Cost: 2 credits. The day's complete AU thoroughbred card as Racing Australia publishes it — meeting, then race, then runners in saddlecloth order — with no dependency on any bookmaker having framed a market. Per runner: saddlecloth number, barrier, jockey, trainer, weight, runner_ref (Racing Australia's code from its latest read of the field; once race day's read has landed it is the code the results and archives carry for the race — RA issues a new code for the same horse each day), Last-10 figures, Hcp Rating, the trainer/raceentry codes, the emergency flag, and scratched plus a scratching block (RA's own scratch time, late flag, "(Eliminated)" marker and note) for a runner already off. Per race: official name, local start time, distance, the conditions line verbatim, class, prize total and per-place breakdown, bonus note, track name/type, field limit, and the counts. Per meeting: rail, track type, condition — as RA's raw track_condition and, parsed from it, going plus going_rating, the same pair /v1/racing/horses/form publishes — weather, penetrometer, its own as_of, emergency_data, scratching_data and withheld_acceptance_data.

Identity (since 2026-10-12). Every runner carries horse_ref, pe:<folded name>: the HORSE, the same value on /v1/racing/form, /v1/racing/horses/form and /v1/racing/horses/runs, stable across dates and meetings. runner_ref is Racing Australia's code for THIS nomination and is new every time the horse is entered; join histories on horse_ref, never on runner_ref.

Fields-page extras (since 2026-09-29). Every runner also carries what Racing Australia prints under the acceptance table for that horse, verbatim and never inferred: age, sex, colour, sire, dam (without the dam's sire), breeder, owner (RA's Owners line), gear_changes for this start ("Tongue Tie OFF FIRST TIME"), the apprentice claim_kg read off the jockey cell ("(a3/52kg)" -> 3; 0 is an apprentice who no longer claims), probable_weight_kg and penalty_kg from their own columns, and the horse's LAST START — the most recent dated start that is not a trial or jump-out: last_start_date, last_start_position, last_start_jockey, last_start_weight_kg, last_start_barrier, last_start_winner (the horse itself when it won) and last_start_rating, the "Rtg" RA prints on that start's line. Null where RA prints nothing; null on every row stored before 2026-09-29 until the meeting is re-read. MEASURED on the first pass that read them (2026-09-28 20:58Z, 10 meetings, 501 acceptances): pedigree/breeder/owner on 501 (100%), a last start on 423 (84%), a last-start Rtg on 261 (52%), a claim on 107, gear changes on 111. last_start_rating is NOT a substitute for hcp_rating: of the runners with no Hcp Rating, 5 carried a last-start Rtg on that pass (0 of 43 on the saved Moe card of 2026-09-08) — the unrated are overwhelmingly horses whose last start printed no Rtg either.

No prices, on purpose. This is the entry list, not a market. It is available about two days out, which is the point of it — /v1/racing/next-to-go and /v1/racing/best-odds carry prices when books open, and this endpoint stays readable long before they do.

Scratched runners are flagged, never removed. They keep their place in the card so it can be reconciled line-for-line against RA's official sheet.

emergency is true, false or null. Null means the meeting was collected before emergency acceptances were recorded (2026-09-08 01:01:39 UTC) — its emergency acceptances are absent from the source table entirely, so "false" would be a fabrication. The meeting says which case it is in emergency_data: "parsed" or "unknown".

scratched is true, false or null too, for the same reason. RA's scratching sheet does not exist for every date in the window — measured 2026-09-08, 2026-09-05 and 2026-09-06 have no scratching row at all while 09-07 has 96 and 09-08 has 77 — and each meeting's sheet is collected separately, so one meeting can be missing its sheet while its acceptances are present. A row from the sheet is the only positive evidence it was read, so each meeting publishes scratching_data:

• "parsed" — the sheet was read for this meeting; scratched is a real boolean and the counts resting on it are real numbers. scratching_sheet_as_of says when.
• "unknown_date" — no meeting on this date has a scratching row. No sheet exists for the day; scratched is null everywhere on it.
• "unknown_meeting" — other meetings on this date have a sheet and this one does not: either nobody was scratched here or this meeting's fetch failed, and nothing on this card tells them apart, so scratched is null. Do not read a null as "not scratched", and do not read scratching_data other than "parsed" as "clean meeting".

