Seven days. 883 races. 213 first-pick winners. That is a 24.1% win rate across the full week, covering everything from Monday sprints to Saturday opens across multiple tracks and grades. The any-top-3 rate over the same period is 60.7%, meaning the winner came from within the model's top three selections in almost two out of every three races. The place rate -- first-pick selections finishing in the first three -- sits at 87.8%.
The standout day in the seven-day window was May 30, where 48 winners emerged from 178 races, a first-pick win rate of 27%. That was also the largest card of the week by some distance. The model appears to perform at its best when working across a high volume of races simultaneously -- more data across the card gives the cross-race calibration more to work with, and the comparative ranking of runners becomes more accurate when there are more benchmarks in play.
The softest day was May 28, where 23 winners from 131 races produced a win rate of 17.6%. Cards with a smaller sample size produce more situations where the model is rating dogs with limited historical data against other unfamiliar runners. The confidence in those assessments is lower, and the results reflect that. A 17.6% first-pick win rate is not a failure -- it is within the normal range of variance for greyhound racing -- but it is the point in the week where the model was working with the thinnest information.
The seven-day totals show no systematic blind spots by grade or track. The any-top-3 place rate of 87.8% -- nearly nine in ten first-pick selections running in the frame -- is the most reliable indicator of model health, and at 87.8% there is nothing to flag. The core form reading is working. The variance in win rates day-to-day is the expected product of greyhound racing's inherent unpredictability rather than anything structural in the model.
For the week ahead, the baseline expectation is a first-pick win rate in the 22-26% range and any-top-3 coverage above 60%. If you see sustained days below those thresholds, it is worth noting -- but this week gave no reason for concern.