Every horse has six hidden ratings: top speed, acceleration, stamina, consistency, late finish, and quick start. When you press race, a small physics-style simulation runs behind the scenes. Each horse gets a performance score built from its ratings plus a random draw of luck, and the fastest finishing time wins.
The winner is decided by that simulation the instant you press the button. The horses on screen are just replaying a result that already happened, so nothing in the animation changes the outcome.
Because the luck term is real but bounded, the best horse does not always win. Over many races the strong horses win more often, but any single race can go to an outsider. That property is what makes the odds meaningful.
Distance changes the weights. In a sprint, speed and acceleration dominate and stamina barely counts. In a marathon, stamina and a strong late finish matter far more. A horse that is unbeatable over a sprint can fade over a longer trip, so switching distance genuinely reshuffles the favourite.
The odds beside each horse are not invented. Before every race the engine runs the same simulation 3,000 times and counts how often each horse wins. That win frequency is the horse's true probability. The payout shown is the fair payout for that probability, reduced by a fixed house margin (next tab).
Each lane shows the horse's name, its live odds, and a form arrow (Up, Even, or Down). Form is a small hidden value that drifts over time, like an animal having a good or bad patch, and it nudges the rating. The picks below also carry a small value tag when the offered price beats what a rational bettor should accept and a trap tag when it does not, so the honest expected-value read is always in front of you.
Yes, and here is how. Randomness comes from a seeded pseudo-random generator. Each race is seeded from the clock and the trial count, so every race gets a fresh, unpredictable draw. The same maths runs for every horse, including the one you backed. The engine never looks at your bet before deciding the result.
Real betting is not a fair coin flip, and this game copies that honestly instead of hiding it. If a horse truly wins one race in four, a fair payout would be four times your stake. This game pays a little less. The shortfall is the house margin.
A 10% margin means that, on average, every 100 coins staked returns about 90 over the long run. You can win, and win big, in the short run. But the longer you play, the more the margin grinds the balance down. That is not a bug; it is the single most important thing the game is built to let you feel.
The bars below compare each horse's true win chance from the simulation with the chance implied by its displayed odds for the current distance. The red mark is the true chance; the green bar is what the odds imply. The odds always imply a slightly higher chance than reality, and that consistent gap is the margin, working exactly as designed.
Green bar = chance implied by the price on offer. Red mark = the real chance from 3,000 simulations. The gap between them is the margin.
When the odds are shown honestly and the maths is fair within a fixed margin, does a real person still bet the way behavioural economics predicts? A perfectly rational bettor would stake small, back only positive-value prices, ignore what just happened, and treat every coin the same. Decades of research say people do none of that. This project measures how far your play departs from the rational baseline, one testable claim at a time.
This is a within-subject, single-case design. There is no separate control group; instead each hypothesis is a within-person contrast that uses your own trials as their own comparison (for example, your bets after a win versus after a loss). Each race is one trial. For every trial the engine records the decision context before the outcome is known and the choice you made, so the comparisons below are defined in advance rather than fished for afterwards.
Every hypothesis reports the size of its effect in plain units (coins, or a fraction of your balance) alongside the number of trials on each side of the contrast. Because this is one person's data, the report is deliberately careful: it will say supported, reversed, no clear effect, or not enough trials yet, and it never claims statistical significance from a handful of races. Real single-case work needs a run of trials before a pattern means anything, so the report keeps telling you how close you are to that.
Nothing is decorative, and each feature exists to make a hypothesis measurable.
Grounded in Kahneman & Tversky, Prospect Theory (1979); Thaler & Johnson on the house-money effect (1990); Reid (1986) and Clark et al. (2009) on near-miss effects; and the classical gambler's-ruin problem.
There is no data collection to consent to. The software runs offline in your browser, stores your progress only on this device, and transmits nothing: no names, emails, location, device identifiers, or gameplay. This is educational research by Raman, not a commercial product, and it involves no real money. If real gambling ever stops feeling like a game for you, support is linked from the end-of-session screen.
Close this and keep playing whenever you like. The experiment exists to see how one person really chooses under risk. The game exists so those choices feel real enough to be worth measuring.
Betting is now locked. This is only a game, but the same rule matters with real money: never chase a loss. If gambling ever stops feeling like fun, help is there for you.
Gambling supportPlease gamble responsibly. In-game coins have no real value.