What Elite Pass Gold is actually trying to solve
Let me start by conceding the thing most racing products spend their lives pretending is not true. The odds are very good. Better than nearly everything anyone builds to beat them.
Odds answer their own question extremely well
A price is the aggregated opinion of everybody who cared enough to have one, weighted by how much they were willing to risk on it. That is a formidable object. We have tested it more ways than I would care to itemise and it keeps coming back accurate.
Split favourites by field size and the market's expectation tracks what actually happens, band by band. Split them by price instead and the same thing. Short ones win far more than long ones, big fields are harder than small ones, and in every cut the expectation and the outcome sit close together. Across 60,671 races with one clear favourite, it won 35% and did not win 65%, which is broadly what the prices said in advance.
So the question odds answer is: how likely is this, according to everyone. On that question they are close to unbeatable and I have no interest in pretending otherwise. We have looked at 227 racing systems and theories now. None of them has validated. That number is not a boast about our standards, it is a warning about the difficulty of the thing.
We are asking a different question
Not how likely. How coherent.
Before a race there is a set of separate readings of it. What the structure of the field looks like. Where the pace is, if it is anywhere. How the horses separate from one another on the evidence available. What the draw is likely to be worth given all of that. These are different lenses looking at the same event, and they were arrived at independently.
Sometimes they all point at the same thing. The structure, the separation and the pace read give you one consistent story about how this race is likely to unfold. More often they do not. The pace read says one thing, the separation says another, and the shape of the field says the question is unanswerable from here.
How strongly the pre-race picture agrees with itself is a measurable property. It is not a probability. It is not a forecast, a confidence level, a value assessment or an opinion about a price, and if a number of ours is ever read as any of those we have failed at writing rather than at research. A race can be perfectly coherent and the favourite can still get turned over, because coherence was never a claim about the result.
Why it could not just be another odds line
The obvious move, and the one I kept being pushed towards, was to produce our own probability for each horse and compare it with the market's. Find the gaps. Sell the gaps.
Two problems with that. The first is that it puts us into direct competition with the single most efficient estimator in the sport, using less information than it has. That is a fight you lose slowly and expensively.
The second is subtler and it is the one that actually settled it. A probability estimate cannot tell you when to say nothing. It always produces an answer, for every horse, in every race, whether the underlying picture supported one or not. Feed it a race where the evidence contradicts itself and it will hand back a confident-looking number built on the contradiction. The uncertainty gets laundered into precision on the way out.
A coherence measure does the opposite. When the readings disagree, that disagreement is the output. It is allowed to come back and say this race does not hold together well enough to be worth describing, which is the single most useful sentence in the whole project.
Most races say nothing, and that is the point
On one development day recently we put 57 races through the full read. Sixteen came out with a score. Forty-one got no read at all. One showed full convergence across everything we look at.
That is not a system underperforming. That is the instrument working. Most races genuinely do not have a clean enough pre-race picture to support saying anything useful about them, and any product that returns a strong opinion on all 57 is manufacturing 41 of them.
To be plain about what that day was: development, not a service. Nothing was issued to anybody, nothing was recommended, and there were no selections because Elite Pass has no validated edge and makes none. What we were testing was whether the thing can tell the difference between a race it can read and a race it cannot, honestly, before the off, without quietly lowering the bar to fill the page.
That is the whole build. Gold early access is open if you want to watch it being put together rather than wait for the tidy version, and I would rather have people in there while it is still being argued about.
Check today’s races
See how Elite Pass reads today’s eligible races.
Related research
More notes
- What two months of testing actually taught usThree of the first five beliefs we tested did not survive. Here is what lived, what died, and what changed about the question we were asking.
- How should a horse racing system actually be tested?Twelve steps, in order, for testing a racing idea properly. Written by someone who has run 227 of them and cleared none.
- Why horse racing systems look brilliant in backtestsFinding a profitable-looking racing system in historical data is easy. Finding one that survives tomorrow is not. The statistics of why.
Elite Pass publishes research, not guaranteed outcomes. Findings are measured against the market and remain subject to replication and sample size. The evidence room and the public ledger carry the full record. All insights.