StableBet
Professor Furlong and Pascal at the AI Lab
THE AI LAB
THE AI LAB · THE OPEN CHALLENGE

Can YOU beat the bookies?

27,936real races say the bookmaker’s margin always wins in the end. The leaderboard says somebody is still on top of it. Pick a strategy, take a virtual £10,000 onto the track at £10 a race, settled at the real Starting Price of races that actually ran, and see where your strategy places.

A research game, not tips. Virtual money only, no account needed. 18+ · please gamble responsibly.

The board to beat

Back the favourite · 3 races a day at £10 a bet · 83 weeks-£1,817
peak +£173

Up +£173 after five weeks, then the margin does its work. That early run is the hook, not an edge.

The board to beat · 28,305 real races

  1. 1Each-way the favourite-8.59%
  2. 2Back the favourite-8.72%
  3. 3Back the second favourite-11.98%
  4. 4A random horse-21.52%
  5. 5Back the outsider-34.69%

24 strategies tested. None finished in profit. No player has beaten them either. The first name on that board is still going.

Virtual £10,000 at real Starting Prices. No account, no real money, ever. 18+ · BeGambleAware.org

Every figure here is the Lab’s published record, not a simulation of one.

Bankroll

£10,000

Pick a strategy to begin

0 / 1,000 to the first board

The other half of the question

Our AI model takes this test too — its value picks run live, every race day, in public. See whether the AI is beating the bookies →

How the game settles its bets

Every bet settles exactly the way the Lab settles its published research: every GB race in the Lab's dataset: flat stakes settled at Starting Price, non-runners void, fallers and pulled-up settled as losses, joint favourites split the stake. Each-way bets pay a quarter of the odds on the places, with the place count set by field size. No prices are invented, no results are simulated — the game replays what the track already decided.

Eighteen systems on the roster, plus your own

Every betting system from the Lab’s published research that this dataset can replay is playable — and the builder lets you set your own rules (pick, bet type, staking plan, price window, going, field size) within the same honest settlement. A few of the Lab’s tests need data the game doesn’t carry (ratings, the draw, prize money); those are shown locked, with their verdicts one click away.

  • Back the Favourite. £10 on the shortest price in every race. Joint favourites split the stake. The Lab’s verdict →
  • Second Favourite. £10 on the second-shortest price. Races with a single price are skipped. The Lab’s verdict →
  • The Outsider. £10 on the biggest price in every race. Joint outsiders split the stake. The Lab’s verdict →
  • Each-Way Favourite. £10 each-way on the favourite: £5 to win, £5 to place at a quarter of the odds. Small fields settle win-only. The Lab’s verdict →
  • The Pin Sticker. £10 on one random runner per race. The pin picks, you watch. The Lab’s verdict →
  • Odds-On Only. £10 on the favourite, but only when it's odds-on (SP below 2.0). Everything else is passed over. The Lab’s verdict →
  • Small-Field Favourite. £10 on the favourite in fields of five or fewer. Fewer rivals, fewer ways to lose. The Lab’s verdict →
  • Big-Field Favourite. £10 on the favourite in fields of sixteen or more. A research run: the dataset holds only a few hundred such races, under the 1,000-bet qualifying line. The Lab’s verdict →
  • Jumps Favourite. £10 on the favourite, jumps racing only. Fences and hurdles included at no extra charge. The Lab’s verdict →
  • Flat Favourite. £10 on the favourite, Flat racing only. No fences between you and the payout. The Lab’s verdict →
  • Soft-Ground Favourite. £10 on the favourite when the ground rides soft or heavy. Mudlarks only. The Lab’s verdict →
  • Each-Way Longshot. £10 each-way on the outsider, but only at 10/1 or bigger. The place part does the heavy lifting. The Lab’s verdict →
  • The Double. Back-to-back favourites in £10 doubles: two must land for anything back. One dead leg kills the slip. The Lab’s verdict →
  • The Four-Fold. Four consecutive favourites in one £10 accumulator. All four or nothing. The Lab’s verdict →
  • The Lucky 15. Four favourites, fifteen £10 bets: four singles, six doubles, four trebles, one four-fold. £150 a slip. The Lab’s verdict →
  • Four-Fold Roulette. Four pin-picked randoms in one £10 accumulator. Pure chaos, compounded. The Lab’s verdict →
  • The Martingale. Back the favourite, £10 base. Double the stake after every loss, reset after a win. The classic chase. The Lab’s verdict →
  • The Fibonacci. Back the favourite with stakes walking the Fibonacci line: 1, 1, 2, 3, 5... One step up per loss, back to the start on a win. The Lab’s verdict →
  • Top-Rated Horse (locked). Needs the official ratings book. The game dataset carries prices and results, not ratings.
  • Handicap Favourite (locked). Needs to know which races are handicaps. Race names aren't in the game dataset.
  • Top-Rated in Handicaps (locked). Needs ratings AND the handicap flag. Double-locked.
  • Lowest Draw (locked). Needs stall numbers. The draw isn't in the game dataset.
  • Festival Favourites (locked). Needs prize money to spot the big meetings. Not in the game dataset.
  • The Old Stagers (locked). Needs horses' ages. Not in the game dataset.
  • Follow the AI (locked). That's the AI's own live test. It plays its version next door, every day.
  • AI's Confident Picks (locked). Needs the model's probabilities. The AI keeps those for its own test.

Questions, answered straight

Is this real money?

No. You play with a virtual £10,000 at £10 a race, and nothing is ever staked for real. Every result comes from a real race that already happened, settled at the official Starting Price — the same convention every experiment in the Lab uses. No account, no deposit, no real money, ever.

Where do the races come from?

Every race is real: 27,936 GB races from 2023-10 to 2026-07, the same dataset the Lab's betting-systems verdicts are computed from. Your batches are drawn in a shuffled order, so no two runs replay the same sequence, but every price and every result is exactly what happened on the track.

How do the leagues work?

Three leagues, nothing in between: 1,000 bets, 2,500 bets and 5,000 bets. Cross a line and your run enters that league, scored exactly as it stood at the line — a run that reaches 5,000 sits in all three. You can mix systems inside one run (switch between batches, or line the whole plan up in advance) and you enter with a nickname, nothing else.

Why did my result differ from the Lab's published figure for the same strategy?

Variance. The Lab's verdict pages measure each strategy over the full dataset; your run is a sample from it. Over 1,000 bets a favourite-backer usually lands within a few points of the published figure — but 'usually' is the point. The gap between your run and the full-sample truth is the most useful thing this game shows.

What happens if I go bust?

The run ends — you can't stake money you don't have, even virtual money. Any league line you'd already crossed keeps its entry, because each league scores your run as it stood at that line, not at the end. Then you pick a new strategy and go again.

Is this telling me how to win at betting?

No, and it can't: no strategy in the Lab's testing has beaten the market over the full sample. This is a research game about what betting strategies really do over time, played on real data. If anything here changes how you bet, let it be the size of the numbers, not the order of the leaderboard. 18+, BeGambleAware.org.

This game uses virtual money and settles on races that have already been run — nothing here involves real gambling. If you do bet, set limits you can afford and stick to them. Support is free and confidential: BeGambleAware.org · 18+.