The counts name their population. An emergency acceptance runs only if it is PROMOTED and RA does not publish whether it was, so a count that mixes emergencies with declared runners overstates the field. Every level (race, meeting, day) carries: runner_count (rows held, scratched and emergencies included — a floor wherever runner_set_status is not "checks_passed"), emergency_count, declared_count (runner_count minus emergency_count), scratched_count, unscratched_count (runner_count minus scratched_count, a reconciliation figure that STILL includes emergencies), declared_scratched_count, and declared_remaining_count — the horses expected to jump barring a promotion, which is the number to render as the field. Worked example, Moe R1 on 2026-09-08: 17 acceptances, 5 emergencies (#1, #14-#17), 12 declared, 6 scratchings all of declared runners — 11 rows are unscratched but only 6 declared horses remain. Anything resting on an unknown fact is null, not a guess, and the day totals are null unless every meeting with rows knows the fact. With NO meeting on the card publishing acceptances there is nothing to total: runner_count is a real 0 and every other day count is null, matching what those meetings each report, and counts_cover_every_published_meeting is false.

On a race that is not "checks_passed", every count is a FLOOR. All seven are computed over the acceptance rows this card HOLDS, and such a race is one the card has itself declared short. Two of them looked exact and were not. scratched_count and declared_scratched_count count sheet rows MATCHED to an acceptance row, so a scratching with no row here is in missing_runners and not in the count: Townsville R6 on 2026-09-08 reads scratching_data: "parsed", scratched_count: 0, while RA's sheet lists #4 Five Star King and #9 Kickoff — at least 2. And declared_remaining_count is the number to render as the field ONLY on a "checks_passed" race: Townsville R5 that day is "gap_detected" with saddlecloth #6 unaccounted for and NOT on the sheet — a horse that may well be jumping — so its 11 is "at least 11", never an exact field. Each such race's runner_set_note states the bound explicitly.

Completeness is cross-checked, and the note says what was OBSERVED. ra_race_fields is keyed (meeting_date_aet, runner_key) without a race number, so a horse accepted in two races on one Sydney day has ONE slot — and since 2026-09-08 the acceptance pass gives that slot to NEITHER race rather than publish one race's barrier and weight against the other race's runner: the horse appears in neither race, not in one of them. Do not de-duplicate by runner_key to recover them — there is nothing to de-duplicate. Read missing_runners instead: every withheld slot is named there, in its own race, with its REAL saddlecloth number, source "withheld_acceptance", and a cause_detail naming the slot published in its place where one was. Each race is checked three ways — for saddlecloth holes, against the independently-keyed scratching sheet where one exists, and against the withheld slots recorded as withheld when the card was collected — then reports runner_set_status ("checks_passed", "gap_detected" or "unverified"), a runner_set_note and any missing_runners. Each missing runner carries the cause actually observed for it — duplicate_runner_key (the collision is visible, not assumed: the same runner_key is held elsewhere on this card, or the slot was recorded as withheld), declared_preferred (a declared acceptance elsewhere on the day won the key and this race's emergency slot was dropped — nothing collided, so do not hunt a second race), eliminated_emergency (RA's sheet marks it an emergency and no acceptance row exists) or not_determinable — plus a cause_detail naming the observation. The endpoint does not guess which of several indistinguishable causes applies.

