Seven days of data, 1,185 races. The model's first pick won 320 times — a 27% win rate. The place rate (finishing first or second) sits at 46.2%, and across 64.8% of races one of the top three picks won outright.
The benchmark: a randomly selected dog in a six-runner field wins 16.7% of the time. At 27%, the model is generating roughly 10 percentage points of edge above random across a large sample. That gap is what matters — it shows the ratings are doing genuine work, not just surfacing short-priced favourites.
The best individual day was yesterday, June 17: 54 wins from 169 races, a 32% strike rate. The top three collectively won 70.4% of the card and placed in 89.9% of races. For anyone using the predictions as a guide to the main contenders, June 17 was a day when the form read clearly.
The toughest day was June 14: 28 wins from 119 races (23.5%). That is below the weekly average but still above the baseline random rate. Looking at the numbers, it appears to be a card that ran against the ratings rather than any systematic error — a smaller race count means a handful of unexpected results move the percentage more visibly.
The any-top-3 win rate of 64.8% and place rate of 88.8% are the most stable indicators across the week. These figures are consistent across every day in the sample — no single day falls significantly below 60% on the top-three win metric. That kind of floor is reassuring; it tells you the model is rarely off the scent entirely, even on its weaker days.
No structural changes to the model this week. The ultra-composite weighting — which blends performance ratings, field speed, suitability and first-bend data — continues to perform as the grid search validated it would.