Thursday produced 147 races across the British and Irish cards, and the models performed well. The first-pick win rate came in at 25.9% — 38 winners from 147 selections — with a further 71 running into the first two positions, giving a first-pick place rate of 48.3%.
The any-top-3 figure is where Thursday really delivered: 98 of those 147 races had one of the model's top three predicted runners finish first, a hit rate of 66.7%. For comparison, a model with no predictive value would expect around 50% of races to see a top-3 pick win (since three out of six runners being picked means any random three have a 50% chance of containing the winner). Hitting 66.7% on 147 races is a meaningful result, not a fluke of a small sample.
It was not a day where everything came easily. The standard distance races produced the cleanest predictions, while a handful of sprint cards threw up results that the form data could not account for — interference in the first bend being the primary culprit, as it often is. Dogs correctly identified as the best in the field occasionally get shuffled out of contention before they have a chance to show it.
The place figures are worth highlighting for punters who approach the sport from an each-way perspective: nearly half of all first picks ran into the top two. Over a full card that suggests consistent value for anyone using the model as a starting point for place betting.