# Changelog of Tennis Abstract Scraper | Match History API (`danthedataman/tennis-abstract-data-api`) Actor - **URL**: https://apify.com/danthedataman/tennis-abstract-data-api/changelog.md - **Full Actor documentation**: https://apify.com/danthedataman/tennis-abstract-data-api.md ## Changelog - 2026-10-08: Trim the public README to its buyer facts in a docs-only rebuild of the deployed 0.1.4 file set; preserve runtime, schemas, prefills, pricing and listing. - 2026-10-07: Refresh public listing and README copy in a docs-only rebuild; preserve runtime, schemas, prefills and pricing. ### 0.2 Adds a tournaments mode: every result table of a Tennis Abstract tournament page as flat dataset rows, beside the unchanged players mode. - WTA biography rows cite the fetched shortlist JS file supplying their bio; match rows and career-only null-biography rows keep their classic-page links (item 20, #5). - Empty players input and invalid run-level tour make no source request, including no warm-up, with cap/span validation precedence preserved (item 21, #6). - Transport exhaustion performs no final sleep, rotation callback or warm-up. A later call on the reused client opens and warms a fresh session before its first attempt; attempt budgets and terminal errors are unchanged (item 24). - The buyer README starts from the reviewed 0.1.5 text and adds the measured tournament rows, caps and errors, with the #5/#6 claims updated (item 23). #### What it does - **A `mode` input**, default `players`. An input without `mode` runs exactly as in 0.1: same rows, same defaults, same inputs read. `tournaments` switches the run to tournament pages, and the players-mode inputs are then not read. - **One tournament page per entry in `tournaments`**, written as the page's own `t=` value (`2024-540/Wimbledon`) or as a pasted `https://www.tennisabstract.com/cgi-bin/tourney.cgi?t=…` URL. The value is kept exactly as given — `M25-Ajaccio+H` keeps its `+` — and a value without its `/Name` part is refused rather than guessed. Repeated entries are fetched and billed once. - **Every result table of the page, in page order.** The main draw and the qualifying draw become `recordType: "tournamentMatch"` rows — round, winner and loser with their site player ids, seeds, entry codes, countries and ranks, score, and each side's printed serve rates and break points saved and faced. The other six tables (seed outcomes, this event's finals, surface speed, stat summaries, winners and errors, other events' finals that week) become `recordType: "tournamentTable"` rows, each carrying its table's own header beside its cell values. On the measured 2024 Wimbledon page that is 127 main-draw rows first and 467 rows in all. - **A `maxRowsPerTarget` cap**, default 500, `0` for every row, applied per page in page order. The buyer's own spending limit stops the run between tournaments and writes a `charge-limit` entry naming the tournaments not fetched, as it does between players. #### Correctness choices worth knowing - **Rates ship as the page prints them.** The `singles-results` and `qual-results` match tables print percentages and dominance ratio as text and publish none of the counts behind them except break points saved and faced. Match rates stay strings (`"5.3%"`) and the break-point pair is split into two integers; no rate is derived. Generic `stat-summaries` rows retain published counts (`Pts`, `A`, `DF`, `BPFaced`, `BPEarned`) as printed text. - **Tournament years are checked before billing.** The requested year is compared with the published title and visible event heading when present. A contradiction or unreadable title year is `unreadable-page`, with zero rows from that target. An absent site title, including a commented-only title, is `blocked`. No event-name or event-id derivation is claimed. - **Table selection uses actual HTML attributes.** The table's `id` selects it regardless of quote style, attribute order or whitespace; `data-id` does not. Classification also uses the actual title and main-table `id`. Comments contribute no identity, table, row or cell text, and cannot consume the row cap. Quoted `>` in attributes does not leak into values or split cells. - **Malformed pasted URLs are per-entry errors.** URL parsing failures are recorded as `unsupported-target`; later valid tournament entries still run. - **Drift is refused, not guessed.** A main-draw or qualifying header that differs from the measured one, a row whose width differs from its header's, a player cell or rank or break-point cell outside its measured shape: the page is reported as `unreadable-page` and no row is taken from it. - **A live source fault is one error, not a retry.** A current event can answer with the site's own Python traceback instead of the page; that is recorded once as `source-error` in `ERRORS`, with no row and no second request. - **No tournaments is an empty result.