Betting systems, tested on real races
We run popular betting systems over thousands of real races with our in-house model and publish exactly what happens — including when they lose. No tips, no “system for sale”. Most clever-sounding betting ideas don't survive contact with the data; here's the proof.
Our in-house model lost 16.8% ROI on the pre-registered Oct-Nov 2024 backtest window.
This page publishes what it predicts and tracks every result. We do this because nobody else does — the methodology is open, the losses are visible, the analysis is honest. The model output is presented as a comparison to the market, not as a recommendation to back, lay, or stake on any runner.
Read the full methodology in our in-house AI horse-racing model write-up. Track the running ledger on the Stablebet track record page.
Gambling can be addictive. Please bet responsibly. Free, confidential support from GamCare, GamStop and BeGambleAware. See our responsible-gambling page for more.
Experiment #1 — Should you bet more when you're confident?
A common idea: stake bigger when your model loves a runner, smaller when it doesn't — like a card counter pressing a hot deck. We tested it head-to-head against flat stakes.
The model's “confidence” doesn't predict profit — so sizing your stakes by it doesn't help.
Across 6,531 real races, the runners the model rated most above the market (its “value” picks) returned -22.4% — its worst bets. The ones it rated below the market (mostly favourites) returned -5.0%, the least-bad. A card counter wins because the count truly shifts the odds; this signal doesn't, so there's no “hot deck” to press.
ROI by how much the model “likes” the horse
Flat-stake ROI for top picks in each model-edge band. If confidence predicted profit the bars would climb as edge grows — instead the most-confident bands are among the worst.
| Model edge | Bets | Strike | Profit / loss | ROI |
|---|---|---|---|---|
| below 0 | 3,632 | 36.4% | -£3,212 | -8.8% |
| 0–3% | 709 | 17.1% | -£993 | -14.0% |
| 3–6% | 664 | 13.0% | -£979 | -14.7% |
| 6–10% | 773 | 10.5% | -£1,734 | -22.4% |
| 10%+ | 753 | 12.7% | -£377 | -5.0% |
Flat stakes vs sizing by confidence (±50%)
The fair comparison is ROI (return per £ staked); absolute profit/loss just tracks how much was staked in total.
| Strategy | Staked | Profit / loss | ROI | Worst drawdown |
|---|---|---|---|---|
| Flat £10 | £65,310 | -£7,294 | -11.2% | -£7,864 |
| Sized by edge (±50%) | £60,725 | -£7,001 | -11.5% | -£7,781 |
What this shows. The model's edge — how much higher it rates a horse than the market does — does not predict profit. Its strongest “value” picks (the 6–10% band) returned -22.4%, while the runners it rated below the market (the 10%+ band — mostly favourites) returned -5.0%, the least-bad of the lot.
Read the head-to-head honestly. Sizing by confidence actually staked £4,585 less overall (it backs off the favourites too), so it lost £293 less in cash, with a slightly smaller drawdown. The fair, per-£ measure — ROI — landed within about a third of a percentage point either way (-11.2% vs -11.5%) — noise at this sample, and it shifts with the exact sizing rule. In plain terms: it made no meaningful difference. The robust signal is the bucket pattern above, not any one staking rule.
The honest takeaway. Betting variation only pays when your signal genuinely predicts results. This one points the wrong way — the model's most-backed picks are its worst — so there was no edge to amplify. The answer to “should you bet more when you're confident?” is, on this evidence, no. We publish it because that's what the data says — see the full track record and the methodology. Backtest of 6,531 races; flat £10 win on the top pick is the baseline.
Experiment #2 — Does the Martingale system work?
The most popular “can't-lose” system: double your stake after every loss so one win recovers everything. We ran it on the model's real top-pick results.
No — Martingale doesn't beat the odds. It just delays the wipe-out.
Doubling your stake after every loss feels bulletproof — until a losing streak hits. In the model's real record the top pick lost 34 times in a row. To keep doubling through that, the next bet alone would have to be £171.8 billion — from a £10 base.
Longest losing streak
34 in a row
at 26.1% strike
Next stake to continue
£171.8 billion
doubling £10, 34 times
£1,000 bankroll
Bust in 16 bets
wiped out 2024-10-03
Why it fails. Martingale wins a little most of the time, which feels like a system that works — but it pays for those small wins by risking everything on the rare long losing run. The doubling is exponential: ten losses turn a £10 stake into £10,240; twenty turns it into millions. No bankroll — and no bookmaker's maximum bet — survives that.
The honest takeaway. A £1,000 bankroll staking £10 was wiped out after just 16 bets. Martingale doesn't change your expected loss — it converts a steady trickle of small wins into one inevitable, catastrophic one. It's the most reliable way to turn a small edge against you into a total loss.
More experiments coming
Next up: is each-way betting worth it, and do favourites actually win — each tested the same honest way. Try the betting calculators to run your own numbers in the meantime.