"not_found\"} for an id that does not exist, and {\"error\":\"no_analysis\"|\"no_market\",\"match_id\":…,\"coverage\":\"none\",\"detail\":…} for a real match we hold nothing for — the same coverage: \"none\" vocabulary /matches/{matchId}/statistics already uses in its 200. New error codes no_analysis, no_market documented on the Error schema."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.8.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.8.0",
+ "title": "1.8.0: GET /history/archive/matches/{archiveId}/tape (getArchiveTape) — the RECONSTRUCTED 2013–2022 point-by-point tape for one archive result: the score sequence…, and 8 more changes",
+ "summary": "Added: GET /history/archive/matches/{archiveId}/tape (getArchiveTape) — the RECONSTRUCTED 2013–2022 point-by-point tape for one archive result: the score sequence behind the published result: reconstructed point sequences, no capture timestamps. ArchiveTape schema. Plus 7 more changes.",
+ "content_html": "
Added
\n
\n
GET /history/archive/matches/{archiveId}/tape (getArchiveTape) — the RECONSTRUCTED 2013–2022 point-by-point tape for one archive result: the score sequence behind the published result: reconstructed point sequences, no capture timestamps. Shipped to production 2026-09-01; the spec was the last place it was missing. Tier: core ULTRA, or any active History plan including Starter (which opens it on a FREE core key). The archive RESULT stays on BASIC, so core BASIC and core PRO read the result and are refused the tape — 403 upgrade_required carrying capability: archive_tape.
\n
ArchiveTape schema. Same envelope as HistoryTape so one parser reads both halves of the tape product, with the differences that are true: match is the winner/loser-shaped archive row (rows are WINNER-FIRST, not p1/p2), profiles is always [], and meta carries archive_match_id rather than match_id — an archive id is not a match id, and passing one to the other's routes resolves a different, real record without erroring. Rows are HistoryTapeRow, reused unchanged.
\n
kind=archive_tape on /history/packages and /history/packages/{period}, and on HistoryPackage.kind. Ten per-year bulk files, period 2013 through 2022, JSONL + CSV, all ready. The JSONL record is byte-for-byte what the per-match endpoint returns; the CSV is the flat per-row view keyed on archive_match_id and deliberately carries no timestamp, win_probability_p1 or danger column. A SEPARATE gate from the per-match tape: ULTRA, a History Pro/Business subscription, or an active one-off package window — a Starter grant reads tapes one at a time and does not download years of them. Core PRO carries neither gate.
\n
\n
Changed
\n
\n
info.description states the provenance and the coverage, thin spots included. Nobody watched these matches: timestamp, win_probability_p1 and danger are null on EVERY row and cannot be filled in later — the production table has no timestamp column at all, so this is a structural fact and not a convention. That is the opposite of the 2023→now tape, which is our own recording, where the rows we actually watched carry a real clock and most of them a model probability. The corpus is 97,901 matches / 14,340,663 rows, seasons 2013–2022 only: 977,903 archive results from 1968–2012 have no tape and never will. Coverage of the era is 19.3% overall and 44.9% of tour-level play — main-draw tour buckets 91.6–98.7%, ATP Challenger main 55.3% and Challenger qualifying 33.6%, slam QUALIFYING 16.0% (ATP) / 18.1% (WTA), ITF and futures effectively zero (25 of 116,575 ATP futures). Stated together, because a strong number published without its thin counterpart sells a corpus nobody has.
\n
ArchiveTape.meta.coverage names the measured split rather than glossing the label.reconstructed_partial (3,594 matches) has TWO causes and does not say which: 3,038 are point-granular tapes of matches that genuinely stopped early (3,027 retirements, 11 defaults) — the larger cause — and the other 556 carry the label only because their tape is per-GAME. Read granularity and the match's own score, not the label, when the question is whether the whole match is there. granularity is point on 99.4% of the corpus; the 556 game tapes are 555 in 2013 and one in 2014.
\n
Coverage says where its vocabulary stops. It describes the 2023+ tape; the archive tape reuses two of its five values and derives reconstructed_partial differently.
\n
ArchiveMatch points at the tape, and GET /history/archive/matches/{archiveId} says the RESULT stays on BASIC either way.
\n
The plain-HTML reference gains an FAQ entry — "Is there point-by-point data before 2023?" — and the history FAQ, llms.txt digest and README plan summary carry the same numbers, so an answer engine reading any one of them gets the corpus, the null clock and the coverage floor together.
\n
info.version is now 1.8.0.
\n
",
+ "date_published": "2026-09-01T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.8.0",
+ "date": "2026-09-01",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "GET /history/archive/matches/{archiveId}/tape (getArchiveTape) — the RECONSTRUCTED 2013–2022 point-by-point tape for one archive result: the score sequence behind the published result: reconstructed point sequences, no capture timestamps. Shipped to production 2026-09-01; the spec was the last place it was missing. Tier: core ULTRA, or any active History plan including Starter (which opens it on a FREE core key). The archive RESULT stays on BASIC, so core BASIC and core PRO read the result and are refused the tape — 403 upgrade_required carrying capability: archive_tape.",
+ "ArchiveTape schema. Same envelope as HistoryTape so one parser reads both halves of the tape product, with the differences that are true: match is the winner/loser-shaped archive row (rows are WINNER-FIRST, not p1/p2), profiles is always [], and meta carries archive_match_id rather than match_id — an archive id is not a match id, and passing one to the other's routes resolves a different, real record without erroring. Rows are HistoryTapeRow, reused unchanged.",
+ "kind=archive_tape on /history/packages and /history/packages/{period}, and on HistoryPackage.kind. Ten per-year bulk files, period 2013 through 2022, JSONL + CSV, all ready. The JSONL record is byte-for-byte what the per-match endpoint returns; the CSV is the flat per-row view keyed on archive_match_id and deliberately carries no timestamp, win_probability_p1 or danger column. A SEPARATE gate from the per-match tape: ULTRA, a History Pro/Business subscription, or an active one-off package window — a Starter grant reads tapes one at a time and does not download years of them. Core PRO carries neither gate."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "info.description states the provenance and the coverage, thin spots included. Nobody watched these matches: timestamp, win_probability_p1 and danger are null on EVERY row and cannot be filled in later — the production table has no timestamp column at all, so this is a structural fact and not a convention. That is the opposite of the 2023→now tape, which is our own recording, where the rows we actually watched carry a real clock and most of them a model probability. The corpus is 97,901 matches / 14,340,663 rows, seasons 2013–2022 only: 977,903 archive results from 1968–2012 have no tape and never will. Coverage of the era is 19.3% overall and 44.9% of tour-level play — main-draw tour buckets 91.6–98.7%, ATP Challenger main 55.3% and Challenger qualifying 33.6%, slam QUALIFYING 16.0% (ATP) / 18.1% (WTA), ITF and futures effectively zero (25 of 116,575 ATP futures). Stated together, because a strong number published without its thin counterpart sells a corpus nobody has.",
+ "ArchiveTape.meta.coverage names the measured split rather than glossing the label. reconstructed_partial (3,594 matches) has TWO causes and does not say which: 3,038 are point-granular tapes of matches that genuinely stopped early (3,027 retirements, 11 defaults) — the larger cause — and the other 556 carry the label only because their tape is per-GAME. Read granularity and the match's own score, not the label, when the question is whether the whole match is there. granularity is point on 99.4% of the corpus; the 556 game tapes are 555 in 2013 and one in 2014.",
+ "Coverage says where its vocabulary stops. It describes the 2023+ tape; the archive tape reuses two of its five values and derives reconstructed_partial differently.",
+ "ArchiveMatch points at the tape, and GET /history/archive/matches/{archiveId} says the RESULT stays on BASIC either way.",
+ "The plain-HTML reference gains an FAQ entry — \"Is there point-by-point data before 2023?\" — and the history FAQ, llms.txt digest and README plan summary carry the same numbers, so an answer engine reading any one of them gets the corpus, the null clock and the coverage floor together.",
+ "info.version is now 1.8.0."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.7.2",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.7.2",
+ "title": "1.7.2: status on Match now documents the lifecycle rule, and 1 more change",
+ "summary": "Changed: status on Match now documents the lifecycle rule. info.version is now 1.7.2.",
+ "content_html": "
Changed
\n
\n
status on Match now documents the lifecycle rule.completed is asserted only for a match we observed being played or whose match-winner market settled decisively; a closed market alone never finishes a match. cancelled with event_status: null means positive evidence the match was not played as scheduled (the market settled void) and no stated reason for why — outcome / fixture reason stay null rather than guessed, and the row upgrades to a completed walkover if a Walk Over with a stated winner lands later. Before 2026-08-23 a void market closure completed the match and stamped Finished; those rows were corrected. No field added or changed type — prose only.
