Most tipping services will give you a horse at any price. We do the opposite with one band in particular: if our pick is sent off between 4/1 and 7/1, the staking gate throws it out, no matter how high it scores. We call it the dead zone, and here is why it exists.
You may have seen it on the Edge Board: a horse scoring 100 out of 100, clearly the model's top-rated runner in its race, carrying a small note that it is not a staked pick because its price sits in the dead zone. That looks odd until you see the backtest behind it.
When we group the model's historical picks by their starting price, almost every band is somewhere between break-even and strongly profitable. The 4/1 to 7/1 band is the glaring exception, a deep trough sitting between two profitable neighbours.
| Pick's starting price | Bets | ROI at SP |
|---|---|---|
| Odds-on (evens or shorter) | 615 | +0.8% |
| 2/1 to 4/1 | 509 | +3.0% |
| 4/1 to 7/1 (the dead zone) | 310 | -21.8% |
| 7/1 to 14/1 | 295 | +11.7% |
| Bigger than 14/1 | 137 | +68.6% |
A trough like that, with profit on both sides, is exactly the shape that makes you suspicious of a fluke. So we did not act on it once. We re-validated it on a walk-forward basis, training on older data and testing forward on data the rule had never seen, and the hole stayed put. That is the difference between a quirk and a signal.
Two things meet in the 4/1 to 7/1 band. At short prices the model and the market broadly agree, so there is little to be gained or lost. At big prices our edge comes from spotting horses the crowd has written off, and the prices are generous enough to pay when we are right. The 4/1 to 7/1 middle is the awkward in-between: short enough that the market has taken our horse seriously, big enough that it loses often, and priced too tightly to make the wins pay for the losses. It is the one place our pick type has historically had no edge at all.
To be clear about what this is and is not: this is a property of our model's selections, not a universal law that every 5/1 shot loses. Across the whole market, 4/1 to 7/1 is just one more point on the broader favourite-longshot bias curve. The dead zone is the specific spot where that bias bites hardest on the kind of horse our model likes to back.
Nothing is hidden. A dead-zone horse still appears on the Edge Board as a research signal, because the model genuinely rates it, and it carries a plain note explaining why it is not a staked pick. The staked board simply skips it. On the day we first wrote this up, the example was a 100-rated horse at 5/1 that the gate filtered out. It finished fourth of five and lost. One race proves nothing, but it is the kind of result the rule exists to avoid over the long run.
Because the most valuable thing a model can tell you is when not to bet. Any service can shout about its winners. Showing you the exact price band we refuse to touch, and the losing record that earned it that status, is the clearest proof we are following the data rather than chasing the most exciting-looking horse on the card.
Discipline is unglamorous. Passing on a 100-score horse because of its price feels wrong in the moment. Over hundreds of bets, it is the difference between a model that holds an edge and one that hands it back.
The Overlay logs every selection before the off, settles in public, and tells you when to sit a race out. The Edge Board surfaces the high-conviction picks the staking rules deliberately skip, dead zone and all. Free value board, bet calculator and a play-money tracker.
Explore The OverlayRelated reading: The favourite-longshot bias · Do favourites win? · all studies.
Figures are the model's historical NAP-equivalent selections grouped by starting price, level stakes, return measured at SP. The 4/1 to 7/1 band covers decimal prices above 5.0 and below 8.0, the exact range the live staking gate excludes. The rule was confirmed walk-forward (trained on earlier data, tested forward on unseen data) so the trough is not curve-fitting. These are our model's picks, not all runners. Historical and for information only.