brucem1967/au-racing-open-data

Australian Racing Open Data

Open, machine-readable datasets for Australian racing — the kind of data that normally sits behind a login, a paywall, or nowhere at all.

Two of these datasets, as far as we can tell, have never existed publicly before: track geometry (turn radii, cambers, straight lengths, first-split distances — gathered by writing to 111 racing clubs and state bodies) and greyhound GPS sectionals at 50-metre resolution.

Everything here is rebuilt and pushed every day. Nothing here is derived from a source that forbids republication — see What is deliberately not here.

Read this first: this data is incomplete by nature and published as it is, not as it ought to be. Coverage is partial, empty cells mean unknown (never zero), values are as supplied by their source and not independently verified, there is no warranty, and none of it is betting advice. Full terms: DISCLAIMER.md.

142,634 rows across 152 venues, 3,444 greyhounds and 116,832 venue-days of weather — 9.6 MB of CSV, rebuilt 2026-09-12.

Dataset	Rows	Coverage
Track geometry	152 venues, 675 sourced values	102 club/body-confirmed
Greyhound sectionals (QLD)	24,405 runs	2025-10-01 → 2026-09-09
Greyhound weights (VIC)	444 weigh-ins	1 race day(s), forward-only
Track weather	116,832 venue-days	2020-01-01 → 2026-08-30

The datasets

Dataset	What it is	Why it's hard to get elsewhere
track_geometry	Turn radii, circumferences, home-straight lengths, cambers, surfaces, sprint lanes and GRV first-split distances for Australian greyhound, harness and thoroughbred venues	No public source exists. Every value was supplied by a club or state body, or derived from OpenStreetMap, and each one carries its own provenance string
greyhound_sectionals_qld	Per-run GPS sectionals from Racing Queensland: 50 m split time, position, distance off the rail, top speed	Published as one CSV per meeting, in a format nobody aggregates
greyhound_weights_vic	Race-day weigh-in weights to 0.1 kg, Victoria	The source page is overwritten daily — miss a day and it is gone forever. This archive only grows forward
track_weather	Daily weather at racing venues since 2020: rain, temperature range, wind and gust maxima	Weather archives are not keyed to racetracks; these are joined to sourced venue coordinates

Each dataset directory has its own README with the full schema, coverage, collection method, known gaps and licence.

Quickstart

git clone https://github.com/brucem1967/au-racing-open-data.git
cd au-racing-open-data
python3 examples/quickstart.py

With pandas:

import pandas as pd

geom = pd.read_csv("datasets/track_geometry/track_geometry.csv")
sect = pd.read_csv("datasets/greyhound_sectionals_qld/greyhound_sectionals_qld.csv")

# Does a tighter first turn slow the 50 m split?
qld = sect.merge(geom, left_on="track", right_on="track_id", how="left")
print(qld.groupby("track")[["split_50m_sec", "turn_radius_m"]].mean())

There is also a dependency-free loader in python/ if you would rather not install anything:

from au_racing_open import load
rows = load("greyhound_sectionals_qld")   # list[dict], stdlib only

Licence and attribution

• Data: ODC-BY 1.0 — use it commercially, build on it, just credit it.
• Code in this repo: MIT.
• Attribution string: Australian Racing Open Data (PandaVision), ODC-BY 1.0

One important carve-out: track_geometry_osm_derived.csv contains values derived from OpenStreetMap and is therefore ODbL, not ODC-BY. It is a separate file precisely so you can take the club-sourced measurements without inheriting share-alike terms. Details in the dataset card.

Under Australian law facts are not copyrightable and computer-generated compilations attract no copyright at all. The licence here is therefore best read as the credit norm we ask you to honour, not a claim of ownership over the facts.

What is deliberately not here

Publishing everything we hold would be easy and wrong. These are excluded on purpose, and the exclusions are enforced in the build script rather than by good intentions:

• Anything from a wagering-conditioned API key — republication would breach the terms that key was issued under.
• UK greyhound (GBGB) data — the UK has a sui generis database right that Australia does not, and it binds downstream users too.
• Betting exchange price/stream data — expressly prohibited by the exchange's terms.
• Commercial form-site scrapes — explicit no-republication terms.
• Source artefacts — the club PDFs, spreadsheets and vendor reports behind these numbers stay private. We publish the values, re-keyed into our own schema.
• Personal data — the people at 111 clubs who answered an email are not part of this dataset. Their addresses, phone numbers and names are stripped automatically, and the build fails if any survive.

How it stays fresh

The datasets are rebuilt from source every day by a scheduled job and pushed here automatically. The build refuses to publish if an exporter fails, if a forbidden source appears in the output, or if anything resembling personal data survives the scrub — so a stale dataset is visible rather than a wrong one being silently shipped.

See CONTRIBUTING.md if you spot an error, have data to add, or run a club that would like its track measured properly.

Why this exists

This data was collected to build racing models. Most of the features tested from it have shown no betting edge at all — which is exactly why it should be public: it is more useful to a community of researchers than it is sitting on one machine. If you find something in here we missed, we would genuinely like to hear about it.

Disclaimer

Short version: incomplete coverage, no warranty, values as supplied and not independently verified, empty means unknown, not betting advice, no personal data, corrections welcome and acted on. The full text is in DISCLAIMER.md — worth two minutes before you build anything on this.

Citation

See CITATION.cff, or:

Australian Racing Open Data (PandaVision), ODC-BY 1.0. https://github.com/brucem1967/au-racing-open-data