\n
info.version is now 1.7.2.
\n
",
+ "date_published": "2026-08-23T00:00:00Z",
+ "tags": [
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.7.2",
+ "date": "2026-08-23",
+ "sections": [
+ {
+ "name": "Changed",
+ "items": [
+ "status on Match now documents the lifecycle rule. completed is asserted only for a match we observed being played or whose match-winner market settled decisively; a closed market alone never finishes a match. cancelled with event_status: null means positive evidence the match was not played as scheduled (the market settled void) and no stated reason for why — outcome / fixture reason stay null rather than guessed, and the row upgrades to a completed walkover if a Walk Over with a stated winner lands later. Before 2026-08-23 a void market closure completed the match and stamped Finished; those rows were corrected. No field added or changed type — prose only.",
+ "info.version is now 1.7.2."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.7.1",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.7.1",
+ "title": "1.7.1: event_status_updated_at on Match, and 1 more change",
+ "summary": "Added: event_status_updated_at on Match. Changed: info.version is now 1.7.1.",
+ "content_html": "
Added
\n
\n
event_status_updated_at on Match. Nullable ISO-8601 UTC (Z) timestamp, right after event_status: when event_status last CHANGED — the instant WE recorded the walkover / retirement / cancellation / postponement / suspension (or its clearing), not when the tournament desk did. It is the field to measure our admin-status latency with. Bumps only on a change of value (a re-read of the same status never moves it; a clear back to null does). Null while event_status has never changed since the field was introduced (2026-08-19) — never backfilled, never guessed. Inherited by every schema built on Match (MatchDetail, HistoryMatch). Additive only — no existing field moved or changed type.
\n
\n
Changed
\n
\n
info.version is now 1.7.1.
\n
",
+ "date_published": "2026-08-19T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.7.1",
+ "date": "2026-08-19",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "event_status_updated_at on Match. Nullable ISO-8601 UTC (Z) timestamp, right after event_status: when event_status last CHANGED — the instant WE recorded the walkover / retirement / cancellation / postponement / suspension (or its clearing), not when the tournament desk did. It is the field to measure our admin-status latency with. Bumps only on a change of value (a re-read of the same status never moves it; a clear back to null does). Null while event_status has never changed since the field was introduced (2026-08-19) — never backfilled, never guessed. Inherited by every schema built on Match (MatchDetail, HistoryMatch). Additive only — no existing field moved or changed type."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "info.version is now 1.7.1."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.7.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.7.0",
+ "title": "1.7.0: basis on GET /matches/{matchId}/points responses, and 1 more change",
+ "summary": "Added: basis on GET /matches/{matchId}/points responses. Changed: info.version is now 1.7.0.",
+ "content_html": "
Added
\n
\n
basis on GET /matches/{matchId}/points responses.live | reconstruction — which base served the page. Live capture is inherently partial: the stream serves what arrived while the match ran, and the match-closing point never streams live. For a COMPLETED match where a measured-complete recorded point sequence exists, the endpoint now serves that complete sequence instead, projected into the same point-frame shape — love-love opener through the match-closing point, seq contiguous 1..N, qualityclean — and says so with basis: "reconstruction". Every other case (every live match, and any completed match without a measured-complete recorded sequence) stays basis: "live" — the persisted stream rows, byte-for-byte what was served before. Completeness beats the partial live capture WHOLESALE — the two sequences are never interleaved (they share no key, so any merge would fabricate an order), the same rule ?points=complete follows on the history read. On projected frames ts is null on every row: the recorded sequence carries no per-point clock and none is fabricated (LivePoint.ts now documents this). after_seq pagination and seq dedup work identically on either basis, but the two bases are different sequences: after a match completes and flips to reconstruction, re-read from after_seq=0 rather than resuming a live cursor into it. Additive only — no existing field moved or changed type.
\n
\n
Changed
\n
\n
info.version is now 1.7.0.
\n
",
+ "date_published": "2026-08-18T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.7.0",
+ "date": "2026-08-18",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "basis on GET /matches/{matchId}/points responses. live | reconstruction — which base served the page. Live capture is inherently partial: the stream serves what arrived while the match ran, and the match-closing point never streams live. For a COMPLETED match where a measured-complete recorded point sequence exists, the endpoint now serves that complete sequence instead, projected into the same point-frame shape — love-love opener through the match-closing point, seq contiguous 1..N, quality clean — and says so with basis: \"reconstruction\". Every other case (every live match, and any completed match without a measured-complete recorded sequence) stays basis: \"live\" — the persisted stream rows, byte-for-byte what was served before. Completeness beats the partial live capture WHOLESALE — the two sequences are never interleaved (they share no key, so any merge would fabricate an order), the same rule ?points=complete follows on the history read. On projected frames ts is null on every row: the recorded sequence carries no per-point clock and none is fabricated (LivePoint.ts now documents this). after_seq pagination and seq dedup work identically on either basis, but the two bases are different sequences: after a match completes and flips to reconstruction, re-read from after_seq=0 rather than resuming a live cursor into it. Additive only — no existing field moved or changed type."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "info.version is now 1.7.0."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.6.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.6.0",
+ "title": "1.6.0: The three-valued draw field on Match, and 5 more changes",
+ "summary": "Added: The three-valued draw field on Match. ?draw=singles|doubles on four listings — GET /matches, /history/matches, /tournaments and /fixtures: the axis the tour filter deliberately collapses; the two compose (?tour=itf&draw=doubles is the ITF doubles slice). Plus 4 more changes.",
+ "content_html": "
Added
\n
\n
The three-valued draw field on Match.singles | doubles | null — same vocabulary as the new ?draw= filter, decided by one shared definition, so filter and field cannot disagree. Evidence order: a doubles-team participant proves doubles over any event type; otherwise the stated event type decides. Null means neither says anything — no stated event type, or a team tie (Davis Cup / BJK Cup / United Cup class), where one event type covers both singles and doubles rubbers and we will not guess which this is. Null is NOT singles. is_doubles stays for compatibility and is now documented as LOSSY, with its evidence order: false also covers "unknown", which is not a claim of singles — prefer draw, whose null says so honestly.
\n
?draw=singles|doubles on four listings — GET /matches, /history/matches, /tournaments and /fixtures: the axis the tour filter deliberately collapses; the two compose (?tour=itf&draw=doubles is the ITF doubles slice). A row whose draw is null matches NEITHER value — null is an answer, not a wildcard. An unknown value is a 400 bad_draw with the allowed values. Two honesty notes carried into the spec: on /tournaments the answer comes from the event type alone (a tournament row has no participants to supply the doubles-team evidence matches have), and ?draw=doubles alone also returns mixed and exhibition doubles that no ?tour= value reaches.
\n
GET /history/coverage (BASIC, or any Historical Data API plan) — the measured completeness rollup per tour × draw bucket: the numbers to read BEFORE choosing what to backtest, in one call instead of paging the archive. A prebuilt snapshot rebuilt nightly right after the completeness ledger reconverges — never computed at read time — so as_of (= built_at) dates every number and ledger_max_computed_at is the newest underlying per-match measurement; 503 coverage_unavailable before the first nightly build. Buckets are atp/wta/challenger/itf/juniors × singles/doubles plus other (team ties, mixed, exhibitions and matches with no stated event type — counted, never dropped, so the totals cannot lie); a bucket with zero completed matches is OMITTED, never emitted as zeros. Each bucket carries the five verifiable numbers (completed, any_tape, point_complete, complete_on_default_read, share), and method states the full measurement rule so every number carries its own definition. As of 2026-08-18: 174,393 completed matches; 91,318 point-complete on the best basis (52.4%) against 81,196 on the default read alone (46.6%); ITF singles 51.1% against ITF doubles 3.5% — do not extrapolate a completeness rate across a tour group.
\n
tape.starts_at_love and tape.computed_at on /history/matches items (both nullable, present where enabled). starts_at_love follows the SAME best-basis rule as points_complete — true when either measured basis opens at the 0-0 state, so if any on-disk sequence opens at love you can obtain one that does. computed_at is when the ledger last measured the match: every field in the tape's measured block is a nightly-reconverged cache, and this is the as-of to quote with any of them. Null on either means the match has not been measured — never a guess.
\n
The complete-basis addendum package files. An affected month's manifest may also list tennis_history_points_complete_<period>.jsonl.gz / .csv.gz (marked compression: gzip; bytes/sha256 cover the compressed bytes — exactly the download): for exactly the matches whose complete point sequence exists only as the on-disk reconstruction, the same tape ?points=complete serves (reconstruction contract: null timestamps and null model fields). The base files keep carrying every match's DEFAULT read — already the complete tape for most point-complete matches — and are never rewritten by the addendum; their bytes and sha256 values do not move.
