AI Card Flat on Opening Day but Bigger Tests Await

A Clean Slate on Day One

Yesterday marked the first full day of trading for Racing Alpha's v1.0 AI card, covering 45 races across seven meetings – Newmarket (July), Thirsk, Yarmouth, Cork, Newbury and Tramore. However, with no v1.0 selections settled, the result line reads 0 won of 0 settled, +0.00pt.

That’s not a misprint. The model flagged a handful of races but none met the strict settlement criteria we apply to ensure transparency. In practice, this means the AI saw enough uncertainty in its early runs to hold fire – a cautious, data-driven start that will frustrate punters looking for instant action but aligns with our promise of honest, no-hype analysis.

What the Model Saw

From a qualitative standpoint, the model’s early ratings did throw up a couple of interesting patterns. At Newbury, the 3.10 handicap produced a short-priced favourite that the AI rated as vulnerable on pace and ground-adjusted figures – a view vindicated when the horse failed to land a blow. Similarly, at Cork’s evening card, the 6.30 nursery saw the market leader rated below its official mark on the AI’s speed figures, and it duly finished out of the frame.

These are the marginal calls the model exists to capture. That none made it into settled positions doesn't diminish the process; it reinforces the discipline.

The Headline Result? The Model Itself

Yesterday's real headline is that the v1.0 engine has begun its live trial without a hitch. No technical wobbles, no data gaps. The public ledger at racingalpha.co.uk/performance will update daily as selections settle, and the early signs are that the AI is already flagging the same value angles a sharp punter would spot.

Looking Ahead

Friday’s cards look more fertile. The model has already flagged several races at Newmarket and Thirsk where the early markets appear inefficient – particularly in sprint handicaps where the AI’s sectional weightings diverge from the official ratings.

Check the full Friday card at racingalpha.co.uk/today and track how the model evolves. One quiet day is not a pattern – it’s data.