The withheld check is the only one that can see a loss above the top of the numbering, which is exactly where Racing Australia numbers its emergencies: drop #13e from a 1..14 field and 1..12 is perfectly contiguous. Measured 2026-09-10, that silence certified Warwick R2 and Bunbury R2 as "checks_passed" with emergency_count 0 while RA listed 14 and 13 entries against the 12 served each — false statements under a schema that calls declared_remaining_count exact on a checks_passed race. Each meeting publishes withheld_acceptance_data: "parsed" (this card placed this meeting's withheld rows, so its discards are accounted for below), "unplaced" (the pass recorded discards here and this card could NOT place a single one — the meeting's acceptance fields are unpublished, or the race the row names is not on this card; withheld_acceptance_slots_unplaced counts them and completeness.withheld_acceptance_slots_unplaced totals the card) or "unknown" (no row at all — either nothing was withheld, or the pass that read this meeting predates the record). A row is the only positive evidence, so do NOT read "unknown" as "nothing was lost"; completeness.withheld_acceptances says whether the whole date holds any record at all.

Every missing_runners entry carries withheld_acceptance (true/false) — the provenance of the loss, and the field completeness.withheld_acceptance_slots counts. Read THAT, not source: one slot can be named by RA's scratching sheet AND recorded in ra_withheld_acceptances, and it is listed once, keeping the sheet's source (the sheet saw the scratching) while carrying the withheld cause and withheld_acceptance: true. A withheld slot whose horse this card already holds in that SAME race under another saddlecloth is not listed at all: RA renders one horse twice in one race under two saddlecloths, only one row can be published, and no runner is missing from the race.

A past-date card is a snapshot and will never complete. date_status is "past", "today" or "future". Acceptances are collected for today and forward only, so on a PAST date: a meeting at acceptances_published: false will NEVER fill in (live on 2026-09-05: sportsbet bundaberg and sportsbet mareeba) — calling again cannot help, and the meeting's publication_note says so; and the scratchings are frozen at the last collection pass of that day, roughly midday race-day, with no later movement recoverable from anywhere, /v1/racing/changes included. On "today" or "future" the next collection pass CAN add both, every two hours — but that is not a promise that the card converges. Racing Australia stops listing a meeting once it is UNDERWAY, so a meeting's rows stop advancing around 12:15-14:15 AEST while its own races run on to 17:00-19:30. Log-verified 2026-09-09: the 02:15Z pass discovered all six of that day's meetings, the 04:15Z pass discovered one, the 06:15Z pass discovered none. The in-band signal is per meeting — its own as_of and scratching_sheet_as_of. A stamp that stops moving while later passes land on other meetings is a meeting that has gone underway, not a quiet one, and from that point /v1/racing/changes carries the race-day movement.

Why 2 credits. A whole day is one request assembled from four reference tables: a midweek card is a few hundred runners over a handful of meetings and a Saturday well over a thousand across seventeen or more (measured: 372 runners / 4 meetings on Tuesday 2026-09-08; 1,259 / 17 on Saturday 2026-09-05). That is materially more than /v1/racing/track-conditions (1 credit, tens of rows) and is priced level with /v1/racing/next-to-go; it does no cross-book price assembly, so it is not a 3. The underlying data only updates every two hours — calling this once or twice a day is the intended shape, not polling it.

Coverage, honestly. Australian THOROUGHBRED meetings only. Racing Australia publishes no acceptance card for greyhounds or harness, and none for New Zealand, so those meetings are absent here — not empty, absent. History is kept from 2026-09-29 on: until then the reference tables were deleted three days after each meeting, so the store begins at the oldest day that survived (2026-09-26) and grows by a day each day. coverage_from is that first day, read from the tables on every call; any date from it up to three days ahead is served, and an earlier one is a free 422 naming the floor rather than a truthful-looking empty card. A past day is the card as the LAST read of that day left it — see the past-date paragraph above.

How current it is. as_of is the newest RA read backing the meetings actually returned (a venue filter narrows it, so one stale meeting cannot borrow another's freshness), and each meeting carries its own plus scratching_sheet_as_of.

A malformed or out-of-window date, and an unknown query parameter, are both a free 422 — nothing is charged.

