StableBet
The Lab · Methodology

How we test betting systems

Every verdict in the Lab comes from the same honest accounting. Here is exactly how we run it, including the bug we caught that turned every “profitable” system back into a loser. If the working does not hold up, neither does the conclusion, so we show it.

The data

We measure each system on 27,909 real British races run between October 2023 and June 2026, settled at the industry Starting Price (SP), the price your bet is actually paid out at. Every runner in those races, with its SP and its finishing position, is in the dataset. No sampling, no cherry-picked window, no “since our good run started”. The same races judge every system, so the comparison is the system, not the sample.

How we settle a bet

This is the part most “systems that win” get wrong, and it is where we caught ourselves out too. A bet is settled the way your wallet settles it:

  • A horse that falls, is pulled up or unseats its rider is a losing bet.You staked your money and it did not win, so you lost it. An early version of our own table quietly dropped those runners, which flattered every favourite, and jumps favourites most of all, because so many of them fall. Counting them honestly is what moved “back the jumps favourite” from a fake +4.6% to its true −14.4%.
  • A genuine non-runner is voided, with the stake returned, because no bet was struck. We separate these from fallers using the official result status, so a withdrawn horse is not punished and a faller is not let off.
  • Joint-favourites split the stake. When two horses share the shortest price we back both for half a stake each, so no result can leak in from picking the one that happened to win.

The bookmaker's margin

Add up the chances implied by every price in a race and they come to more than 100%. That extra is the over-round, the bookmaker's built-in cut, and it runs at roughly 12% per race across British racing, climbing towards 30% in a big field of 16 or more. You pay a slice of it on every bet, win or lose. On top of that sits the favourite-longshot bias: the market overprices big outsiders and underprices short ones, so the longer the price you back, the worse the value. That is why backing the favourite leaks least and backing the outsider leaks most, and why neither makes money.

Why no system wins

Across the twenty systems we have tested, not one returns a profit. The least-bad is backing odds-on favourites at about −7%, the gentlest version of the same leak. Backing the favourite in every race loses about 12.5%. Accumulators, by stacking the margin on every leg, lose 40% and up. None of this is bad luck or a small sample, it is the over-round doing exactly what it is designed to do. A staking plan changes how the loss arrives, not whether it arrives.

What we do not claim

Our own AI model is a part of this, judged by the same rules. Backing its top pick blind loses about 11%, much like the favourite. It is a genuinely accurate, calibrated read on a race, which is useful for understanding the shape of a card, but it does not beat the Starting Price, it is nota value detector, and we will never sell you its picks. We would rather show you this page. You can follow the model's day-to-day results on the track record and read how it is built in the model write-up.

See it in action: every system ranked, with the numbers and the working, in Pascal vs the Professor.