Seven days of data, 970 races, 225 first-pick winners. That is a 23.2% first-pick win rate across the week — sitting within the model's expected operating range and consistent with the long-run average.
The week had a clear standout day on June 21, where 39 winners came from 120 races for a 32.5% first-pick win rate. That is a strong performance — nearly one in three top picks winning is well above average, and it typically reflects cards where fields were more consistent: established grade levels, dogs with solid local form, races where the form held up rather than being scrambled by traffic or bad luck.
June 22 ran a very different pattern. It was the week's biggest card at 152 races, but only 30 first-pick winners came from it (19.7%). Large mid-week cards often include meetings at tracks where the model's data depth is thinner, meaning it has less to lean on. The any-top-3 win rate for June 22 was still around 60%, suggesting the model was identifying the right dogs — it was just less certain about finishing order within competitive fields.
Yesterday (June 24) returned the week's lowest absolute count at 7 from 36, but it was also the lightest card by a significant margin. In percentage terms, 19.4% is slightly below average rather than anything alarming.
The numbers that reflect the model's real value over the full week are the any-top-3 figures: 60.7% win rate and 87.8% place rate. Nearly 88% of top-3 selections finishing in the top two across 970 races shows strong overall calibration. The model is consistently identifying the right dogs; the exact finishing order is where racing does what racing does.