One thoroughbred's runs in our own settled results

Cost: 3 credits. Every run we hold for one AU thoroughbred, newest first, out of our own settled results.

THIS IS NOT A CAREER RECORD. Each run here is a race WE collected and settled: the same rows /v1/racing/results serves, flattened one-per-horse. The store begins at the first complete card on source.coverage_from and source.coverage_to is the latest race in it, so runs_total is this horse's record SINCE THAT DATE, not its lifetime. The same run never changes once written, so the record only grows — build a longer history forward from here.

WHY YOU WOULD USE IT INSTEAD OF /v1/racing/horses/form. That endpoint is Racing Australia's own form page: deeper (career splits, sectionals, position-in-running, the price path to SP) but dependent on RA serving us the page. Since 2026-09-21 RA's edge answers every fresh form read with a human-verification challenge, so horses we have not already stored cannot be read at all, and we do not automate that challenge. This endpoint has no such dependency — it reads only what we already hold — and it is the honest answer to "what do you have on this horse today". Use both: form for depth when RA lets us in, runs for availability always.

RESOLVING horse — PASS THE NAME, OR ITS horse_ref. A name is folded to letters and digits (the trailing country parenthetical stripped first) and must match in full — the same rule /v1/racing/horses/form applies, so one name resolves identically on both. pe:custo, the horse_ref published on every thoroughbred runner, placing, live-board entry and archive row, is the same fold already done and resolves to exactly the same runs. A Racing Australia horsecode (ra:MjgwNjM0ODg2NDA, or the bare code) is accepted too, but it finds one run rather than a horse: Racing Australia issues a new code for the same horse each day, so a code filter returns the one run that carries it. runner_refs lists every code published under the name; more than one is normal. The other side of grouping on a name is that a registered name reused by two horses returns the union of both. A horse we hold no run for is a 404 and costs nothing; an unknown query parameter is a 422 and costs nothing.

READING A RUN. position is the finishing place, null when the horse did not start (status is scratched) or started and did not finish — abnormal is then RA's own reason cell, verbatim and from no closed list. sp is the official starting price and sp_note RA's favouritism marker; null sp means unknown, never zero. margin is RA's own text in lengths. time_s is the WINNER's time only: RA prints one time per race, in its race header, and no per-runner time in the results table, so since 29 September 2026 a position-1 run carries the race time and every other run's time_s is null. Every run of a race also carries the race-level race_time_s and last_600m_s (RA's header, seconds) and prize_total, and each placed run its prize_won off RA's prize line (dead heats share the places they tie for; 0 outside the paid places); all four are null on runs stored before that date until the meeting is re-read, and before 29 September no run carried any time (0 of 17,344 stored runs, measured 2026-09-28). trainer and jockey are the ones of that day, which is what the result published — profile.trainer on the form endpoint is the CURRENT trainer. field_size counts the horses that ran. weight_kg is the allotted weight and carried_kg the weight actually carried when RA printed one (a claim under it, or an overweight rider over it); race_class is RA's own leading conditions term ('Maiden', 'BenchMark 64'). All three are kept since 28 September 2026. On origin: tabtouch runs, and on runs from 10 August to 27 September 2026 whose result lacked them, race_class and weight_kg come from TABtouch's race page instead: its own class wording ('BM58 HCP', 'MDN-SW') and its printed weight, with carried_kg null. result_id joins back to /v1/racing/results, except on origin: history and origin: tabtouch runs, whose synthetic 'hist:...' id joins to nothing.

stats is computed over every held run, not just the limit returned, and counts only starts (status: ran). runs_total is that same count of starts; held_rows_total is every held row, scratchings included, and runs is the newest limit of those rows -- compare len(runs) with held_rows_total to know whether you have them all. Before 2026-10-13 runs_total counted the scratchings too.

Cached 10 minutes per query; the store itself refreshes hourly at :59.