Intelligence Analytics
Signal quality, feature importance, and historical calibration
How Does FormRace's Racing Intelligence Perform?
Across 15,048 Australian flat races in a walk-forward backtest (Nov 2025 – Jul 2026), the model's top-rated runner won 31.1% of races. The starting-price favourite won 34.7% of the same races. The model is therefore 3.6 points behind the market on raw strike rate, and backing its top pick flat returned -17.1%. It beat the favourite in 1 of 9 months — best in Dec 2025, worst drawdown in Apr 2026 at −$528.3 flat units.
Where the ratings do earn their place: the winner sits inside the model's top two runners in 50.2% of races and its top pick places 62.9% of the time — enough to cut a 12-runner field down to a short list in seconds, across 391 features per runner you would otherwise read by hand. Use it to spend your attention well, not to outsource the bet.
Frequently Asked Questions
How should I read FormRace's racing intelligence?
As research, not as a promised winner. Across 15,048 walk-forward races (Nov 2025 – Jul 2026) the model's top-rated runner won 31.1% of races while the starting-price favourite won 34.7% of the same races. The model is behind the market by 3.6 points overall and beat the favourite in 1 of 9 months. Use the ratings to narrow a field and to see where price and probability disagree — not as a betting instruction.
What data goes into a FormRace intelligence brief?
391 engineered features per runner across 32 groups: recent form lines, measured sectionals and closing splits, pace projection, weight, barrier draw, class movement, track-condition history, jockey and trainer records and their combination history, gear changes, stewards' race notes, distance aptitude, days since last start, and live market movement.
Does FormRace tell you which bets to place?
No. The value-bet gate was retired on 30 July 2026 after out-of-sample testing showed flagged runners losing money — worse than backing the top pick in every race (-17.1%). Nothing on the board recommends a stake. What the model does publish is a false-favourite read: across 1,895 scored races, favourites flagged suspect won 26.3% against a 32.7% base rate. Where to bet is your call.
How is FormRace different from horse racing analysis sheets?
A tip sheet gives you a selection. FormRace gives you the working: the model's probability beside the market's price, the reasons a favourite looks vulnerable, sectional and pace evidence, live market movement, and a public post-race record of what the board said before the jump. You can disagree with it and see why.
Is FormRace's model backtested honestly?
Every figure here comes from a walk-forward backtest: for each test month the model is trained only on races that finished before that month, and the favourite baseline is measured on exactly the same races. Flat-stake ROI is a $1 win bet at the settled price. We previously published a "+24.1% ROI over 2,798 races" claim; an internal audit found the backtest that produced it was broken, and the claim has been removed from the product rather than restated.