Over the last seven days of tracked results (993 races from 11 July to 17 July), the model's top pick has won 249 times — a 25.1% first-pick win rate — with a further 444 places, giving a place rate of 44.7%. Widen that to any of the top three predicted dogs and the numbers step up sharply: 637 winners came from the top three picks across the week, a 64.1% any-top-3 win rate, with 87.4% of races seeing a top-three selection finish in the frame.
Day to day, the picture has been consistent rather than spiky. Yesterday was the strongest single day of the week for raw win totals (60 first-pick winners from 223 races), but the underlying strike rate held fairly steady across the seven days — nothing that looks like a collapse in one direction or a lucky purple patch propping up the average. That consistency across a busier week (223 races on the 17th versus 106 on the 12th) is itself a reasonable signal — the model isn't just working on quiet cards with fewer moving parts.
Where the numbers are most reassuring is the gap between the first-pick rate and the any-top-3 rate. A 25.1% first-pick win rate against a 64.1% any-top-3 rate tells you the model is generally identifying the right small group of contenders even when it doesn't nail the exact order — which, given how tight greyhound racing finishes tend to be, is arguably the more useful signal for punters building shortlists rather than relying purely on the single top selection.