\n
\n
Changed
\n
\n
info.version is now 1.6.0.
\n
",
+ "date_published": "2026-08-18T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.6.0",
+ "date": "2026-08-18",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "The three-valued draw field on Match. singles | doubles | null — same vocabulary as the new ?draw= filter, decided by one shared definition, so filter and field cannot disagree. Evidence order: a doubles-team participant proves doubles over any event type; otherwise the stated event type decides. Null means neither says anything — no stated event type, or a team tie (Davis Cup / BJK Cup / United Cup class), where one event type covers both singles and doubles rubbers and we will not guess which this is. Null is NOT singles. is_doubles stays for compatibility and is now documented as LOSSY, with its evidence order: false also covers \"unknown\", which is not a claim of singles — prefer draw, whose null says so honestly.",
+ "?draw=singles|doubles on four listings — GET /matches, /history/matches, /tournaments and /fixtures: the axis the tour filter deliberately collapses; the two compose (?tour=itf&draw=doubles is the ITF doubles slice). A row whose draw is null matches NEITHER value — null is an answer, not a wildcard. An unknown value is a 400 bad_draw with the allowed values. Two honesty notes carried into the spec: on /tournaments the answer comes from the event type alone (a tournament row has no participants to supply the doubles-team evidence matches have), and ?draw=doubles alone also returns mixed and exhibition doubles that no ?tour= value reaches.",
+ "GET /history/coverage (BASIC, or any Historical Data API plan) — the measured completeness rollup per tour × draw bucket: the numbers to read BEFORE choosing what to backtest, in one call instead of paging the archive. A prebuilt snapshot rebuilt nightly right after the completeness ledger reconverges — never computed at read time — so as_of (= built_at) dates every number and ledger_max_computed_at is the newest underlying per-match measurement; 503 coverage_unavailable before the first nightly build. Buckets are atp/wta/challenger/itf/juniors × singles/doubles plus other (team ties, mixed, exhibitions and matches with no stated event type — counted, never dropped, so the totals cannot lie); a bucket with zero completed matches is OMITTED, never emitted as zeros. Each bucket carries the five verifiable numbers (completed, any_tape, point_complete, complete_on_default_read, share), and method states the full measurement rule so every number carries its own definition. As of 2026-08-18: 174,393 completed matches; 91,318 point-complete on the best basis (52.4%) against 81,196 on the default read alone (46.6%); ITF singles 51.1% against ITF doubles 3.5% — do not extrapolate a completeness rate across a tour group.",
+ "tape.starts_at_love and tape.computed_at on /history/matches items (both nullable, present where enabled). starts_at_love follows the SAME best-basis rule as points_complete — true when either measured basis opens at the 0-0 state, so if any on-disk sequence opens at love you can obtain one that does. computed_at is when the ledger last measured the match: every field in the tape's measured block is a nightly-reconverged cache, and this is the as-of to quote with any of them. Null on either means the match has not been measured — never a guess.",
+ "The complete-basis addendum package files. An affected month's manifest may also list tennis_history_points_complete_.jsonl.gz / .csv.gz (marked compression: gzip; bytes/sha256 cover the compressed bytes — exactly the download): for exactly the matches whose complete point sequence exists only as the on-disk reconstruction, the same tape ?points=complete serves (reconstruction contract: null timestamps and null model fields). The base files keep carrying every match's DEFAULT read — already the complete tape for most point-complete matches — and are never rewritten by the addendum; their bytes and sha256 values do not move."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "info.version is now 1.6.0."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.5.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.5.0",
+ "title": "1.5.0: The as-of Elo tape, and 7 more changes",
+ "summary": "Added: The as-of Elo tape. covers_from_start on GET /matches/{matchId}/points (bool|null): whether the persisted stream opens at the match's 0-0 opener (seq 1 is the love-love state) — strictly affirmative; null only when the match has no rows at all. Plus 6 more changes.",
+ "content_html": "
Added
\n
\n
The as-of Elo tape.GET /rankings?system=elo (ULTRA, both modes) — our own computed Elo as point-in-time records: per-player as-of (also via archive_player for the ~62,000 rated people outside the roster, tour required there) and a leaderboard (tour required — the ATP and WTA walks are disjoint; exactly one surface, default overall; min_matches default 20 and activity_weeks default 52 / max 104, both echoed in meta.coverage.qualified). Four independent ladders (overall/hard/clay/grass), ATP from 1877, WTA from 1968, main tours plus challengers plus the futures tier. rating always; rank leaderboard-only; points always null; matches is the ladder-scoped count. meta.coverage gains newest_available, players_rated / players_linked, qualified and the model block (publication_lag_days: 14 — a week's results become effective 14 days after the week begins, so the failure direction is staleness, never look-ahead). The corpus head is frozen at 2026-05-25 and does not advance; elo is never included implicitly. The free current Elo on GET /players/{id} is a DIFFERENT scale (~150 Elo of per-player standard deviation apart) — never present a rating from one scale against a rating from the other. kind=elo yearly bulk packages join /history/packages.
\n
covers_from_start on GET /matches/{matchId}/points (bool|null): whether the persisted stream opens at the match's 0-0 opener (seq 1 is the love-love state) — strictly affirmative; null only when the match has no rows at all.
\n
The point channel family enters the /ws-token vocabulary.point:match:{match_id} / point:slate are documented as point_match / point_slate in the channel vocabulary, listed only when the point feed is enabled server-side and the key's plan carries the point surface. A channel named in that response is a promise; a missing one will not deliver.
\n
\n
Changed
\n
\n
Measured statistics stop rounding the truth.MatchStatisticsMeasured now states: match totals only (no per-set measured statistics); absent = not measured while a present 0 is a real measured zero; the three coverage tiers with their hard zeros — the winners/unforced/forced-errors family historically on ~43% of ATP singles, ~24% of WTA singles and ~47% of tour doubles, and NONE of Challenger, ITF or juniors (24,552 payloads, 2026-07-31) — and that the family has not been available since 2026-07-12 (0 of 4,513 August payloads carry it, measured 2026-08-17).
\n
errors_total is documented as the total of FORCED errors, with the evidence (3,766 payload sides, June–July 2026: equals the per-stroke error sum in 96.2%; smaller than unforced_errors_total in 11.7% of sides — impossible for a superset; per-match points accounting closes only under the forced reading, median residual 0 over 367 matches). Total errors = errors_total + unforced_errors_total; the *_errors shot family is the forced-error breakdown, and groundstroke_errors = forehand_errors + backhand_errors, a rollup rather than an addition.
\n
UTR honesty.system=utr is documented as observed from UTR's public search: withheld ratings are ABSENT, never 0; per-player as-of only — no listing by design (a table of only the players we happen to track would be a fake leaderboard); history since 2026-07-29; and the sweep's deliberate rankless / no-Elo (ITF-skewed) bias with its measured coverage (2026-08-17, players active in the last 60 days): ITF 931 of 5,606 (16.6%), Challenger 197 of 1,903 (10.4%), WTA 43 of 573 (7.5%), ATP 15 of 525 (2.9%).
\n
*(recorded retroactively)* On 2026-08-16 the /ws-token description was expanded in place to teach the push-feed protocol (Centrifugo v2 connect/subscribe/heartbeat, token-per-reconnect, the SDK PushStream pointer) without a version bump; this entry is the changelog record of that change.
\n
info.version is now 1.5.0.
