← API reference
API
Changelog
Customer-visible changes to the API, most recent first. Internal refactors, performance work and data-quality fixes that leave a response unchanged are deliberately not listed. Each date is anchor-able, so an entry can be linked directly.
2026-11-03
- GET /api/v1/racing/meetings no longer answers a database abort with 503. The last board built for the same request shape is kept for 30 minutes and served instead, as an ordinary 200 carrying meta.stale: true, meta.stale_age_seconds (the real age of the board you were given) and the X-Stale-Age header, under Cache-Control: private, no-store so no cache keeps a stale board. A request shape with no stored board still answers 503, with meta.degraded and meta.failed_sources naming what failed. meta.tier and meta.rate_limit are always rebuilt from your own request, so a stale board can never carry another key quota. Additive: no field was removed or renamed.
- A racing card is no longer short a race because the first card feed we merge was. Where the card is missing a race that another feed has runners and prices for, the race is now added with its runners, its own race identity and its price rows, so it pages, joins and settles like every other race on the card. A race published under a placeholder name (VENUE R07 03/11) now takes the real name where a feed has one. Nothing is removed or renumbered: the card stays in race order, a race already on the card keeps its own name and runners, and a feed race with no runners is not published as an empty shell.
- A read that runs out of its time budget now answers 503 with retry_after instead of 500. A deadline was being classified as an internal fault, so a brownout on our side was reported to you as a server error; it is now classified as transient, like any other dropped read, and carries the same retry hint. This covers the routes that fan out per request, including /teams, /results, /racing/futures, /racing/runner-stats, /racing/field-form and /players/compare.
2026-11-01
- GET /api/v1/racing/sectionals no longer serves a race-level split ladder on a past run dated before 2026. One collector files a whole race as a single past-form entry — its segment times are the winner's, so a ladder served beside any other runner of that race is that runner's name on the winner's time. Those rows now serve split_times: null and split_ranks: null and carry only the runner's own finish, margin_lengths and sp. meta.pre2026_run_level_split_rows reports how many rows were withheld. Runs dated 2026 and later are unaffected, as is every other collector.
- GET /api/v1/racing/sectionals rows carry the past run's own finish, margin_lengths and sp. Additive — no field was removed or renamed.
2026-10-28
- GET /api/v1/racing/sectionals serves the last-200/400/600/800-metre splits of each past run and a per-split rank, beside the 600 m split it already served. split_times.l200_s, l400_s, l600_s and l800_s are the time for that closing section of the run; last200_s…last800_s carry the same values under the names already in use. A split is served only when the run's own segment times tile that whole window — a run with a gap in it, or without a segment reaching the finish, answers null for that split rather than a number that is too fast. split_ranks ranks each split within the race field (1 is fastest; ties share the rank) and is null on a runner-only lookup, where there is no field to rank against. section_ranks carries the source's own per-segment rank. distance_points now includes the run's finishing position. split_times.last600_s and l600_s are always the derived sum of the trailing three 200 m segments; the 600 m figure the form feed publishes is a different measurement about 2.5 s longer on the same run, so it is no longer served and a run carrying only that figure is not returned. Additive — no field was removed or renamed.
- The same route serves roughly twice as many past runs. The 600 m split and the full segment array behind it are now read together, so a run carrying only the array is served instead of being dropped — on the busiest track-and-distance cards the array covers about twice as many past runs as the published split does.
2026-10-23
- Every v1 error body carries the documented envelope. A handful of hand-rolled routes answered a bare {"error": "..."} object with no success key, while every other route answered {"success": false, "error": {...}} — so a client branching on success saw undefined and fell through its error handling. Verified live: GET /api/v1/racing/field-form?race_id=<unknown> answers 404 with success: false, and GET /api/v1/racing/sectionals/<unknown> answers 404 with success: false. Additive: no error code changed.
- GET /api/v1/racing/results/{date} no longer drops the tail of a big day. The tote-card read behind the dated form was bounded in a way that truncated a large card, so the back end of a busy day could be missing. The read is now paged to completion.
- GET /api/v1/racing/odds-history only caches a healthy window. The shaped window was cached without checking that the read behind it succeeded, so a short or failed read could be served from cache for its full TTL. A degraded window is no longer stored.
