# Our Model Disagrees With the Favourite in 26 of Today's 27 Races

_Across Goodwood, Beverley, Cork and Yarmouth our model's top pick is not the favourite in 26 of 27 races. What that means, and why it is not a signal._

*Source: [https://stablebet.co.uk/racing-news/model-vs-market-30-august-2026/](https://stablebet.co.uk/racing-news/model-vs-market-30-august-2026/) · last updated 2026-08-30*

## The model backs a different horse to the market in 26 of 27 races today

Sunday's card covers four meetings: Goodwood, Beverley, Cork and Yarmouth. Twenty-seven races in total.

In twenty-six of them, our model's top-rated runner is not the market favourite.

That sounds like a headline. It mostly is not, and the reason why is more useful than the number.

Before the reason, a caveat that matters. Eight of those 27 races have a price snapshot we do not trust: the quoted odds sum to less than a complete book, which no real market does, so the snapshot is stale or missing runners. Restricting the count to the 19 races where the book is coherent, the model's top pick differs from the favourite in 18 of 19. The pattern is the same either way, which is why we are still writing about it, but 18 of 19 is the figure with evidence behind it.

Our [AI Race Predictor](/ai-race-predictor/) estimates every runner's chance from the form data alone: career runs, recent finishing positions, course and distance records, official ratings. It never sees the odds while it is doing that. Only afterwards do we line its numbers up against the market's.

When two forecasters work from different information and neither is confident, they will disagree about who is fractionally in front most of the time. That is what today looks like. In the [Goodwood Group 3](/racing-news/goodwood-prestige-fillies-2026-preview/) the model spreads nine fillies between 14.8% and 9%. In the [Beverley Listed sprint](/racing-news/beverley-bullet-2026-preview/) it spreads eight runners between 19.5% and 9.6%. When the gaps between horses are that small, the top of the list changes on very little, and "the model disagrees with the favourite" becomes close to the default outcome rather than a finding.

The days worth paying attention to are the ones where the model and the market disagree by a wide margin on a single horse, not the ones where they shuffle the order of a bunched field.

Today has one of those, in the 14:50 maiden fillies' stakes at Beverley. The market makes **Eleven Eighty Two** a 24.3% chance. The model makes her 5.7%. That 18.6-point gap on a favourite is the widest on the card among the races we can price honestly, and it comes in a 13-runner Class 4 maiden on heavy ground, which is close to the least predictable shape a race can take.

## What a disagreement is actually worth

There is a temptation to read "the model likes a 10/1 shot the market has missed" as a tip. It is not one, and the reason is worth understanding because it applies to every tipping service you will ever read.

A probability estimate is not a prediction of the result. When our model says Argentine Tango has a 19.5% chance in the Beverley sprint, it is not saying he wins. It is saying that if this race were run a hundred times under these conditions, it expects him to win about twenty of them. That is a number describing a horse that loses four times out of five.

That distinction is where most betting confidence goes wrong. A disagreement about probability gets read as a disagreement about outcome, and a horse rated "better than the market thinks" gets backed as though it were likely to win.

### The margin is the part that actually costs you

Here is a number from today's card that matters more than any selection.

The bookmaker margin on the Goodwood Group 3 is about 19%. On the Cork Listed sprint later the same afternoon, it is about 4%.

That is the same sport, the same afternoon, and roughly a fivefold difference in the cost of placing a bet. The margin is the amount by which the prices on offer add up to more than a fair book, and it comes out of punters' returns whichever horse wins.

A 19% margin means the prices in that race are, collectively, well short of fair value. You can pick the winner and still be playing a game that is heavily tilted. Over a season the margin does more damage to a betting balance than any individual selection decision.

We wrote up how this works in the [betting margin explainer](/ai-race-predictor/lab/understand-odds/). It is the single most useful thing on this site for anyone who bets regularly, and it has nothing to do with picking horses.

## The model's actual record, which is a losing one

Any site publishing model output owes you its results. Here are ours, taken from the [live track record](/ai-race-predictor/track-record/) rather than from memory.

Across **9,111 races**, backing the model's top-rated pick with a flat £10 win stake at industry starting price returns a **loss of 13.7%**. The strike rate is 24.6%, so roughly one pick in four wins. Over the last 30 days the picture is worse, not better: a loss of 20.7% from 1,151 races.

On forecasting accuracy the model scores a Brier of 0.100. The market scores 0.092. Lower is better, so **the market is the more accurate forecaster**, and by a clear margin on this measure.

That is the honest position, and we publish it because a model that quietly stopped reporting its losses would be worthless as a second opinion.

### So why run it at all

Because losing to the market is the finding, not the failure.

A calibrated model that cannot beat starting prices is direct evidence of how efficient racing markets are. The prices you see already contain almost everything the public form can tell you, which is why systems built on public form do not make money. Our [betting systems research](/ai-race-predictor/lab/strategies/) tests that repeatedly, and none of the systems we have tested turns a profit over a serious sample.

The model earns its keep in two narrower ways. It gives a second read on races where the market's confidence looks thin, which is most of today's card. And it makes the cost of betting visible, race by race, by showing what a fair book would look like next to the one actually on offer.

Neither of those is a route to profit. Both are more useful than a tip.

If you take one thing from today's eighteen verified disagreements, make it this: two independent estimates of the same race landed in different places, and the one with the better long-run accuracy record is still the bookmaker's.

> **Responsible betting reminder:** Set a budget before you bet and stop when you reach it. Free, confidential help: [BeGambleAware.org](https://www.begambleaware.org) or the National Gambling Helpline on 0808 8020 133.