\n
",
+ "date_published": "2026-08-17T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.5.0",
+ "date": "2026-08-17",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "The as-of Elo tape. GET /rankings?system=elo (ULTRA, both modes) — our own computed Elo as point-in-time records: per-player as-of (also via archive_player for the ~62,000 rated people outside the roster, tour required there) and a leaderboard (tour required — the ATP and WTA walks are disjoint; exactly one surface, default overall; min_matches default 20 and activity_weeks default 52 / max 104, both echoed in meta.coverage.qualified). Four independent ladders (overall/hard/clay/grass), ATP from 1877, WTA from 1968, main tours plus challengers plus the futures tier. rating always; rank leaderboard-only; points always null; matches is the ladder-scoped count. meta.coverage gains newest_available, players_rated / players_linked, qualified and the model block (publication_lag_days: 14 — a week's results become effective 14 days after the week begins, so the failure direction is staleness, never look-ahead). The corpus head is frozen at 2026-05-25 and does not advance; elo is never included implicitly. The free current Elo on GET /players/{id} is a DIFFERENT scale (~150 Elo of per-player standard deviation apart) — never present a rating from one scale against a rating from the other. kind=elo yearly bulk packages join /history/packages.",
+ "covers_from_start on GET /matches/{matchId}/points (bool|null): whether the persisted stream opens at the match's 0-0 opener (seq 1 is the love-love state) — strictly affirmative; null only when the match has no rows at all.",
+ "The point channel family enters the /ws-token vocabulary. point:match:{match_id} / point:slate are documented as point_match / point_slate in the channel vocabulary, listed only when the point feed is enabled server-side and the key's plan carries the point surface. A channel named in that response is a promise; a missing one will not deliver."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Measured statistics stop rounding the truth. MatchStatisticsMeasured now states: match totals only (no per-set measured statistics); absent = not measured while a present 0 is a real measured zero; the three coverage tiers with their hard zeros — the winners/unforced/forced-errors family historically on ~43% of ATP singles, ~24% of WTA singles and ~47% of tour doubles, and NONE of Challenger, ITF or juniors (24,552 payloads, 2026-07-31) — and that the family has not been available since 2026-07-12 (0 of 4,513 August payloads carry it, measured 2026-08-17).",
+ "errors_total is documented as the total of FORCED errors, with the evidence (3,766 payload sides, June–July 2026: equals the per-stroke error sum in 96.2%; smaller than unforced_errors_total in 11.7% of sides — impossible for a superset; per-match points accounting closes only under the forced reading, median residual 0 over 367 matches). Total errors = errors_total + unforced_errors_total; the *_errors shot family is the forced-error breakdown, and groundstroke_errors = forehand_errors + backhand_errors, a rollup rather than an addition.",
+ "UTR honesty. system=utr is documented as observed from UTR's public search: withheld ratings are ABSENT, never 0; per-player as-of only — no listing by design (a table of only the players we happen to track would be a fake leaderboard); history since 2026-07-29; and the sweep's deliberate rankless / no-Elo (ITF-skewed) bias with its measured coverage (2026-08-17, players active in the last 60 days): ITF 931 of 5,606 (16.6%), Challenger 197 of 1,903 (10.4%), WTA 43 of 573 (7.5%), ATP 15 of 525 (2.9%).",
+ "*(recorded retroactively)* On 2026-08-16 the /ws-token description was expanded in place to teach the push-feed protocol (Centrifugo v2 connect/subscribe/heartbeat, token-per-reconnect, the SDK PushStream pointer) without a version bump; this entry is the changelog record of that change.",
+ "info.version is now 1.5.0."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.4.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.4.0",
+ "title": "1.4.0: Live per-point events, and 6 more changes",
+ "summary": "Added: Live per-point events. WebSocket points signal. Plus 5 more changes.",
+ "content_html": "
Added
\n
\n
Live per-point events.GET /matches/{matchId}/points (ULTRA) — the live per-point event stream of one match in seq order, paged with ?after_seq= (the resume cursor; up to 500 rows per page, continue on last_seq while has_more). It is the REST catch-up for the WebSocket point frames, which are best-effort with no replay. Per-match pbp_coverage states honestly whether the match has a true per-point stream (point) or only the snapshot score path (game — points empty, an answer rather than an error); per-point coverage is never promised slate-wide. New schemas MatchPoints, LivePoint and PointFrame; new error codes bad_after_seq and points_disabled (the surface switched off server-side).
\n
WebSocket points signal. The native /ws feed's signals array may now name points to opt into one point frame (schema PointFrame) per persisted point of the subscribed matches. Config-gated, off by default — the subscribed ack echoes the signals actually active. The push feed carries the same frames on their own channel family (point:match:{match_id}, point:slate), deliberately separate from the score channels; webhooks gain the matching point event (one POST per live point, X-LTAPI-Event: point). Point frames are events, not states — a missed one does not self-correct; recovery is the REST catch-up read.
\n
Point-complete tape reads.?points=default|complete on GET /history/matches/{matchId}: complete opts out of observed-rows-first precedence and serves a whole-match reconstruction WHOLE, in point order, where one exists (point_winner on every row, null timestamps/model fields per the reconstruction contract); where none exists the response is the default read plus meta.points — no error. Cannot combine with sequence=clean (400 bad_combination); an unknown value is a 400 bad_points; where not yet enabled, complete answers 400 points_read_disabled rather than silently serving the default. The tape's meta.points block (schema PointsMeta) reports the measured point-completeness of exactly the sequence returned — computed at read time, per match, never a stored blanket claim.
\n
Point-completeness on the listing.HistoryMatch.tape gains points_complete (bool|null — the nightly ledger's best-basis verdict; null means not yet measured, never a guess) and completeness (0..1|null), and GET /history/matches gains the ?points_complete=true|false filter (applied after the page is cut, exactly like ?coverage=; anything but true/false is a 400 bad_points_complete).
\n
\n
Changed
\n
\n
Model win-probability copy tells the truth about density. The plan and FAQ copy no longer claims the tape carries the model win-probability "at every point" / "per point": the model stamp is best-effort and rides the rows where the model ran — meta.model_rows is the count, and null is the honest value elsewhere. The current-score snapshot is described as overwritten on every score commit, not "on every point".
\n
HistoryTapeRow.point_winner is documented on ?points=complete rows as well as ?sequence=clean (there the served order IS point order).
\n
info.version is now 1.4.0.
\n
",
+ "date_published": "2026-08-16T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.4.0",
+ "date": "2026-08-16",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "Live per-point events. GET /matches/{matchId}/points (ULTRA) — the live per-point event stream of one match in seq order, paged with ?after_seq= (the resume cursor; up to 500 rows per page, continue on last_seq while has_more). It is the REST catch-up for the WebSocket point frames, which are best-effort with no replay. Per-match pbp_coverage states honestly whether the match has a true per-point stream (point) or only the snapshot score path (game — points empty, an answer rather than an error); per-point coverage is never promised slate-wide. New schemas MatchPoints, LivePoint and PointFrame; new error codes bad_after_seq and points_disabled (the surface switched off server-side).",
+ "WebSocket points signal. The native /ws feed's signals array may now name points to opt into one point frame (schema PointFrame) per persisted point of the subscribed matches. Config-gated, off by default — the subscribed ack echoes the signals actually active. The push feed carries the same frames on their own channel family (point:match:{match_id}, point:slate), deliberately separate from the score channels; webhooks gain the matching point event (one POST per live point, X-LTAPI-Event: point). Point frames are events, not states — a missed one does not self-correct; recovery is the REST catch-up read.",
+ "Point-complete tape reads. ?points=default|complete on GET /history/matches/{matchId}: complete opts out of observed-rows-first precedence and serves a whole-match reconstruction WHOLE, in point order, where one exists (point_winner on every row, null timestamps/model fields per the reconstruction contract); where none exists the response is the default read plus meta.points — no error. Cannot combine with sequence=clean (400 bad_combination); an unknown value is a 400 bad_points; where not yet enabled, complete answers 400 points_read_disabled rather than silently serving the default. The tape's meta.points block (schema PointsMeta) reports the measured point-completeness of exactly the sequence returned — computed at read time, per match, never a stored blanket claim.",
+ "Point-completeness on the listing. HistoryMatch.tape gains points_complete (bool|null — the nightly ledger's best-basis verdict; null means not yet measured, never a guess) and completeness (0..1|null), and GET /history/matches gains the ?points_complete=true|false filter (applied after the page is cut, exactly like ?coverage=; anything but true/false is a 400 bad_points_complete)."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Model win-probability copy tells the truth about density. The plan and FAQ copy no longer claims the tape carries the model win-probability \"at every point\" / \"per point\": the model stamp is best-effort and rides the rows where the model ran — meta.model_rows is the count, and null is the honest value elsewhere. The current-score snapshot is described as overwritten on every score commit, not \"on every point\".",
+ "HistoryTapeRow.point_winner is documented on ?points=complete rows as well as ?sequence=clean (there the served order IS point order).",
+ "info.version is now 1.4.0."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.3.1",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.3.1",
+ "title": "1.3.1: kind=rally bulk packages documented",
+ "summary": "Added: kind=rally bulk packages documented.",
+ "content_html": "
Added
\n
\n
kind=rally bulk packages documented. The /history/packageskind enum gains rally — the charted rally corpus (shot-by-shot) as YEARLY exports, ULTRA, period = YYYY like archive. The kind has been live in the API (it shipped alongside the per-match rally endpoints); the spec simply did not list it. Additive only: the enum, the HistoryPackage.kind field, and the package-shape prose now match the served surface — kind accepts tape (default), rankings (ULTRA), rally (ULTRA) and archive.
