Seven days, 696 races, and the model's first pick won 167 of them for a 24.0% win rate. That's a touch below the long-run target of 25%, but within the normal range of weekly variance. The place rate for first picks was stronger at 45.7%, which means nearly half the time the top selection finished in the first two. That's where the model continues to earn its keep for each-way punters.
The standout day was Saturday, where 33 winners from 113 races gave a 29.2% first-pick win rate. Friday was the busiest card of the week with 254 races, and the model found 61 winners (24.0%) with 118 places (46.5%). Wednesday was tough: just 2 winners from 12 races (16.7%), though the tiny sample makes that hard to read much into.
Broadening to the top three selections, at least one of them won in 63.4% of races and at least one placed in 89.2% of races across the full week. That 89.2% place coverage is particularly useful for anyone running place-based systems, because it means nine times out of ten, the model is putting at least one of its top three into the frame.
Second picks won 154 races (22.1%) and third picks won 120 (17.2%), showing a clear and logical drop-off in predictive power as you move down the rankings. The gap between first and second pick isn't enormous, which suggests the model is good at identifying a group of contenders rather than always pinpointing the exact winner. That's a realistic outcome for any ratings-based approach in a sport where first-bend trouble can derail the best dog in the field.
Overall, a steady week. The model is doing its job.