Timeform, the Racing Post and us: what we measured

We use the professional forecasts every day and we measured ourselves against them. The rating beats the Timeform forecast on held-out racing; the market beats everyone; the blend of all three lands the most forecasts. All of it is published.

We are not here to tell you the professionals are wrong. We capture the Timeform card and the Racing Post tissue every day, we show both on every race page, and we blend them into our own forecast because we measured that the blend is better than any one of us. What we can offer is something neither of them publishes: a measured comparison, held out, with the losing numbers included.

The test

Take every race from 2026 that the model had not seen in training, 8,012 races and 44,830 runners. Give every rating system its fair probability for each dog: ours from the model, the Timeform forecast and the market from their prices with the overround removed, the GreyhoundStats rating through a fitted mapping. Then score each system two ways. Log-loss, which rewards putting high chances on winners and punishes confidence on losers. And the plain one: how often the top-rated dog won.

The result

Two things to read there, and we insist on both.

The rating beats the published expert forecast. The gap between 1.610 and 1.657 is significant: the 95% interval on the difference runs from 0.038 to 0.056, nowhere near zero. The top pick winning 30.5% against 27.7% is the same fact in plainer clothes. That is a true claim and we make it.

The market beats all of us. The Betfair closing price is a better probability than any rating, ours included, and by a distance. We say this everywhere, because a rating service that hides it is asking you to bet against a number that already contains everything it knows. There is no pre-off edge in our model against the closing price. We tested that a dozen ways over the summer and killed it, and the tests are on the methodology page.

So what is the point of a rating that the market beats

Three things, and they are the product.

A second opinion that is independent. Our model is trained on official times and results, not on Timeform or the market. When it disagrees with the forecasts, that disagreement is information. The selection classes on the site are exactly those disagreements, named and recorded: when our model and Timeform's capture model both make a dog ten points likelier than the price, that is Synergy, and its record is published, wins and losses.

A better forecast than any one source. On 4,466 settled races since 1 August, a straight forecast from our model alone landed 11.0% of the time, from the Timeform forecast alone 9.5%, from the Racing Post alone 9.0%, and from an equal blend of the three 11.7%. Reverse forecasts and trios follow the same order. The blend is what the race page shows, with the measurement in the tooltip.

A record you can check. The forward test has run every racing day since 10 June 2026, 89 days and 59,911 settled bets so far, at the Betfair closing price, with nothing edited after the fact. When a class runs cold, the page shows it cold.

Where the professionals are still ahead

Timeform's running comments are better than anything we generate, and we show them. The Racing Post's daily comment on a dog is on our dog pages because it is worth reading. Neither publishes a calibrated chance or a held-out record, and that is the gap we fill, not a claim to have replaced them.