\n
",
+ "date_published": "2026-08-07T00:00:00Z",
+ "tags": [
+ "Added"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.3.1",
+ "date": "2026-08-07",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "kind=rally bulk packages documented. The /history/packages kind enum gains rally — the charted rally corpus (shot-by-shot) as YEARLY exports, ULTRA, period = YYYY like archive. The kind has been live in the API (it shipped alongside the per-match rally endpoints); the spec simply did not list it. Additive only: the enum, the HistoryPackage.kind field, and the package-shape prose now match the served surface — kind accepts tape (default), rankings (ULTRA), rally (ULTRA) and archive."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.3.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.3.0",
+ "title": "1.3.0: Shot-level charting, and 7 more changes",
+ "summary": "Added: Shot-level charting. Push-feed token. Plus 6 more changes.",
+ "content_html": "
Added
\n
\n
Shot-level charting.GET /charting/players (ULTRA) — career serve/return profile: serve placement (deuce/ad × wide/body/T), return depth and outcomes, net and serve-and-volley conversion, clutch break/game/set-point serving and returning, winners and unforced errors by wing, rally-length and shot-direction tendencies, summed over the player's charted matches (name keyed, gender=men|women disambiguates, ambiguous fragments refused with candidates). GET /charting/matches/{chartingMatchId} (ULTRA) — every stat family for one charted match, both players, per-set split. Coverage is curated: 11,646 charted matches back to the 1960s, concentrated on the majors, not full-slate.
\n
Push-feed token.GET /ws-token (ULTRA): mints a short-lived signed token plus the push WebSocket URL and channel vocabulary — match:{match_id} per-match streams and slate:all for every live score frame. A separate high-fan-out surface from the native /ws feed.
\n
H2H stat splits. On ULTRA, GET /h2h adds a per-player stats block: serve/return/break-point aggregates over the pairing — archive_serve (serve-side, from 1991) and current (2023+, adding return and break-point conversion, aces and winners), each with meetings_with_stats.
\n
Abuse throttle documented. The 429 family now documents all three body shapes: the per-minute limit (rate_limited with upgrade_url/tier/ price), the per-day quota (rate_limited with scope: "day", limit_per_day and resets_at — an absolute ISO instant), and abuse_throttled with retry_at_epoch — a 24-hour block for clients hammering far past their cap, which a well-behaved retry loop never sees.
\n
\n
Changed
\n
\n
2026-08-06 quota grid re-set (recorded here retroactively). On 2026-08-06 the daily quotas were cut, with no grandfathering: FREE 100/day (was 1,000), BASIC 1,000/day (was 10,000), PRO 10,000/day (was 100,000); ULTRA unchanged at 500,000/day; per-minute limits unchanged (30/60/300/600). The shipped 1.2.0 spec text was edited in place on that date without a version bump — this entry is the changelog record of that change. Older entries below quote the pre-cut grid as it stood then.
\n
Tours. Coverage phrasing is now the five tours everywhere — ATP, WTA, Challenger, ITF and juniors — matching the tour filter enum (atp, wta, challenger, itf, juniors).
\n
WebSocket copy. The subscribe frame is documented as {"topics":["live-scores"]} (+ optional signals) — the previously shown action key is not read by the server. Score frames are documented as carrying the ULTRA model fields (win_probability_p1, danger) live — a null means the model had no output for that point. The 2-connections-per- key limit is stated.
\n
info.version is now 1.3.0.
\n
",
+ "date_published": "2026-08-07T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.3.0",
+ "date": "2026-08-07",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "Shot-level charting. GET /charting/players (ULTRA) — career serve/return profile: serve placement (deuce/ad × wide/body/T), return depth and outcomes, net and serve-and-volley conversion, clutch break/game/set-point serving and returning, winners and unforced errors by wing, rally-length and shot-direction tendencies, summed over the player's charted matches (name keyed, gender=men|women disambiguates, ambiguous fragments refused with candidates). GET /charting/matches/{chartingMatchId} (ULTRA) — every stat family for one charted match, both players, per-set split. Coverage is curated: 11,646 charted matches back to the 1960s, concentrated on the majors, not full-slate.",
+ "Push-feed token. GET /ws-token (ULTRA): mints a short-lived signed token plus the push WebSocket URL and channel vocabulary — match:{match_id} per-match streams and slate:all for every live score frame. A separate high-fan-out surface from the native /ws feed.",
+ "H2H stat splits. On ULTRA, GET /h2h adds a per-player stats block: serve/return/break-point aggregates over the pairing — archive_serve (serve-side, from 1991) and current (2023+, adding return and break-point conversion, aces and winners), each with meetings_with_stats.",
+ "Abuse throttle documented. The 429 family now documents all three body shapes: the per-minute limit (rate_limited with upgrade_url/tier/ price), the per-day quota (rate_limited with scope: \"day\", limit_per_day and resets_at — an absolute ISO instant), and abuse_throttled with retry_at_epoch — a 24-hour block for clients hammering far past their cap, which a well-behaved retry loop never sees."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "2026-08-06 quota grid re-set (recorded here retroactively). On 2026-08-06 the daily quotas were cut, with no grandfathering: FREE 100/day (was 1,000), BASIC 1,000/day (was 10,000), PRO 10,000/day (was 100,000); ULTRA unchanged at 500,000/day; per-minute limits unchanged (30/60/300/600). The shipped 1.2.0 spec text was edited in place on that date without a version bump — this entry is the changelog record of that change. Older entries below quote the pre-cut grid as it stood then.",
+ "Tours. Coverage phrasing is now the five tours everywhere — ATP, WTA, Challenger, ITF and juniors — matching the tour filter enum (atp, wta, challenger, itf, juniors).",
+ "WebSocket copy. The subscribe frame is documented as {\"topics\":[\"live-scores\"]} (+ optional signals) — the previously shown action key is not read by the server. Score frames are documented as carrying the ULTRA model fields (win_probability_p1, danger) live — a null means the model had no output for that point. The 2-connections-per- key limit is stated.",
+ "info.version is now 1.3.0."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.2.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.2.0",
+ "title": "1.2.0: Results archive (1968–2022), and 11 more changes",
+ "summary": "Added: Results archive (1968–2022). Head-to-head. Plus 10 more changes.",
+ "content_html": "
Added
\n
\n
Results archive (1968–2022). The history product now runs in two continuous, non-overlapping halves: the point-by-point tape (2023→now) and the results archive (1968–2022). Four new endpoints, all BASIC (or any Historical Data API plan): GET /history/archive/matches (winner/loser- shaped results — ATP and WTA, main draws, qualifying and the ITF/futures tiers, 1968 through 2022, with final score, seeds, ranks at the time; filters tour, name, from/to, round, level; its own id space; event_date is the TOURNAMENT START date; ends 2022-12-31, exactly where the tape begins), GET /history/archive/matches/{archiveId} (one result, with per-match serve statistics where the era recorded them — null before 1991 mostly, never synthesised), GET /history/archive/players (bios + career-high rank and the week it was first reached), and GET /history/archive/career (career aggregates — sums and ratios of sums only; serve.matches_with_stats states the serve-stat coverage). New schemas ArchiveMatch, ArchivePlayer, ArchivePlayerBio, ArchiveCareer.
\n
Head-to-head.GET /h2h?p1=&p2= (BASIC, or any History plan): the record between two players across both halves of the product. Name-keyed; an ambiguous fragment is refused with the candidate list (400 ambiguous_name); totals count meetings with a known winner and undecided counts the rest; every meeting carries outcome so walkovers and retirements can be excluded. New schema HeadToHead.
\n
Rally construction.GET /rally/matches, GET /rally/matches/{rallyMatchId} and GET /history/matches/{matchId}/rally (all ULTRA): shot-by-shot charted data — serve direction, every stroke with wing/direction/depth, rally length, how the point ended. Its own id space (rally_match_id); 404 not_charted distinguishes "we hold the match but nobody charted it" from "no such match". New schemas RallyMatch, RallyPoint, RallyShot.
\n
Tournament catalogue.GET /tournaments and GET /tournaments/{tournamentId} (FREE): the stable id space Match.tournament_id joins, with city/country from a curated table and category only where the catalogues agree unambiguously — never derived from the name. New schema Tournament.
\n
Rankings listing mode.GET /rankings without player (PRO) returns the FULL published table in rank order for exactly one system — rows carry player_name as published and a null player_id outside our roster, so a top-N has no silent holes; meta.coverage.effective_date names the week served; utr has no listing. Per-player as-of records stay ULTRA. RankingRecord gains player_name, previous_rank (ATP/WTA) and rank_movement (ITF).
\n
List filters./matches gains player (repeatable, max 50, either-participant), country (IOC-style lowercase 3-letter codes — NOT ISO-3166), from/to (UTC day boundaries, every status) and keeps tour; /history/matches gains tour, player and country alongside its existing from/to/coverage.
