Seven days, 830 races, and the headline numbers are holding steady. The first pick has won 209 of those 830 races at a 25.2% win rate. For context, a random selection from a typical six-runner field would produce a 16.7% win rate — the models are running about 8 percentage points above chance. The place rate for first picks sits at 43.5%, meaning the top selection finished in the first three in nearly half of all races run across the week.
Open it up to any of the top three predictions and 60.1% of races were won by a model selection. Place coverage across the top three reaches 86.7%, meaning the eventual winner was identified somewhere in the predicted top three in almost nine out of ten races.
The standout day was Sunday (May 31), which returned a 32% first-pick win rate from 97 races — comfortably above the weekly average. Thursday (May 28) was the difficult session: 23 wins from 131 races, a 17.6% rate well below the trend. The causes of a weak day tend to cluster around lightly-raced dogs with limited form data and races featuring course debutants — situations where the models work with thinner information and the form signals are harder to read. A busy Thursday card with a lot of open-race runners fits that pattern.
What's worth noting is how the any-top-3 rate held up even on the poor day. Even on May 28, 71 of 131 races (54.2%) were won by one of the top three predictions — well above the random baseline. This suggests that on difficult days, the models are usually in the right area of the form book even when the precise rank ordering misfires.
The week overall represents consistent, unremarkable performance in the best sense: no dramatic outliers in either direction, and the any-top-3 coverage in particular sitting at 86.7% is a number that reflects well on the underlying methodology.