What a speed figure is
Raw finishing times don't compare across races. A win in 1:10 flat on a fast, sealed surface is a very different performance from 1:10 at another track on a tiring one, and neither one compares to a race run over a different distance. A speed figure strips those differences out so a single number means the same thing everywhere. The higher the figure, the faster the horse ran, regardless of where, when or how far.
How we build it
Our figures are velocity-based and self-calibrated. We don't borrow constants from any older figure system. Everything comes from our own database of North American races. The outline:
- Velocity, not raw time. Every performance becomes a speed (distance ÷ adjusted time) and gets compared to a standard velocity curve fitted across all of our races at that distance and surface. Normalising each distance by its own spread is what puts a sprint figure and a route figure on one scale.
- Every runner earns a figure, winner or not. Beaten lengths convert to time using the actual speed of the race at that point (Equibase's 8.167 ft per length). A horse beaten two lengths is rated on how fast it ran, not on where it happened to finish.
- Class-adjusted par times. We compute a baseline "par" for each combination of track, surface, distance and class level. Thin samples get blended toward broader baselines, so a lightly-raced cell doesn't throw off a noisy par.
- Track-day variant. This is the single biggest correction. Some days a track plays fast or slow for everyone. We measure that day's bias separately for sprints and routes, reading it partly from how the card ran as a whole and partly from the expected time given the known ability of the horses in each race. Whatever is left over is the track's speed that day, and we subtract it.
The scale
Figures run on a 0-140 integer scale. Two well-populated class levels anchor it, which keeps the numbers meaningful over time:
- ~52 is a bottom-level claimer.
- ~110 is an open, Grade 1 performance.
- Most winners of ordinary races land somewhere between the two.
Confidence tiers
Not every figure deserves equal trust, so we flag each one rather than bury the uncertainty:
- A, high. A well-sampled par, a full field, a fast or firm surface, and enough same-type races that day to measure the variant cleanly.
- B, medium. Sound, with one or two of those conditions weaker.
- C, low. Read it as indicative only. An off track, a short field, a thin par cell, an uncontested wire-to-wire winner, or a card with too few races to pin down the variant.
When the underlying time is missing or garbled, or the distance won't sit on the scale, we show no figure instead of guessing.
Pace figures
About a third of races record split times. For those, we also build pace figures that describe how the race was run: fast early and hanging on, or slow early and finishing hard. The speed figure tells you how fast overall. The pace figure tells you the shape of the effort. Races without splits are marked as such and carry a speed figure only.
What we claim, and what we don't
We're honest about the limits. No external "true" figure exists to calibrate against, so every check we run is internal. Do better races produce higher figures? Does the same horse vary a credible amount from start to start? Do the variants average out to zero over a season? And the real test: does a figure help predict a horse's next start?
On that basis we say our figures are internally coherent and predictive. We do not say they match any other publisher's numbers. A TurfAtlas 90 is not "the same as" a Beyer 90 or any other branded figure, and was never meant to be. It's a TurfAtlas figure, built end to end from our own data.