How Our Predictor Works: From Screenshot to Top-3 Picks
A transparent look at the technology powering Racing Ready Australia.
Racing Ready's predictor turns a single screenshot into a full race analysis in under 30 seconds. Here's a transparent, step-by-step breakdown of exactly what happens from the moment you hit upload to the moment your top-3 picks appear.
< 30s
Average time from upload to top-3 results
1,200+
Variables evaluated per runner in the prediction engine
100%
Screenshot-based - no manual data entry required
Step 1 - Screenshot to Structured Data (OCR)
When you upload a race card screenshot, our Optical Character Recognition (OCR) pipeline immediately goes to work. It parses the image to extract runner numbers, horse names, barrier positions, jockeys, trainer names, weights, and the current live odds - directly from the visual layout of your TAB or Racing.com screenshot.
TAB App (iOS & Android)
TAB.com.au web screenshots
Racing.com race cards
Punters.com.au race pages
For best results: take the screenshot in portrait mode, include the full runner list, and ensure the odds column is visible.
Step 2 - Data Enrichment
Once the runner names are extracted, our system instantly cross-references them against our historical racing database. For each runner, we pull in:
- Last 12 starts: finishing positions, distances, track conditions, class, prize money
- Track and distance records: win rate and place rate at today's specific venue and distance
- Barrier statistics: win rate from today's barrier at this track
- Jockey statistics: win rate for this jockey at the venue and at this distance
- Trainer statistics: trainer win rate at the distance over the last 12 months
- Weight carried: how today's allocated weight compares to the horse's historical norm
- Going preferences: how the horse performs on Good, Soft, and Heavy going
- Current market odds: the live TAB price at the time of upload
Step 3 - The Prediction Engine
With enriched data loaded for every runner, our ensemble machine learning model calculates a win probability for each horse. The model was trained on over five years of Australian TAB race results across all states and has been calibrated to minimise systematic bias by track type, distance class, and going condition.
The result isn't 'Horse X will win.' It's 'Horse X has a 38% true probability of winning - and the market is only pricing it at 25%.'
Step 4 - Value Score and Confidence Rating
The final output for each runner combines two metrics: the Value Score (model probability versus market implied probability) and a Confidence Rating (how strong and consistent the underlying data signals are). A HIGH confidence, positive value score is the combination most worth acting on.
TOP PICK badge: the runner with the highest model probability in the field.
Value Score > 0: model thinks the horse is paying more than it should.
HIGH confidence: the underlying data signals are consistent and strong.
MEDIUM confidence: the horse has a genuine case but the data has some noise.
LOW confidence: treat this as a lottery ticket - possible, but speculative.
Frequently Asked Questions
Does the predictor work for harness and greyhound races?
The current version is optimised for thoroughbred flat racing at TAB-recognised Australian venues. Support for harness and greyhound racing is on the development roadmap.
What if my screenshot is blurry or partially cut off?
The OCR pipeline requires the runner names and odds columns to be clearly visible. If the upload fails to extract data cleanly, you'll be prompted to try again with a clearer screenshot. Landscape screenshots often cut off key columns - portrait mode works best.
Does the model update with today's late scratchings or gear changes?
The model uses data available at the time of upload, including the live odds. If a key runner is scratched after you upload, re-uploading a fresh screenshot will re-run the analysis on the updated field.