Seven days of data covering 1,194 races gives a solid picture of how the prediction models have held up through the middle of June. The headline figure is a first-pick win rate of 26.6%: the model's top-rated dog won roughly one in every four races across the week. The place rate (top pick finishing in the top two) came in at 47.7%, meaning close to half of all first picks were at least in the frame.
The seven days are largely consistent without the kind of wild swings that undermine confidence in a rating system. Monday the 15th was the strongest single day at 28.4% from 155 races; Saturday the 14th was the softest at 23.5% from 119 races. A five-point spread between the best and worst days over a week is reasonably tight, suggesting the model is not heavily sensitive to card size or day-of-week variation. The mid-week days with the largest race volumes (246 races on the 13th, 222 on the 12th) both returned win rates above 26%, which is encouraging.
The any-top-3 metrics give the broader accuracy picture. Across 1,194 races, the model's top three picks included the winner 763 times, a rate of 63.9%. Place coverage from the top three hit 88.3%: in nearly nine out of ten races last week, at least one of the model's three selections finished in the top two positions. That is the figure that tells you the model is reading the form landscape accurately even when it gets the precise winner wrong.
The week did not reveal an obvious structural blind spot, though the Yarmouth A3 on Monday (where the predicted winner finished fourth and the actual winner was rated sixth) is a reminder that front-running rail performances are a recurring challenge for systems built on field-relative speed and historical form patterns. That is not a flaw to fix so much as an honest limitation of any pre-race rating approach.
The live accuracy tracker is updated throughout each racing day if you want to follow today's results in real time.