\n
Match fields.Match gains tour (the same vocabulary as the filter, null never guessed), tournament_id, round_code (controlled vocabulary F…ER, null when unrecognised), a documented event_status enum (Retired | Cancelled | Walk Over | Postponed | Interrupted, with the null-ambiguity caveat) and withdrew (1|2 on Retired/Walk Over, withdrawer = loser by rule); winner is now served for the full archive age.
\n
Fixture fields.Fixture gains start_time (null until the order of play assigns a time), player1_id/player2_id (exact-key roster resolution, never a name match) and round_code.
\n
Tape additions. Clean-sequence tape rows gain point_winner (derived from single-point transitions, never guessed; absent on raw); the tape detail gains a top-level tiebreaks array (observed terminal tiebreak scores per 7-6 set) and meta.model_rows; the history listing's tape object gains model_rows.
\n
Archive bulk packages./history/packages and /history/packages/{period} gain kind=archive — the results archive (1968–2022) as YEARLY exports (period = YYYY), gzipped, alongside the monthly tape packages; the file manifest documents compression.
\n
Statistics final. The in-play statistics coverage vocabulary gains final — the closing figures of a completed match; a finished match cannot be "stale", so age_seconds is null there.
\n
\n
Changed
\n
\n
info.version is now 1.2.0; info.description states the two history halves and the updated tier deltas (tournaments on FREE, the archive family and /h2h on BASIC/History plans, the rankings listing on PRO, rally and per-player as-of rankings on ULTRA).
\n
",
+ "date_published": "2026-08-03T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.2.0",
+ "date": "2026-08-03",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "Results archive (1968–2022). The history product now runs in two continuous, non-overlapping halves: the point-by-point tape (2023→now) and the results archive (1968–2022). Four new endpoints, all BASIC (or any Historical Data API plan): GET /history/archive/matches (winner/loser- shaped results — ATP and WTA, main draws, qualifying and the ITF/futures tiers, 1968 through 2022, with final score, seeds, ranks at the time; filters tour, name, from/to, round, level; its own id space; event_date is the TOURNAMENT START date; ends 2022-12-31, exactly where the tape begins), GET /history/archive/matches/{archiveId} (one result, with per-match serve statistics where the era recorded them — null before 1991 mostly, never synthesised), GET /history/archive/players (bios + career-high rank and the week it was first reached), and GET /history/archive/career (career aggregates — sums and ratios of sums only; serve.matches_with_stats states the serve-stat coverage). New schemas ArchiveMatch, ArchivePlayer, ArchivePlayerBio, ArchiveCareer.",
+ "Head-to-head. GET /h2h?p1=&p2= (BASIC, or any History plan): the record between two players across both halves of the product. Name-keyed; an ambiguous fragment is refused with the candidate list (400 ambiguous_name); totals count meetings with a known winner and undecided counts the rest; every meeting carries outcome so walkovers and retirements can be excluded. New schema HeadToHead.",
+ "Rally construction. GET /rally/matches, GET /rally/matches/{rallyMatchId} and GET /history/matches/{matchId}/rally (all ULTRA): shot-by-shot charted data — serve direction, every stroke with wing/direction/depth, rally length, how the point ended. Its own id space (rally_match_id); 404 not_charted distinguishes \"we hold the match but nobody charted it\" from \"no such match\". New schemas RallyMatch, RallyPoint, RallyShot.",
+ "Tournament catalogue. GET /tournaments and GET /tournaments/{tournamentId} (FREE): the stable id space Match.tournament_id joins, with city/country from a curated table and category only where the catalogues agree unambiguously — never derived from the name. New schema Tournament.",
+ "Rankings listing mode. GET /rankings without player (PRO) returns the FULL published table in rank order for exactly one system — rows carry player_name as published and a null player_id outside our roster, so a top-N has no silent holes; meta.coverage.effective_date names the week served; utr has no listing. Per-player as-of records stay ULTRA. RankingRecord gains player_name, previous_rank (ATP/WTA) and rank_movement (ITF).",
+ "List filters. /matches gains player (repeatable, max 50, either-participant), country (IOC-style lowercase 3-letter codes — NOT ISO-3166), from/to (UTC day boundaries, every status) and keeps tour; /history/matches gains tour, player and country alongside its existing from/to/coverage.",
+ "Match fields. Match gains tour (the same vocabulary as the filter, null never guessed), tournament_id, round_code (controlled vocabulary F…ER, null when unrecognised), a documented event_status enum (Retired | Cancelled | Walk Over | Postponed | Interrupted, with the null-ambiguity caveat) and withdrew (1|2 on Retired/Walk Over, withdrawer = loser by rule); winner is now served for the full archive age.",
+ "Fixture fields. Fixture gains start_time (null until the order of play assigns a time), player1_id/player2_id (exact-key roster resolution, never a name match) and round_code.",
+ "Tape additions. Clean-sequence tape rows gain point_winner (derived from single-point transitions, never guessed; absent on raw); the tape detail gains a top-level tiebreaks array (observed terminal tiebreak scores per 7-6 set) and meta.model_rows; the history listing's tape object gains model_rows.",
+ "Archive bulk packages. /history/packages and /history/packages/{period} gain kind=archive — the results archive (1968–2022) as YEARLY exports (period = YYYY), gzipped, alongside the monthly tape packages; the file manifest documents compression.",
+ "Statistics final. The in-play statistics coverage vocabulary gains final — the closing figures of a completed match; a finished match cannot be \"stale\", so age_seconds is null there."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "info.version is now 1.2.0; info.description states the two history halves and the updated tier deltas (tournaments on FREE, the archive family and /h2h on BASIC/History plans, the rankings listing on PRO, rally and per-player as-of rankings on ULTRA)."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ },
+ {
+ "id": "https://docs.livetennisapi.com/changelog.html#v1.1.0",
+ "url": "https://docs.livetennisapi.com/changelog.html#v1.1.0",
+ "title": "1.1.0: Usage endpoint, and 24 more changes",
+ "summary": "Added: Usage endpoint. As-of rankings. Plus 23 more changes.",
+ "content_html": "
Added
\n
\n
Usage endpoint.GET /usage (FREE — any tier): your own durable daily usage vs quota — tier, limits, today's calls and a 30-day history. Calls to it are quota-exempt. New schema Usage.
\n
As-of rankings.GET /rankings (ULTRA): ranking records as they stood on a date — player (required, repeatable, max 50), as_of, and system (atp, wta, itf_jt, itf_mt, itf_wt, utr). Systems are never collapsed into a single "rank"; UTR carries a rating with null rank/points. meta.coverage.oldest_available gives the earliest date each system can answer for. New schemas RankingRecord, RankingListMeta.
\n
In-play match statistics.GET /matches/{matchId}/statistics (ULTRA): two families, deliberately never merged — DERIVED (rebuilt from the point-by-point record: hold/break %, break points, service & return points) and MEASURED (counted during the match: aces, double faults, the serve split, winners/unforced errors). Each family carries its own coverage, as_of and age_seconds; absent measured fields are omitted, never zero-filled; coverage: none on both families is a 200 with null players, not a 404; a divergence guard withholds measured values when the families disagree. New schemas MatchStatistics, MatchStatisticsSide, MatchStatisticsMeasured, MatchStatisticsFreshness, MatchStatisticsFamily.
\n
Webhooks.POST /webhooks, GET /webhooks, DELETE /webhooks/{webhookId} (ULTRA, direct keys only): we POST the same frames the WebSocket sends to your HTTPS endpoint. Deliveries carry X-LTAPI-Signature (sha256=<hex> HMAC-SHA256 over the raw body), X-LTAPI-Timestamp and X-LTAPI-Event; up to 3 webhooks per key (409 webhook_limit); auto-disable after 25 consecutive failures. The signing secret is returned exactly once, on the 201. New schema Webhook.
\n
Bare price ticks.GET /matches/{matchId}/prices (PRO): recent ticks of the mapped match-winner market without the market wrapper — limit caps at 500, minutes bounds the lookback, 404 when the match has no mapped market.
\n
Tape coverage and sequence./history/matches gains ?coverage= (items are now HistoryMatch — Match plus a tape coverage object); /history/matches/{matchId} gains ?sequence=raw|clean, documents that it works on a LIVE match, and its meta now carries coverage, point_source, raw_rows, unique_states and sequence. Tape rows are now the explicit HistoryTapeRow (null timestamp marks a reconstructed row). New schemas Coverage, HistoryMatch, HistoryTapeRow.
\n
Rankings packages./history/packages and /history/packages/{period} gain ?kind=tape|rankings (default tape) and the listing gains ?year=YYYY for the year-archive view; HistoryPackage documents the kind field and that JSONL is one line per match while CSV is one row per point.