- GET /api/v1/racing/futures attributes the Cox Plate to Flemington, where it is run. It was served under Moonee Valley, which runs a different feature on the same card.
- GET /api/v1/racing/meetings no longer reports a settled status for a race with no result behind it, and two cards at one venue on one day that are different race types are no longer collapsed into a single meeting.
- A superseded archive copy of a race is never served in place of the live one, on GET /api/v1/racing/results or its dated form.
- One operator's board is no longer served under another operator's name on five routes, including the all-bets board.
- GET /api/v1/racing/results serves a race once. A meeting whose stored name is a service or sponsor variant of a track we already publish — "Doncaster Bags" is the same card as "Doncaster", "Rosehill Gardens" the same as "Rosehill" — is keyed to the canonical track slug instead of being filed as a second meeting, so one race is not served twice under two venue slugs. Duplicates written before the fix are retired from the response: the duplicate groups on 1–19 October are down from 502 to 108 and continue to fall.
- Every metered response now carries X-Credits-Cost — the weighted debit for that request — beside X-Credits-Limit, X-Credits-Remaining and X-Credits-Reset. GET /api/v1/racing/meetings, GET /api/v1/odds-history and GET /api/v1/sports did not send it; they do now. A request that returns no rows is refunded, so a balance can rise between two calls. GET /api/v1/racing/predictions costs 1 credit; the docs said 2.
- Results rows carry meeting_id: the meeting's own identity, not the day its race was filed under. meeting_key is built on the Sydney day of the jump, so a UK or Irish card that runs after midnight in Sydney carried the next day's key, and two meetings at one venue on adjacent days collapsed onto one key. meeting_id is built on the venue's own day, so it matches the meeting the meetings endpoint buckets. Additive — meeting_key is unchanged and still served.
- meta.data_note explains a partial results page instead of leaving it to be inferred: whether races on the date are still pending (status "pending", no finishing order claimed) and how much of the page carries a jump time — "scheduled_time populated for N of M rows". It is null when there is nothing to say, and the notes combine rather than one replacing another.
- A results row whose stored document carries no track name is no longer served with an empty race_key_v2 and a null scheduled_time: the race key the board filed it under is used as a fallback, so the documented join key is populated on every row. Measured on 2 October: 0 of 500 served rows carry a blank race_key_v2, where 70 of 70 rows on 1–4 October were blank before the fix.
- Rugby league NRLW results are served under the live competition rather than a retired one, which served no completed 2026 fixtures at all. The season now serves 93 rows, 71 of them completed 2026 fixtures, every one of them matching the source draw (0 fixtures unmatched in either direction), each carrying odds_event_id to reach the odds history.
2026-10-22
- A failed read answers 503 with retry_after instead of 200. Routes that could not complete a read previously answered 200 with an empty or partial body, which was indistinguishable from a genuine empty result. 503 now means the read failed and is retryable; an empty data array still means no rows. Applied to the racing routes, the opportunities routes and the shared route factory, so the whole surface answers the same way.
- GET /api/v1/racing/field-form distinguishes a failed lookup from an unknown race. An unknown race_id is a 404; a failed lookup used to answer the same 404, claiming the race does not exist when we simply could not read it. A failed read is now a 503.
- GET /api/v1/export serves every dataset GET /api/v1/bulk advertises. The two routes disagreed on the dataset list, so datasets the bulk endpoint advertised were unreachable through export. The route also reads the primary store rather than failing when the archive flag is off.
- GET /api/v1/smart-money/stats read the wrong table and answered 500 on every call. It now reads the line-movement table.
- GET /api/v1/h2h read a raw data blob instead of the columns the table holds, so the response was built from the wrong shape.
- GET /api/v1/status guards its sync probes and no longer re-runs the fetcher on a cache hit. A probe failure is reported rather than propagated.
- Registering a webhook validates the URL and the body. A malformed URL and a malformed JSON body are both rejected with a 400 instead of being stored.
- The endpoints hint block on the opportunities umbrella pointed at a path that did not resolve. Corrected.
2026-10-21
- GET /api/v1/results no longer drops a row whose stored data carries no excluded key. Such a match was silently absent from the response; it is now served like any other.
