Ranked field
Every active runner appears exactly once in model order.

AI-powered horse racing predictions, fair odds and explainable rankings—from Ascot to Saratoga, built for a global racing audience.
Saratoga · Race 8
1⅛m · Dirt · 9 starters · Tote market
Copper State
3-112
5/1
+43%
Midnight Circuit
21-43
6/1
+33%
Bluegrass Theory
5-221
9/2
−25%
Top AI prediction edge
Copper State · tote 43% above fair
Global coverage, verified
UK, Irish and Hong Kong cards form the live foundation. North American racing is the next launch priority, with tote-aware pricing and US track coverage planned from the start.
Connected regions
Provider status · right now
Great Britain
Live
Ireland
Live
Hong Kong
Live
International
Selected cards
North America
Launch priority · feed pending
Planned for major US and Canadian circuits with pari-mutuel odds, fractional display and state-aware wagering context.
Australia
Expansion roadmap
Expansion labels describe product direction, not live availability. A region activates only after its licensed feed and settlement workflow are verified.
Explainable AI predictions
True Overlay turns verified race data into structured AI predictions—then shows the probability, fair price, evidence and uncertainty behind every call.
Inspect our AI methodologyEvery runner is ranked. No cherry-picked tip, no missing longshot, no hiding the inconvenient call.
We turn estimated win probability into a fair decimal price, then show the market gap in plain sight.
Source fields, freshness, model version and uncertainty stay attached to every published analysis.
Decision guardrail
Pre-commitment keeps a useful model from becoming an emotional shortcut. Define the bankroll, stake and stop point first.
Only use a separate bankroll you can afford to lose.
Set a fixed maximum stake before opening a card.
Pause when the session limit you chose is reached.
Live AI predictions
Predictions appear only after a complete, fresh racecard is stored. No feed means no fabricated picks.
—
Live cards
Integration ready · credentials pending
Connect the database and licensed racing feed to begin importing real cards. Until then, every live-data surface remains honestly unavailable.
View platform status →Analysis standard
Every active runner appears exactly once in model order.
Probabilities are validated and normalized to 100%.
Calculated deterministically from the normalized probability.
Shown only when a permitted, fresh market price is stored.
Evidence labels resolve to non-empty source fields for that runner.
Missing inputs and data completeness remain visible beside the estimate.
Model ID, prompt version, input hash, source time, and generation time are retained.
Official results close the record without deleting unsuccessful predictions.
Operating lifecycle
Fetch licensed cards and results behind server-only credentials.
Map provider records into stable races, runners, prices, and result identities.
Create one structured shared analysis for the current source input.
Check identities, evidence, totals, arithmetic, freshness, and completeness.
Lock the prediction with its model, prompt, hash, and timestamp.
Attach official results while retaining every successful and unsuccessful call.
Institutional standards
A mature analysis product needs operating rules that remain intact after a win, a loss, a correction, or a missing feed.
An analysis is published only when its racecard exists in the normalized store and every cited field can be traced to that record.
The product presents estimated chances, fair prices, market gaps, and uncertainty. It never turns an estimate into a promised outcome.
Shared pre-race predictions are immutable. Once official results arrive, wins and losses settle into the same public ledger.
Incomplete form, ratings, comments, or prices reduce the displayed completeness score; the system does not invent replacements.
Reference library
The complete generation and validation contract.
Region registry, feed cadence, and freshness rules.
Product interpretation and derived API reference.
Public configuration readiness at request time.
Architecture, key handling, and trust boundaries.
Provider, AI, price, performance, and affiliate context.
The Blog
How overlays work, what the model measures, and how to bet an edge without going broke finding out.
Photo: Keith LukeFundamentals7 min read
A price-led approach asks a different question from ordinary tipping: are the available odds larger than the estimated chance? Here's what that means in practice.
Photo: Mathew SchwartzModel Inputs6 min read
A horse that finishes fourth can still have run a revealing sectional. Timing data can add context that the finishing position alone misses.
Photo: Taylor SondgerothStrategy8 min read
An edge only pays if you're still solvent when it shows up. How to size stakes so a normal losing run never takes you out of the game.