** A tournaments run with no entries fetches nothing and is not back-filled. An unknown `mode` or an unusable `maxRowsPerTarget` ends as an empty result with one `unsupported-target` entry in `ERRORS`, like the 0.1 input errors. #### Not part of this release - The page's "Extreme Performances" block (prose, not a table) and its event header (name, date, surface and draw size) are not exported. - The WTA tournament page is not read: it was never measured as a complete link. The per-player charting tables and the rankings and Elo reports remain planned. ### 0.1 — 2026-09-11 First release: ATP and WTA career match logs from Tennis Abstract as flat typed dataset rows. #### What it does - **One row per match**, with date, tournament, surface, level, round, score, opponent, and the raw per-match serve and return counts the source stores — 51 columns on an ATP row, 48 on a WTA row. Absent values are `null`, never `0`, and the serve/return block blanks as a whole when the source recorded no statistics for that match. - **A whole career in one request.** The ATP career page embeds the full match array, so a career is a single HTTPS GET — 1,516 matches for Novak Djokovic, 2003-01-06 to 2026-08-31, on the measured capture. A WTA career is assembled the way the source's own page assembles it, from the shortlist file plus the older-history file: 130 + 418 = 548 rows for Iga Swiatek. - **Many players per run**, addressed as bare slugs, as pasted player-page URLs in either `p=` form, or through the discovery tokens `@elo-top:N`, `@leaders:1-50` and `@wta-leaders:1-50`, which resolve names through the slug rule the source's own JavaScript publishes. - **Filters over rows already fetched** — season, surface, level, round, tournament, result and head-to-head opponent — so a head-to-head costs the same as the career it came from, and no filter value is ever sent to the source as a request parameter. - **An opt-in player record row** (`includePlayerRecord`) carrying current and peak rank, birth date, hand, backhand, country, active flag, Twitter handle, and the match total the source served for that player. - **Two run ceilings**: `maxMatchesPerPlayer` (default 100, `0` for all) and `maxPlayers` (default 50, maximum 200, applied after discovery so a token cannot fan out past it; a higher value is clamped with a warning, not refused). A run also stops when the buyer's own Apify spending limit is reached, writing a `charge-limit` entry to `ERRORS` naming the targets it did not fetch. #### Correctness choices worth knowing - **Column names come from the page**, not from a hard-coded list: rows are mapped through the header array the source publishes, so an inserted column cannot shift the data. - **Identity is checked against the page's own published name.** The source answers `200` for a slug it does not have and serves a different player's page; such a page is reported as `wrong-player` and no row is taken from it. Where a surface publishes no name of its own, the rows are still delivered and the target carries an `identity-unchecked` **note** — a note beside delivered rows, never a failure. - **Failures never enter the dataset.** Every per-target failure goes to the run's `ERRORS` key-value record with four keys (`target`, `slug`, `errorClass`, `message`); the class vocabulary is ten values, eight page-level and two run-level. - **Empty results exit cleanly.** No players, filters matching nothing, or an input value the Actor cannot act on: the run succeeds with an empty dataset, one `ERRORS` entry where there is something to say, and guidance in the log — never a traceback and never a failed run. - **No derived statistic is emitted.** The source's displayed rates are computed at render time; the README gives the formulas so they can be derived from the raw counts instead of guessed. - **Raw counts only, no doubles.** The measured doubles history file answers `404` and the career page never fills its doubles array, so no doubles switch is offered rather than one that silently does nothing. #### Not part of this release - Store listing craft — icon, published example tasks and Store SEO fields — is the operator's pre-publication step, not something this release performs. - Tournament draws, the per-player charting tables and the rankings/Elo reports are planned for 0.2; charted point logs and draw forecasts for 0.3. Both row families ship into the same dataset, which is why every row already carries `recordType`. - The leaders files are read for names only. Their bulk match arrays are deferred to 0.3 as a mode of their own, so no match datum is lost — the same matches ship per player through `@leaders:` and the career route. - The site's recent-results fragment, `/jsfrags/`, is excluded as a route rather than as data: every row it carries is already on the allowed career page, and the hrefs it adds are on allowed pages too.