2026-10-20
- GET /racing/field-form resolves every race id /racing/meetings serves. A request for one race used to be answered with the next race on the card (190 of 229 races on 20 Oct) or 404 (39 of 229): race_id is now matched against the race record itself, by race_key_v2, numeric race id, date+venue+race number, or {meetingId}_R{n}.
- GET /racing/field-form keeps scratched runners in data.runners with "scratched": true instead of dropping them, matching /racing/meetings. data.field_size is the starter count and data.entrant_count the row count, so a race whose card is fully scratched returns 200 with field_size 0 rather than 404.
2026-10-18
- Races with an empty box no longer list a placeholder runner, and a runner with no named jockey is served with "jockey": null rather than a placeholder string. Check for null before displaying a jockey — on 18 October 2,789 of 5,473 runners (51.0%) carried jockey: null, and no placeholder runner name or placeholder jockey string was served at all.
2026-10-17
- /api/v1/racing/results with a date no longer fails with a server error. A read of a heavy date returns its full race list — 737 races and 5,473 runners for 18 October — on both the query form and the dated path, rather than an error page.
- /api/v1/odds-history pages are bounded: a page carries at most 1 MB of JSON and ends with meta.next_cursor. Walk meta.next_cursor until meta.truncated is false rather than computing pages from limit — a page can end before the limit when it would exceed the cap.
- /api/v1/odds-history returns the headline h2h market of each bookmaker by default. Send ?markets=full for the complete market set (spreads, totals, props). meta.markets and meta.market_outcomes_omitted say which body you got, so a summary is never mistaken for the full card. opens, latest and movement are identical either way.
2026-10-16
- GET /racing/connections/season-stats resolves a jockey or trainer name to every stored spelling of that person before aggregating, and lists the spellings it merged in data.spellings. C.J.Waller, C Waller and Chris Waller are one trainer, 428 starts and 46 wins this season (10.75%); a lookup by a single spelling used to answer that spelling alone (147 starts, 10.88%). Different people who share a surname are still kept apart.
2026-10-15
- Racing meetings responses carry a live exchange overlay per runner — exchange.back / lay (with resting sizes), plus the projected starting price exchange.sp_near / sp_far, refreshed while the market is open (pre-off only). Price keys are omitted rather than nulled when the feed has no value for them.
- Racing odds-history responses are gzip-compressed. A dated page that was served as ~42 KB of JSON now arrives as ~5 KB on the wire (measured 5,111 bytes against 41,762 uncompressed) with content-encoding: gzip and vary: Accept-Encoding, so a client that sends Accept-Encoding: gzip gets the smaller body without changing anything.
- /api/v1/results rows carry a new is_final boolean, and the status mapping is documented: resulted and completed are final, pending is not. status is never null.
- /api/v1/racing/results rows carry local_date and jump_time_utc. Results are filed under the AU date of the jump, while local_date is the meeting's own local date, so UK/IE meetings keep their local day.
- scheduled_time on /api/v1/racing/results/{date} is filled from the board archive, so rows that previously carried null now carry the scheduled jump time.
- Cricket results are aliased to their /odds-history id, so a fixture served by /api/v1/results resolves on the odds-history lookup, and cricket tournament fixtures are served under their format key rather than being unreachable.
- AFLW fixtures are filed under their own sport key with bookmaker prices served, instead of appearing on the AFL board with no prices.
- id_match_rate corrected. It was capped at 50% for fully joined sports.
- Cricket: domestic first-class matches are filed as domestic first-class rather than as Test matches.
2026-10-14
- A read outage on the odds archive no longer returns 401, which reads as a dead key. A database or read failure is now 503 with retry_after; a spent monthly allowance is 429 with retry_after set to the seconds until reset; 402 stays reserved for insufficient credits. Applied across every v1 route.
- /api/v1/odds-history only caches the primary history read, and cache keys are versioned. A primary outage serves the last known-good page, otherwise 503 with retry_after. Rows with no team identity are never served as current data. meta.truncated now reports true at the limit cap.
- /api/v1/odds-history accepts order=asc|desc and commence_from. A malformed commence_from is a 400 rather than a silently widened window.