{
  "as_of": "2026-09-08T02:16:38.306760Z",
  "completeness": {
    "counts_cover_every_published_meeting": true,
    "known_missing_runners": 6,
    "meetings_acceptances_unpublished": 0,
    "meetings_emergency_data_unknown": 0,
    "meetings_scratching_data_unknown": 0,
    "meetings_scratching_sheet_absent_for_whole_date": 0,
    "meetings_withheld_acceptance_data_unknown": 4,
    "races_checks_passed": 24,
    "races_gap_detected": 5,
    "races_unverified": 0,
    "runners_without_emergency_data": 0,
    "runners_without_scratching_data": 0,
    "withheld_acceptance_slots": 0,
    "withheld_acceptance_slots_unplaced": 0,
    "withheld_acceptances": "unknown_date"
  },
  "coverage_from": "2026-09-05",
  "date": "2026-09-08",
  "date_status": "today",
  "declared_count": 334,
  "declared_remaining_count": 272,
  "declared_scratched_count": 62,
  "emergency_count": 38,
  "meeting_count": 4,
  "meetings": [
    {
      "acceptances_published": true,
      "as_of": "2026-09-08T02:16:38.306760Z",
      "declared_count": 89,
      "declared_remaining_count": 59,
      "declared_scratched_count": 30,
      "emergency_count": 18,
      "emergency_data": "parsed",
      "race_count": 7,
      "races": [
        {
          "conditions": "Maiden, Set Weights, Three-Years-Old and Upwards, Apprentices can claim.",
          "declared_count": 12,
          "declared_remaining_count": 6,
          "declared_scratched_count": 6,
          "distance_m": 1108,
          "emergency_count": 5,
          "field_limit": "12 + 6 EM",
          "missing_runners": [],
          "prize_breakdown": {
            "1st": 17600,
            "2nd": 5760,
            "3rd": 2880
          },
          "prize_total": 32000,
          "race_class": "Maiden",
          "race_name": "Moe Denture Clinic Maiden Plate",
          "race_number": 1,
          "runner_count": 17,
          "runner_set_status": "checks_passed",
          "runners": [
            {
              "age": 5,
              "barrier": 14,
              "breeder": "Mr A Harvey",
              "colour": "bay",
              "dam": "Casino Rose",
              "emergency": false,
              "horse_ref": "pe:backupbarbie",
              "jockey": "Teo Nugent",
              "last10": "2x3x4",
              "number": 2,
              "owner": "S S Racing, E Ralph, S Trovatello, A Wilson, C Taranto, A D Hore-Lacy, J Savage, A Williams, W Williams & V Wilson",
              "raceentry_code": "MzE2ODU1MTEwOTY",
              "runner_key": "backupbarbie",
              "runner_name": "Back Up Barbie",
              "runner_ref": "ra:MTA3MjIxMjU2MzI",
              "scratched": true,
              "scratching": {
                "emergency": false,
                "late": false,
                "scratched_at": "2026-09-07T21:29:00Z"
              },
              "sex": "mare",
              "sire": "Toronado (IRE)",
              "trainer": "Ciaron Maher",
              "trainer_code": "NDU4MDA0NzI",
              "weight_kg": 57.5
            }
          ],
          "scratched_count": 6,
          "start_time_local": "2:00PM",
          "track_name": "Main",
          "track_type": "Turf",
          "unscratched_count": 11
        }
      ],
      "runner_count": 107,
      "runner_set_status": "checks_passed",
      "scratched_count": 34,
      "scratching_data": "parsed",
      "scratching_sheet_as_of": "2026-09-08T02:16:38.306760Z",
      "track": {
        "dual_track": "N",
        "going": "Heavy",
        "going_rating": 9,
        "penetrometer": "5.96",
        "rail": "Out 3m Entire Circuit",
        "track_condition": "Heavy 9",
        "track_info": "Inspection 5:30AM 8/9;",
        "track_type": "Turf",
        "weather": "Fine"
      },
      "unscratched_count": 73,
      "venue": "Moe",
      "venue_id": "moe",
      "venue_norm": "moe",
      "venue_site": "moe",
      "withheld_acceptance_data": "unknown",
      "withheld_acceptance_slots_unplaced": 0
    }
  ],
  "note": "…",
  "race_count": 29,
  "runner_count": 372,
  "scratched_count": 72,
  "unscratched_count": 300
}