Seven days, 1,171 races, and the headline number is 25.1% for first-pick wins. With six dogs in a race, a random selection would produce 16.7% — so the model is identifying the winner at a rate roughly 50% above chance. That is the baseline against which everything else should be measured.
The week had clear variation. Tuesday June 17 was the standout with 54 wins from 169 races — a 31.9% rate — a day when the predictions ran hot across a full card. At the other end, Monday (June 22) came in at 19.7% from 152 races, and June 19 produced 42 wins from 219 races (19.2%). Both of those lower-scoring days coincided with larger-than-usual race cards, which tends to bring in more variable races at the margins and naturally puts pressure on the strike rate.
The any-top-3 figures are probably the best indicator of underlying model health: include the second and third picks alongside the top selection, and those three dogs won 62.4% of 1,171 races and placed in 88.3% of them. The placing rate — nearly nine races in ten where at least one of the top three picks finished in the top two — suggests the model is consistently identifying the right dogs even on days when the exact winner proves harder to nail.
One note worth flagging: June 19 was affected by a data pipeline issue that has since been corrected. The numbers from June 20 onwards should be read as a cleaner picture of where the model currently sits. The three days from June 20 to 22 average out to 24.1% for first picks, which is solid if short of the week's best.