Seven days of racing, 1,260 races, and a first-pick win rate of 23.8%. That is the week in a single number, and it is worth unpacking what it actually means.
The baseline expectation in a six-runner race is 16.7% — one in six — for a model with no predictive ability at all. A first-pick win rate of 23.8% across 1,260 races means the model's top selection is winning roughly 40% more often than pure chance would predict. The any-top-3 win rate of 61% is equally telling: three out of five races saw one of the model's top three picks win, against a theoretical baseline of 50%.
The week had genuine variation. June 6th was the strongest day by volume — 63 first-pick winners from 253 races, a 24.9% rate on the busiest card of the seven days. June 8th was the toughest: 30 first-pick wins from 160 races, an 18.8% rate that dipped below the model's usual output. Looking at the run comments from that day, interference in the early strides of races was a recurring theme — dogs checked or baulked in the first bend, removing their ability to run to form before the race had properly started. That is a variable the model cannot predict, and when it clusters on a single day it will always drag the numbers down.
The place rates tell a quieter but useful story. The first-pick place rate (into the top two) across the week was 44.8%, and the any-top-3-pick place rate was 84.8%. Those figures have been consistent across all seven days, even on June 8th when the win numbers dipped. This suggests the model is correctly identifying the competitive structure of races — which dogs belong near the front — even when the precise outcome is disrupted by on-track incidents. For punters approaching greyhound racing through each-way or place markets, those place rates point to consistent underlying value.