Why we publish the ideas that failed
Three of the first five beliefs we tested did not survive. They are still on the site, written up as carefully as the two that did.
Nobody shows you the graveyard
Every system you have ever been sold arrives fully formed and undefeated. You see the method that worked. You do not see the eleven that were tried and quietly buried on the way to it, and you are not told how many races were looked at before somebody stopped looking.
That is not necessarily dishonesty. It is how people write about their own work. Nobody puts the failures in the brochure.
The trouble is that the failures are what tell you whether the finished thing means anything. If I test twenty ideas and show you the one that worked, I have shown you a number that is almost guaranteed to exist by chance alone. If I test five and show you all five, including the three that died, you can see the shape of the search. Those are very different pieces of evidence, even when the surviving claim is word for word identical.
What died, and how
The draw went first, and it went cleanly. Badly drawn horses genuinely do underperform in the right conditions. Anyone who has watched a big field over five furlongs at a sharp track knows this and is not imagining it. So we took the draw on its own, as a way of beating the market, and measured it on held-out races the method had never seen. It returned a clean null. Not a weak result. Nothing.
Following the money went next. Prices genuinely move, and they move a long way in the last hour. We recorded our own timestamped snapshots rather than reconstructing it afterwards from memory and hope. Then we compared horses that had shortened against horses that had drifted, at the same final price. The advantage disappeared. What had looked like a signal about steamers turned out to be a fact about which price band they had ended up in.
The third was more painful, because it started out true. Strong recent form is genuinely predictive. It really does find good horses. But when we held the final price constant and asked whether form still added anything, most of what it was telling us had already been absorbed into the price by the time the stalls opened. The form was right. The price was right. That is an awkward combination if you were hoping one of them would beat the other.
What a failed idea is actually worth
More than people assume, and in three specific ways.
It closes a door. Racing has a small number of well-worn approaches and a very large number of people quietly retrying them. Knowing that the draw on its own is dead means I will not spend another six weeks on it, and neither will anyone who reads the write-up. Time is the scarce resource in this work. A confirmed null buys you some.
It usually tells you where the interesting version is. The draw did not die because it does not matter. It died because it is one of the most discussed features in British racing, so by the time a price settles, a strong draw effect is normally already in it. That failure points somewhere quite precise. Not draw. Draw with pace, position and the shape of the course, which is much harder to see and therefore much harder to price. The dead idea drew the map.
And it calibrates you. Three losses out of five is a useful thing to carry around. It makes you slower to believe the next promising pattern, which is the single most expensive habit in this sport.
The rule we set before we had anything to protect
Every claim gets settled against Betfair starting prices, and the result is published whichever way it lands. We decided that early, when there was nothing on the books to embarrass us, which is the only time it is easy to decide.
It has already cost us. Three of the five headline beliefs we put up were ideas I liked and expected to survive. Publishing them as failures is not a pose. It is genuinely annoying, and there were mornings I would rather have written something else.
But the alternative is worse. If we only showed you the survivors you would have no way of judging them, and neither would I, because a person who never records his misses eventually cannot remember having any.
A settled claim means the outcome was checked against the price that was actually available, not against how it felt at the time.
What survived, stated carefully
Two things, and both are less exciting than they sound.
Favourites are priced about as well as we can measure. Across 60,671 races with one clear favourite, it won 35% and did not win 65%, which is close to what those prices implied. Short-priced favourites win more often, almost exactly as often as their price says they will. Favourites are more vulnerable in big fields, and the market already prices that too.
Notice that neither of those is an edge. They are findings about how good the market already is. Elite Pass has no validated betting edge and does not claim one. What survived is a clearer picture of where the crowd's enormous effort is concentrated, and therefore where it is thinnest.
Favourite wins 35% · does not win 65% · n = 60,671
The uncomfortable version
If we keep testing at this rate, most of what we publish will be negative. That is not a failure of the programme. On a subject this heavily studied, by this many capable people, with this much money involved, a high null rate is what an honest process looks like from the outside.
If you ever see a run of results here where everything works, be suspicious. I will be.
Related research
- The race that changed the questionA friend's horse had everything in its favour at Newmarket. The race went somewhere else entirely, and the question I left with was not about the horse.
- Why we timestamp our own pricesA price you read after the race has already been contaminated by the result. So we write ours down before the off, and never touch them again.
- 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.
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.