1,735AU races quoted, 7 days
12 / 12Australian books quoting now / served
12Median books per race
99.93%Results with every runner placed, where a full-field source exists
72,971Races in the closing-line archive
99.56%Odds API checks passed, 30 days, external authenticated check
When each figure was measured
- Current Racing coverage and feed freshness: measured 16 Nov 00:10 UTC (21 min ago)
- Current Results completeness: measured 16 Nov 00:20 UTC (12 min ago)
- Current Closing-line archive measurement: measured 16 Nov 00:20 UTC (12 min ago)
- Current Sports consensus check: measured 16 Nov 00:20 UTC (12 min ago)
- Current Results latency and price sources: measured 16 Nov 00:23 UTC (8 min ago)
- Being written Closing-line archive: newest record from 15 Nov 23:12 UTC
Availability, freshness and correctness are separate questions: /status says whether the API is answering, this list says whether each measurement and the archive are current, and the sections below say what the checks found. A current measurement is not proof the data is correct; see what these checks cannot show. Machine-readable as measurements in /data-quality.json.
Racing coverage and feed freshness
Australian races over the trailing 7 days, measured 16 Nov 2026 00:10 UTC and recomputed every 30 minutes.
Degraded now:
none.
"Time since last read" is how old our last successful read of that bookmaker was, over the last 24 hours. It is not the age of the price itself — capture is change-only, so a price can be minutes old while the read confirming it is seconds old. The per-book price ages, the books-per-race distribution and the SQL behind every figure are on the coverage report. New Zealand: 130 races from 12 books in the same window.
Brands and price sources. The 12 brands a customer key can read are 11 independent price sources.
Ladbrokes and Neds closed at the same price on 90.7% of the runners both quoted, so each such pair is one source.
Closing win prices compared to the cent over the trailing 7 days to 16 November 2026; a pair identical on at least 85% of runners counts as one source.
Results completeness
Races from 2026-11-10 to 2026-11-16. A result has two parts, measured separately below. The result — placings, and dividends where the bookmaker publishes them — is what settles a bet: 4,150 of the 4,152 races that ran have a final one. The full field — every runner's finishing position, and box or barrier, time and margin where the source prints them — comes from a separate source per code: of the 4,152 races that have one, 4,149 carry it (99.93%), 2 are pending and 1 are overdue.
How the table adds up. Every stored result is in exactly one of full field, pending, overdue, past retry, abandoned or no source: 4,167 = 4,149 + 2 + 1 + 0 + 15 + 0. "Pending" means the race is younger than its collector's normal pass; "overdue" means the pass has run without writing a field but the collector keeps trying; "past retry" means no scheduled pass still looks at the race. "Result final" and "Dividends" count the same races by whether the placings are final and whether any dividend is stored, so neither is part of that sum; where the settling feed publishes no dividends, availability.dividends on each race reads not_published. Races we priced but hold no result for are not in the table, because it counts stored results; the next section accounts for every priced race.
Priced races joined to a result
The denominator is every Australian and New Zealand race in the closing-line archive that started between 10 Nov 2026, 00:00 and 16 Nov 2026, 08:18 Sydney time, at least 3 hours before this reading: 1,465 races. 1,427 carry their result under their own race_id, and 1 reach it through a recorded alias (1 with a rescheduled start): 97.47% joined. 1 is held for review, where the evidence is ambiguous, and never joined automatically (1 filed under another racing code). 36 have no stored result.
A race joined through an alias answers to either race_id: GET /v1/racing/results?race_id= accepts both and lists the other in race_id_aliases. Each figure is result_join on /v1/racing/results/coverage, defined term by term in result_join.definition.
Read from GET /v1/racing/results/coverage, the same endpoint a customer calls.
How soon results arrive
Minutes from the scheduled start until a race's result can be read from /v1/racing/results, over the trailing 7 days to 16 November 2026. A race that jumps late makes the figure longer, never shorter.
Measured from when our results row was first written (race_results.created_at) against the scheduled start.
Closing-line archive
3,084,789 rows across 72,971 races from 24 bookmakers, 2026-08-04 to 2026-11-15. A row is a genuine closing line (is_closing_line) when the book was still quoting that price inside 300 seconds of the jump: 91.28% of rows — 96.78% from 2026-09-29, when the archive also began confirming a price that had not moved in those last minutes, and 85.03% before it, when only a price change inside them counted. Every other row carries the last price seen before the book stopped quoting, and says so. Of the 21,079 Australian and New Zealand races a result can exist for (from each code's first collected result, started at least 3 hours ago), 98.04% carry one.
Rows the archive flags about itself — flagged, not silently dropped: 4,542 where one meeting may have been split under two venue spellings, 1,145 with a suspect runner name, 1,134 whose race appears to duplicate another race record, and 19,084 for runners the result marked scratched (hidden from /v1/racing/closing-lines by default). These flags catch known shapes of bad row. They cannot catch a bookmaker feed that was wrong at source — see the limits below.
Price-quality filter (version 2026-11-09.1): 2,806,476 rows (90.98% of the whole archive) are a genuine closing line, not scratched, and carry no venue-split, runner-name or duplicate flag. This is about the price only: a row can pass it and have no result, and most international rows have no results source. Over 2026-10-17 to 2026-11-16, 916,045 of 1,063,136 rows pass it and have a result (research_filter=research_ready); 107,085 pass the price filter with no result. A worked example that applies it and prints what it excludes is in historical racing odds in Python.
Read from GET /v1/racing/closing-lines/coverage.
Where each bookmaker's closing prices start is under when each dataset starts.
Sports price sanity check
Every 15 minutes, each bookmaker's head-to-head price on each game is compared with the median of the other bookmakers quoting that game. A price more than 20% away is flagged as a likely stale or bad quote. Games need three or more books to be checked.
Read from GET /v1/data-quality/audit/latest. This is a consensus check, not a comparison with each bookmaker's own site, and it covers sports only.
Odds API availability
An authenticated request is made to the API from outside the production host every five minutes, the same path a customer's request takes; a check passes only when the response is valid and, during the Australian racing day, carries fresh prices, monitored since 2026-10-13.
99.85% of checks passed over 7 days (2,015 checks) · 99.56% of checks passed over 30 days (8,633 checks).
Internal on-host health checks have a longer record and are shown separately on the status page; they are not used for this figure. These are observed measurements, not a contractual SLA. Every surface, the blind periods and the method are on the status page.
When each dataset starts
From each date, the data is collected for every race in its scope. Earlier races may carry some of it. How much of the latest week has arrived is in the tables above.
The closing-line and results dates are read live, from archive_from on GET /v1/racing/closing-lines/coverage and coverage_from on GET /v1/racing/results/coverage; a bookmaker that joined the archive later is dated by the earliest race it has in the archive. Price history older than the live window is served from the same permanent archive.
What these checks cannot show
- Racing freshness (age_seconds) is the time since our last successful read of that bookmaker, not proof the bookmaker's own price moved. A feed that is frozen or wrong at source can look fresh.
- A connector marked healthy is reachable and writing prices. That is liveness, not correctness.
- There is no continuous, automated check yet that compares each racing bookmaker's served prices with the bookmaker's own website. Report a price that looks wrong and it is investigated against the source.
- The sports consensus check covers head-to-head markets on the sports listed, and flags only prices more than 20% away from the other books. It does not see racing, and it does not catch small errors.
Test it against your own use case
The free tier needs no credit card, and the sandbox needs no key at all.
Get a free API key
See pricing