Over the last seven days the model has covered 1,105 races, with the top-rated selection winning 25.3% of them and placing in 47.6%. Widen that out to any of the top three picks and the numbers climb to a 65.5% win rate and an 89.4% place rate, meaning the correct winner is inside the model's top three picks roughly two times out of three, and one of them finishes in the frame in almost nine races out of ten.
Day by day it's been a bit of a mixed bag rather than a flat line, which is normal and honestly a healthier sign than suspiciously consistent numbers would be. July 19th was the strongest single day, the top pick landing 30.8% of its 120 races, while July 14th was the softest, down at 21.4% from 112 races. The two most recent days, July 19th and 20th, have both come in above the weekly average, which is a reasonable sign the model's been reading conditions well over the busier midweek cards rather than just catching a lucky run on a quiet night.
There's no obvious pattern of specific blind spots in this stretch, no single track or grade dragging the numbers down consistently, more a normal week's spread of races where form held and races where it didn't. That kind of variance around a solid baseline is exactly what you'd expect from a model working off real data rather than one either overfitting to recent results or missing something structural. The full breakdown for every race is always available for anyone who wants to check the model's working for themselves.