\n
List meta.ListMeta now declares total (nullable — null when the set cannot be counted cheaply) and has_more (page on this, not on count).
\n
Error hints.Error now declares detail (human-readable explanation) and allowed (accepted values on a rejected enumerated parameter).
\n
Price provenance.Price gains price_source and synthetic, so a quote synthesised from mid is never mistaken for a live order book.
\n
Profile provenance. The Analysis profile (and ModelProfile) gains stage (pregame | live | null = unknown), model_version, and — on Analysis — input_state, the score an in-play forecast actually saw.
\n
CORS documented.info.description now states that the REST surface sends Access-Control-Allow-Origin: * (GET/OPTIONS, no credentials mode) and that a FREE key in browser code is acceptable while a paid key belongs server-side.
\n
WebSocket subscribe frame.info.description now shows the full subscribe frame keys — action, topics and the optional signals list.
\n
\n
Changed
\n
\n
info.version is now 1.1.0.
\n
GET /matches/{matchId}/score documents that it is a point-in-time snapshot, pointing at /history/matches/{matchId}?sequence=clean for the sequence of states and /matches/{matchId}/statistics for in-play statistics.
\n
\n
The entries below were previously listed as Unreleased and ship in this release.
\n
Added
\n
\n
History endpoints documented./history/matches/{matchId} (the full point-by-point tape with the per-point model win-probability — BASIC, or Historical Data API Starter+), /history/packages and /history/packages/{period} (pre-built monthly bulk downloads, manifest + ?format=jsonl|csv — PRO, Historical Data API Pro+, or a one-off package pass), plus the from/to date-range filter and the 400/403 responses on /history/matches. New schemas HistoryTape, ModelProfile, HistoryPackage. Purely additive.
\n
Concrete plan deltas.info.description now states exactly what each tier adds over the one below it, with its rate limits (FREE 30/min · 1,000/day; BASIC 60/min · 10,000/day; PRO 300/min · 100,000/day; ULTRA 600/min · 500,000/day), and documents the standalone Historical Data API plans (Starter / Pro / Business / one-off passes). /matches documents that status=completed requires BASIC — the gate was always enforced, just not written down.
\n
WebSocket break-point signals. The /ws subscribe frame now documents an optional signals array; naming break_point opts the connection into two new frames — break_point (schema BreakPoint) the instant a break point arises and break_point_result (schema BreakPointResult) when it resolves. Both are ULTRA-only and purely additive: an existing subscriber that sends no signals sees exactly the frames it saw before. Documented in info.description, the two new component schemas, and the rendered reference.
\n
FREE tier. Self-serve with no card at https://livetennisapi.com/subscribe/free (30 req/min, 1,000 req/day). Covers live and upcoming matches, scores, players and fixtures — the six endpoints now tagged (FREE) in their summary. Purely additive: no endpoint, field, or type changed, and every paid tier keeps exactly the access it had. /history/matches remains BASIC; market prices stay PRO; analysis, live model fields and the WebSocket feed stay ULTRA.
\n
operationId on all 12 operations, so generated clients get stable method names.
\n
info.contact, info.license (MIT) and info.termsOfService.
Rendered reference: correct plan labels + plans/FAQ pages. The tier parser labelled GET /matches/{matchId} as ULTRA (highest-tier-wins scan over "(FREE; +market PRO, +analysis ULTRA)") and did not recognise FREE at all, so every FREE endpoint showed "Plan required: —". It now takes the first tier named. The Plans section gained the FREE row, per-tier daily caps, the standalone Historical Data API plans (Starter / Pro / Business / one-off passes), the Break-point Alerts plans (Free vs Pro), and a FAQ ("How much data can I access on each plan?", "How far back does history go?", "What's in the point-by-point tape?"). Operation descriptions are now rendered, and llms.txt mirrors all of it.
\n
info.title is now Live Tennis API, matching the product name.
\n
",
+ "date_published": "2026-08-02T00:00:00Z",
+ "tags": [
+ "Added",
+ "Changed"
+ ],
+ "_livetennisapi": {
+ "about": "https://docs.livetennisapi.com/llms.txt",
+ "version": "1.1.0",
+ "date": "2026-08-02",
+ "sections": [
+ {
+ "name": "Added",
+ "items": [
+ "Usage endpoint. GET /usage (FREE — any tier): your own durable daily usage vs quota — tier, limits, today's calls and a 30-day history. Calls to it are quota-exempt. New schema Usage.",
+ "As-of rankings. GET /rankings (ULTRA): ranking records as they stood on a date — player (required, repeatable, max 50), as_of, and system (atp, wta, itf_jt, itf_mt, itf_wt, utr). Systems are never collapsed into a single \"rank\"; UTR carries a rating with null rank/points. meta.coverage.oldest_available gives the earliest date each system can answer for. New schemas RankingRecord, RankingListMeta.",
+ "In-play match statistics. GET /matches/{matchId}/statistics (ULTRA): two families, deliberately never merged — DERIVED (rebuilt from the point-by-point record: hold/break %, break points, service & return points) and MEASURED (counted during the match: aces, double faults, the serve split, winners/unforced errors). Each family carries its own coverage, as_of and age_seconds; absent measured fields are omitted, never zero-filled; coverage: none on both families is a 200 with null players, not a 404; a divergence guard withholds measured values when the families disagree. New schemas MatchStatistics, MatchStatisticsSide, MatchStatisticsMeasured, MatchStatisticsFreshness, MatchStatisticsFamily.",
+ "Webhooks. POST /webhooks, GET /webhooks, DELETE /webhooks/{webhookId} (ULTRA, direct keys only): we POST the same frames the WebSocket sends to your HTTPS endpoint. Deliveries carry X-LTAPI-Signature (sha256= HMAC-SHA256 over the raw body), X-LTAPI-Timestamp and X-LTAPI-Event; up to 3 webhooks per key (409 webhook_limit); auto-disable after 25 consecutive failures. The signing secret is returned exactly once, on the 201. New schema Webhook.",
+ "Bare price ticks. GET /matches/{matchId}/prices (PRO): recent ticks of the mapped match-winner market without the market wrapper — limit caps at 500, minutes bounds the lookback, 404 when the match has no mapped market.",
+ "Tape coverage and sequence. /history/matches gains ?coverage= (items are now HistoryMatch — Match plus a tape coverage object); /history/matches/{matchId} gains ?sequence=raw|clean, documents that it works on a LIVE match, and its meta now carries coverage, point_source, raw_rows, unique_states and sequence. Tape rows are now the explicit HistoryTapeRow (null timestamp marks a reconstructed row). New schemas Coverage, HistoryMatch, HistoryTapeRow.",
+ "Rankings packages. /history/packages and /history/packages/{period} gain ?kind=tape|rankings (default tape) and the listing gains ?year=YYYY for the year-archive view; HistoryPackage documents the kind field and that JSONL is one line per match while CSV is one row per point.",
+ "List meta. ListMeta now declares total (nullable — null when the set cannot be counted cheaply) and has_more (page on this, not on count).",
+ "Error hints. Error now declares detail (human-readable explanation) and allowed (accepted values on a rejected enumerated parameter).",
+ "Price provenance. Price gains price_source and synthetic, so a quote synthesised from mid is never mistaken for a live order book.",
+ "Profile provenance. The Analysis profile (and ModelProfile) gains stage (pregame | live | null = unknown), model_version, and — on Analysis — input_state, the score an in-play forecast actually saw.",
+ "CORS documented. info.description now states that the REST surface sends Access-Control-Allow-Origin: * (GET/OPTIONS, no credentials mode) and that a FREE key in browser code is acceptable while a paid key belongs server-side.",
+ "WebSocket subscribe frame. info.description now shows the full subscribe frame keys — action, topics and the optional signals list."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "info.version is now 1.1.0.",
+ "GET /matches/{matchId}/score documents that it is a point-in-time snapshot, pointing at /history/matches/{matchId}?sequence=clean for the sequence of states and /matches/{matchId}/statistics for in-play statistics."
+ ],
+ "notes": [
+ "The entries below were previously listed as Unreleased and ship in this release."
