Over the last seven days the model has covered 1,097 races, and the headline numbers hold up well: a 25.2% first-pick win rate and a 46.1% first-pick place rate. Widen it to any of the top three rated runners landing the win and that climbs to 64.7%, with 89.2% of races producing a top-three pick finishing in the frame somewhere.
Day to day there's been real variation, which is normal and worth being upfront about rather than smoothing over. July 18th was the strongest day of the week by volume, 232 races with 61 first-pick winners, a strike rate a shade over 26%. July 17th followed a similar pattern at 223 races and 60 winners. The quieter days, July 14th and 15th, saw the first-pick strike rate dip into the low 20s, 24 winners from 112 races and 29 from 143 respectively, though the any-top-3 numbers held up better on both, 74% and 60% respectively, suggesting the model was still reading these fields correctly even when the exact top selection didn't cross the line first.
That pattern, a steadier any-top-3 rate even on days when the outright top pick cools off, is usually a good sign rather than a bad one. It means the model isn't badly misreading races on quieter days, it's more that the gap between the first and second or third rated dog was tighter, and the placed money went to a near-miss rather than a total surprise.
Across the full week the numbers sit comfortably above what you'd expect from chance alone in fields of five and six runners, and the consistency of the any-top-3 figures in particular, rarely dropping below 60%, points to a model that's holding its edge through a genuinely varied run of racing rather than riding a hot streak.