Guides · 29 June 2026 · 7 min read
How horse racing ratings work - and how ours are different
The ratings you already know
Every horse in Britain and Ireland carries an official rating (OR) - the handicapper's estimate of its ability, expressed in pounds of weight. Alongside it live speed figures like Topspeed, form ratings like Racing Post Ratings (RPR), and countless private handicaps. Each captures one lens on ability, and each has the same weakness: a single number can't tell you whether today's conditions suit.
A horse rated 85 might be 95 on soft ground at a left-handed track and 70 on quick ground in a big field. The rating isn't wrong; it's incomplete. The question that decides races is not 'how good is this horse' but 'how good is this horse today'.
What a modern scoring system does differently
The Racing Bot's answer is to score the whole context, not just the ability. Thirteen signals per runner, every day: recent form, trainer form and jockey form over their last runs, the horse's win record at today's course and on today's going, the trainer's and jockey's records at the course, the trainer-jockey combination, the market move since opening, Racing Post tips, named tipster support, the official rating itself, and whether the horse is rated top in the race.
Each signal earns points scaled by how strongly it leans, and the points sum to one score - typically somewhere between double digits and the mid-500s, with a season average around 270. The top score in each race becomes the pick. Crucially, every score opens up: tap any runner and you see the exact points from every signal, plus a marker on each signal the horse leads outright.
Transparent beats clever
A rating you can't interrogate is a rumour with a number attached. Transparency is what lets you learn from the system: when a pick loses, the results page shows what the scoring liked about the horse that beat it - and over hundreds of races you start seeing which signals carry the day in which conditions. It also keeps the system honest: the record (13,000+ picks in the last 12 months, roughly one winner in four at the top of the card) is public, dated and includes every loser.
If you think the weights are wrong, you can prove it: the simulator re-runs a season of results with your own weighting of every signal, and reports the win rate, place rate and yield your version would have produced. That is the real difference - not a better black box, but no box at all.