+ ]
+ },
+ {
+ "name": "Added",
+ "items": [
+ "History endpoints documented. /history/matches/{matchId} (the full point-by-point tape with the per-point model win-probability — BASIC, or Historical Data API Starter+), /history/packages and /history/packages/{period} (pre-built monthly bulk downloads, manifest + ?format=jsonl|csv — PRO, Historical Data API Pro+, or a one-off package pass), plus the from/to date-range filter and the 400/403 responses on /history/matches. New schemas HistoryTape, ModelProfile, HistoryPackage. Purely additive.",
+ "Concrete plan deltas. info.description now states exactly what each tier adds over the one below it, with its rate limits (FREE 30/min · 1,000/day; BASIC 60/min · 10,000/day; PRO 300/min · 100,000/day; ULTRA 600/min · 500,000/day), and documents the standalone Historical Data API plans (Starter / Pro / Business / one-off passes). /matches documents that status=completed requires BASIC — the gate was always enforced, just not written down.",
+ "WebSocket break-point signals. The /ws subscribe frame now documents an optional signals array; naming break_point opts the connection into two new frames — break_point (schema BreakPoint) the instant a break point arises and break_point_result (schema BreakPointResult) when it resolves. Both are ULTRA-only and purely additive: an existing subscriber that sends no signals sees exactly the frames it saw before. Documented in info.description, the two new component schemas, and the rendered reference.",
+ "FREE tier. Self-serve with no card at https://livetennisapi.com/subscribe/free (30 req/min, 1,000 req/day). Covers live and upcoming matches, scores, players and fixtures — the six endpoints now tagged (FREE) in their summary. Purely additive: no endpoint, field, or type changed, and every paid tier keeps exactly the access it had. /history/matches remains BASIC; market prices stay PRO; analysis, live model fields and the WebSocket feed stay ULTRA.",
+ "operationId on all 12 operations, so generated clients get stable method names.",
+ "info.contact, info.license (MIT) and info.termsOfService.",
+ "Redocly lint + a structural contract check in CI.",
+ "Rendered reference published to https://docs.livetennisapi.com."
+ ],
+ "notes": []
+ },
+ {
+ "name": "Changed",
+ "items": [
+ "Rendered reference: correct plan labels + plans/FAQ pages. The tier parser labelled GET /matches/{matchId} as ULTRA (highest-tier-wins scan over \"(FREE; +market PRO, +analysis ULTRA)\") and did not recognise FREE at all, so every FREE endpoint showed \"Plan required: —\". It now takes the first tier named. The Plans section gained the FREE row, per-tier daily caps, the standalone Historical Data API plans (Starter / Pro / Business / one-off passes), the Break-point Alerts plans (Free vs Pro), and a FAQ (\"How much data can I access on each plan?\", \"How far back does history go?\", \"What's in the point-by-point tape?\"). Operation descriptions are now rendered, and llms.txt mirrors all of it.",
+ "info.title is now Live Tennis API, matching the product name."
+ ],
+ "notes": []
+ }
+ ],
+ "client_libraries": []
+ }
+ }
+ ]
+}
diff --git a/docs/historical-results-archive.html b/docs/historical-results-archive.html
index 47fad6c..7131df4 100644
--- a/docs/historical-results-archive.html
+++ b/docs/historical-results-archive.html
@@ -202,7 +202,7 @@
Live Tennis API DocsGet a free key ↗Explore the documentation
-
Live Tennis API · docs · version 1.13.88
+
Live Tennis API · docs · version 1.13.89
Historical tennis results API — 1968 onward
How far back does the historical tennis data go, and what is in it? The results archive covers 1968 to 2022: GET /history/archive/matches for results, /archive/players for bios, /archive/career for career aggregates and GET /h2h for head-to-head across both the archive and our own completed matches. All BASIC. Matches from 2013 onward additionally carry a reconstructed point-by-point tape at /archive/matches/{archiveId}/tape (ULTRA).
6 endpoints on this page. Base URL
diff --git a/docs/index.html b/docs/index.html
index a1374f6..e54c44e 100644
--- a/docs/index.html
+++ b/docs/index.html
@@ -23,6 +23,8 @@
+
+
diff --git a/docs/live-scores.html b/docs/live-scores.html
index b53325e..4988730 100644
--- a/docs/live-scores.html
+++ b/docs/live-scores.html
@@ -202,7 +202,7 @@
Live Tennis API DocsGet a free key ↗Explore the documentation
-
Live Tennis API · docs · version 1.13.88
+
Live Tennis API · docs · version 1.13.89
Live tennis scores API
How do you read live tennis scores from the API? Poll GET /matches?status=live for the slate, then GET /matches/{matchId}/score for the lowest-latency read of one match. Both are on the FREE plan. Scores carry sets, games, points and the server; /events and /points give the same match as a stream of changes rather than a snapshot, so a client can catch up without re-reading state.
11 endpoints on this page. Base URL
diff --git a/docs/llms.txt b/docs/llms.txt
index bc6724a..d53f8f6 100644
--- a/docs/llms.txt
+++ b/docs/llms.txt
@@ -9,6 +9,10 @@
Base URL: https://api.livetennisapi.com/api/public/v1
Full text reference: https://docs.livetennisapi.com/reference.html
Changelog (dated, versioned): https://docs.livetennisapi.com/changelog.html
+Changelog feed (Atom): https://docs.livetennisapi.com/changelog.atom
+Changelog feed (JSON Feed 1.1): https://docs.livetennisapi.com/changelog.json
+Each JSON Feed item has a _livetennisapi object with the version and the date. It also holds
+the sections (name, items, notes) and the client library versions named (client, package, version).
OpenAPI spec: https://docs.livetennisapi.com/openapi.yaml
OpenAPI spec (JSON): https://docs.livetennisapi.com/openapi.json
Website: https://livetennisapi.com
diff --git a/docs/openapi.json b/docs/openapi.json
index 842efb8..252dbbc 100644
--- a/docs/openapi.json
+++ b/docs/openapi.json
@@ -2,7 +2,7 @@
"openapi": "3.1.0",
"info": {
"title": "Live Tennis API",
- "version": "1.13.88",
+ "version": "1.13.89",
"contact": {
"name": "Live Tennis API",
"url": "https://livetennisapi.com"
diff --git a/docs/players-and-tournaments.html b/docs/players-and-tournaments.html
index 5cca83b..51a138f 100644
--- a/docs/players-and-tournaments.html
+++ b/docs/players-and-tournaments.html
@@ -202,7 +202,7 @@
Live Tennis API DocsGet a free key ↗Explore the documentation
-
Live Tennis API · docs · version 1.13.88
+
Live Tennis API · docs · version 1.13.89
Tennis players, tournaments and rankings API
How do you look up a player, a tournament or a ranking?GET /players searches by name and GET /players/{playerId} returns one player's bio, current ranking and cached statistics — both FREE. GET /players/{playerId}/stats (PRO) is the player's win/loss record over our completed matches, split by surface, level and season, with ULTRA adding summed serve and return statistics. GET /tournaments is the catalogue that Match.tournament_id joins against, so a match can be resolved to its event without a second lookup. GET /rankings serves rank-ordered listings on PRO. The per-player as-of records, including the Elo tape, are ULTRA.
8 endpoints on this page. Base URL
diff --git a/docs/point-by-point-history.html b/docs/point-by-point-history.html
index d199d27..6323c54 100644
--- a/docs/point-by-point-history.html
+++ b/docs/point-by-point-history.html
@@ -202,7 +202,7 @@
Live Tennis API DocsGet a free key ↗Explore the documentation
-
Live Tennis API · docs · version 1.13.88
+
Live Tennis API · docs · version 1.13.89
Point-by-point tennis data API
Which endpoints return point-by-point tennis data, and how complete is it?GET /history/matches/{matchId} returns the tape for one completed match: the ordered score states we hold for it. The model's win probability sits on the rows that were observed live, and coverage is declared per match. GET /history/matches lists completed matches with their tape coverage, and GET /history/coverage is the measured completeness rollup per tour and draw bucket — a number we publish rather than estimate, with its own as_of. GET /history/packages serves the same data as pre-built monthly bulk files.
8 endpoints on this page. Base URL
diff --git a/docs/push-feed-and-webhooks.html b/docs/push-feed-and-webhooks.html
index 268d508..33fdd5a 100644
--- a/docs/push-feed-and-webhooks.html
+++ b/docs/push-feed-and-webhooks.html
@@ -202,7 +202,7 @@
Live Tennis API DocsGet a free key ↗Explore the documentation
-
Live Tennis API · docs · version 1.13.88
+
Live Tennis API · docs · version 1.13.89
Tennis WebSocket feed and webhooks
How do you receive tennis data as it happens instead of polling? Two ways, both ULTRA. GET /ws-token mints a short-lived token for the high-fan-out push feed, which streams score and point changes as they are written. POST /webhooks registers an outbound HTTP callback for the same events; GET /webhooks lists your registrations and never returns the signing secret. Webhooks require a direct key, not a marketplace one.
4 endpoints on this page. Base URL
diff --git a/docs/reference.html b/docs/reference.html
index cbf6955..8bbc802 100644
--- a/docs/reference.html
+++ b/docs/reference.html
@@ -207,7 +207,7 @@