Data-driven greyhound predictions — how the algorithm works
RateThatDog's prediction engine is built on five years of UK greyhound racing data and validated against 188,000+ pre-race snapshots. It doesn't pick horses. It doesn't tip football. It does one thing: uses structured data to find the greyhounds most likely to win, at every track, every day.
Why greyhound predictions need a different approach
Greyhound racing runs over 100 races per day across 19 UK venues. Traditional form-reading at this volume is impossible — there isn't enough time to process each card properly before the traps open. Most tipping services respond to this by reducing everything to a single speed rating or a simple win-rate lookup. That misses most of the signal.
The conditions of each race vary enormously: track geometry, trap assignment, distance, class level, pace profile, trainer form and recent sectional trends all interact. A dog that is genuinely fast at Romford's 400m is not automatically a contender at Towcester's 500m. A Trap 1 specialist at Hove is not the same proposition at Newcastle, where trap bias is minimal. Meaningful predictions require each of these factors to be modelled separately and then combined in a way that reflects how they actually interact in race results.
That is what the RateThatDog composite score does. It was built by training on five years of historical race data, tuned by grid-search across weight combinations, and validated out-of-sample on a held-back 30% of the 188,000+ snapshot dataset. The weights differ by distance band — sprint, standard, long — because different factors dominate at different trip lengths.
The six components of the prediction model
Composite Score
The core prediction number — a distance-weighted blend of performance rating, field speed, suitability and bend speed. Grid-searched and validated out-of-sample across 188k race snapshots. Sprint races weight performance and suitability; standard-distance (380–480m) races lean heavily on field speed. The composite lifts top-pick accuracy from 23.4% (legacy) to 25.5%.
How composite score works →Field Speed Rating
A per-race ranking that measures how fast a dog is relative to its field — not in absolute terms, but normalised to the race. Field speed is the single biggest predictive factor at standard distances (380–480m), where early pace sets the race. A dog ranked #1 for field speed in its race wins significantly more often than the raw form line implies.
Filter by field speed in Dog Selector →Suitability Models
Four independent suitability scores — track, distance, trap and class — each built on win and place rates at the relevant conditions, confidence-adjusted for sample size. A dog with high track and trap suitability on a biased course like Hove or Monmore commands a very different probability than a course first-timer.
Understanding suitability scores →Performance Rating
A 0–100 score built from five factors: pace (30%), finishing position (25%), beaten distance (20%), sectional progression (15%) and grade adjustment (10%). It measures how well a dog performed — controlling for the quality of race — rather than just recording result.
Performance rating explained →Hot Dogs Algorithm
Hot Dogs is a filtered subset of the composite model: the system surfaces only dogs with composite score 60 or above that have no close rival in their race. This tighter filter removes marginal picks and delivers a 28.3% strike rate — validated on the same 188k+ snapshot dataset. Typically 8–15 selections per day across 19 tracks.
Today's Hot Dogs →Strong Picks
A broader shortlist built on consistent top-2 composite rankings across multiple prediction models. Strong Picks includes dogs that score well under both the composite method and the form-suitability model, providing a higher-volume selection pool for each racing day.
Today's Strong Picks →How to use the predictions
The quickest route to the day's best predictions is the Hot Dogs page. It refreshes before each race and shows every dog meeting the composite 60+ threshold with no close rival. These are the picks where the model is most confident — and the historical 28.3% strike rate reflects that confidence.
For a broader shortlist, the Strong Picks page surfaces dogs that score consistently across multiple models. These are higher-volume picks — useful if you're building systems or following a particular track.
For deeper research on a specific race or dog, the Dog Selector lets you filter the full day's runners by composite score, field speed rank, trap suitability, trainer form and more. You can save filters as betting systems and track their performance over time.
Individual race pages show the full composite breakdown for every runner — composite score, performance rating, suitability scores (track/distance/trap/class), bend ranking, and the race confidence score. The race confidence number tells you how predictable a race is — high confidence means the model's top pick has a meaningful edge over the field; low confidence means the race is genuinely open.
See today's predictions
Predictions update live throughout the day as races approach. Hot Dogs refreshes before each card.
Frequently asked questions
Are these AI greyhound predictions free?↓
Yes. All predictions on RateThatDog — including Hot Dogs, Strong Picks, racecards and the Dog Selector — are free to use.
What is a composite score in greyhound racing?↓
The composite score is RateThatDog's core prediction metric: a number from 0–100 that combines performance rating, field speed, suitability (track/distance/trap/class) and bend speed into a single, distance-weighted ranking. A composite of 60+ in a race where no other dog is close is the threshold for a Hot Dog pick.
How accurate are the greyhound predictions?↓
The algorithm achieves a 25.5% top-pick strike rate overall, validated on 188,000+ pre-race snapshots using a 70/30 train/test split by race date. The Hot Dogs filter (composite 60+, sole leader) achieves 28.3%. Blind baseline for a six-runner race is 16.7%.
Which greyhound tracks are covered?↓
All 19 licensed GBGB venues in the UK: Hove, Romford, Nottingham, Sunderland, Newcastle, Monmore, Yarmouth, Towcester, Central Park, Doncaster, Sheffield, Kinsley, Harlow, Oxford, Suffolk Downs, Dunstall Park, Star Pelaw, and others.
How are greyhound predictions different from tipsters?↓
RateThatDog publishes data and model outputs, not tipster opinions. Every prediction is produced by the same consistent algorithm applied identically to every race — no cherry-picking, no retrospective selection. The strike rates are validated on historical data, not self-reported.