- No response names its upstream data source. Provenance is served as a neutral class: primary, secondary or exchange.
2026-10-13
- Gameday routes resolve sport aliases on every endpoint, so a key such as basketball_nba resolves its aliases everywhere rather than on a subset.
- /api/v1/racing/predictions serves real per-runner predictions.
- Player props serve combo prop markets, and every served row guarantees commence_time.
- Racing runner-form and field-form serve the union of the live form board and the served results history again (the union had silently returned no rows).
2026-10-12
- Racing runner-form serves a union of the live form board and the served results history (backfilled to 2018), matched on normalised runner name plus racing code; each row carries form_source ("recent" | "history") and a live row wins a tie.
- Racing responses carry tote dividends per runner: tote_win and tote_place are {price, status, provenance, captured_at, seconds_before_jump} or null. Sectionals are addressable by runner name, race id or venue plus meeting date, with no horse code required.
- An API key's historicalDays window is read from either jsonb level, so a key minted under the legacy nested shape is clamped as labelled instead of serving its full history.
- /api/v1/status exposes per-table racing data freshness.
2026-10-05
- Racing results are identity-gated: rows without runner identity are never served, a partially-identified race comes back as status "pending" with winner null, and status is never null.
- Racing prices expose bookmaker_type (fixed | tote | exchange | group | derived | capped), group rows carry brands[], and ?include=tote,exchange opts in to the non-default price lines.
- Added GET /api/v1/me/usage (tier, credits used/remaining/limit, usage %, reset) and completed the X-Credits-* and X-RateLimit-* response headers.
- /api/v1/racing/results carries SP, tote and per-field provenance.
- Racing board rows serve runner notes[].
- pending_count describes the response, not just the board fill.
- A raced day is never answered with an empty results 200, and the type= alias is accepted.
- /api/v1/odds-history resolves a bare hex id against the prefixed legacy id.
2026-10-03
- meta.truncated is reported on all 35 v1 routes that previously omitted it.
2026-10-02
- /api/v1/results ?event_id= returns one row per fixture and resolves a legacy odds id.
2026-10-01
- Added GET /api/v1/racing/flucs, per-book price fluctuation history for one race.
- /api/v1/sports exposes coverage flags: results_published, id_match_rate and in_season.
- /api/v1/results excludes rows published under an odds id, adds an event_id lookup and accepts aflw/nrlw aliases.
- A database timeout returns 503 with Retry-After rather than 500.
- /api/v1/odds-history resolves id lookups by legacy_id as well as id.
- Racing runners carry implied_win_pct and tab_rating_rank from the canonical runner API.
2026-09-29
- Results and meetings rows carry per-race data_quality.
- A truncated signal is added to all paginated routes, with a limit+1 fetch so it is accurate.
2026-09-28
- /api/v1/racing/results carries race_key_v2 for meeting disambiguation.
2026-09-24
- /api/v1/racing/results carries race_key_v2 and source_id for meeting disambiguation.
- A country= filter is added, and unsupported params return 400.
- The bookmaker registry includes additional books and adds a neutral source field.
- The TAB bookmaker label is renamed from "Fixed Odds" to "TAB".
2026-09-22
- The racing meetings date filter default limit is raised from 20 to 200, with truncation meta.
2026-09-21
- Accuracy figures include n (sample size).
2026-09-19
- /results accepts a date query param.
2026-09-18
- "abandoned" is removed from the final (resulted) status set.
- Meetings sort ascending and results accept a race_type filter.
2026-09-17
- Live exchange prices are wired into the v1 meetings API.
- Naive AEST/AEDT jump times are converted to UTC before status checks.
2026-09-14
- Added GET /api/v1/data-quality for odds-vs-results mapping.
2026-09-05
- /racing/meetings accepts an opt-in date filter.
2026-09-04
- Meetings responses surface bib, gear and allowedBetTypes.
- Racing results expose isScratched.
2026-09-03
- Published this OpenAPI 3.1 spec and rebuilt this reference from a full read of every v1 route.
- /odds-history falls back to 16 bookmaker feeds when the primary archive is unavailable.
- MCP server documentation corrected: URLs, OAuth endpoints and client configs.