Once a week we open the books on the prediction models, and the last seven days cover a decent sample: 903 races between July 10 and July 16.
The headline numbers first. The model's first pick won 220 of those races, a strike rate of 24.4 percent, and hit the places at 44.2 percent. Cast the net across all three rated picks and at least one of them won 65.7 percent of the time, with the top three covering the frame in 88.4 percent of races. Against a baseline of roughly 17 percent for picking blind in six-dog fields, the first pick is running well clear of chance, which is the first thing we check every week.
Consistency was the theme rather than fireworks. The best day was Sunday July 12, when 33 winners from 106 races pushed the daily strike rate past 31 percent. The leanest was Tuesday the 14th at 21.4 percent, and Thursday came in a shade under par at 22.7. That is a tight band. No blow-up days, no miracle days, just a model doing roughly what it says on the tin, day after day.
Where is the room for improvement? The gap between the first pick winning 24.4 percent and the top three winning 65.7 percent tells you the model is excellent at identifying the right cluster of dogs and merely good at ordering them. Ranking the front two of a race correctly is the hardest problem in this sport, because it is decided by first-bend luck as much as ability. That is where our development effort is focused. Full daily breakdowns are on the results page, updated live, and we will check back in next week.