Originally published February 2026. Updated July 2026.
Soft ground costs speed. The question is how going affects racehorse stride — is it stride length, stride frequency, or both? We analysed stride profiles across 100 horses on different going conditions to find out which.
The Dataset
All 100 horses were selected from truly run races, filtered for peak performance using time ratings. Without those filters, the noise in stride data drowns out the signal.
Each horse has stride measurements from two runs on different going, spanning between one and four steps on the UK going scale. The dataset covers 35 sprinters (5f–7f), 31 milers (7.1f–9.9f), and 34 middle-distance/stayers (10f+). Where track configuration could distort a comparison — different gradients, bends versus straights — we flag it in the data rather than correct for it.
Stride Shortens. Rhythm Holds.
The speed loss comes almost entirely from one place: stride length.
Across the dataset, stride length drops by an average of 1.2% per single step of going change, rising to 3.9% for a two-step change and 6.6% for three steps. Stride frequency — how fast the legs are turning over — barely moves regardless of how much the going changes.
The mechanism is intuitive. Softer ground absorbs energy on every footfall, so each stride covers less distance. But the legs keep turning at roughly the same rate. Stride frequency was more stable than stride length in 79% of horses. Rhythm is the constant. Stride length is what gives.
Because speed is stride length multiplied by frequency, and frequency holds steady, the speed loss tracks closely with the stride loss. A one-step going change costs around 1.7% of speed. Two steps costs 4.3%. Three steps costs 7.6%. Those numbers match the slower finishing times punters see on soft ground — this is the mechanical explanation for why.
| Going Change | Avg Stride Length Loss | Avg Speed Loss |
|---|---|---|
| 1 step (e.g. good to firm → good) | -1.2% | -1.7% |
| 2 steps (e.g. good → soft) | -3.9% | -4.3% |
| 3 steps (e.g. good to firm → soft) | -6.6% | -7.6% |
At one step of going change, a third of horses show no measurable stride loss. By two steps, 90% lose stride. At three steps, every horse in the dataset lost stride length. The bigger the going shift, the more certain the effect.
How Going Affects Racehorse Stride by Distance Type
The overall pattern holds across all three distance types. But the degree of exposure differs.
| Distance Type | Avg Stride Length Loss | Avg Speed Loss |
|---|---|---|
| Sprinters (5f–7f) | -3.6% | -4.2% |
| Milers (7.1f–9.9f) | -2.3% | -2.6% |
| Stayers (10f+) | -3.4% | -4.0% |
These averages combine all going step sizes — one-step through three-step changes — within each distance category.
Sprinters are the most exposed. They run with faster leg turnover and longer strides than milers or stayers in our dataset — 7.7m average versus 7.3m for both other groups. They’re already at their mechanical limits. Soft ground takes away stride length and there’s less margin to absorb it.
Their rhythm responses are unpredictable too. Some horses speed up their legs trying to compensate. Others slow down. Unlike stayers, there’s no consistent pattern.
Stayers respond more predictably. Their leg turnover is disciplined and repeatable — 38% of stayers held their rhythm perfectly stable on softer ground. The going shortens their stride, but it doesn’t disrupt the consistency of their leg speed.
Milers appear most resilient, losing less stride than either sprinters or stayers. The honest answer is we don’t know why yet. At 31 horses the pattern is clear, but the explanation isn’t. Worth noting. Not worth building conclusions on.
What This Means for Prediction
The pattern is consistent enough to have practical consequences for how the models are built.
StridePredictor’s models use stride biomechanics to predict each horse’s optimal racing distance. When we first tried to improve predictions by standardising for going — adjusting every horse’s stride data back to an equivalent good-ground baseline — accuracy dropped. The models got worse, not better.
This study explains why. The models already see the effect of going through stride length. When a horse runs on softer ground and its stride shortens, that shorter stride enters the model directly. Applying a blanket correction on top risks removing information the model is already using.
But the research revealed something more important.
The relationship between stride and distance isn’t identical across all going conditions. Good and Good to Firm show the expected difference in stride length, but the relationship between stride and distance remains consistent enough for models trained across those conditions to predict accurately. As the ground becomes softer, that relationship begins to change. On Soft ground, stride compression becomes substantial enough that a model calibrated on better ground can start to interpret the shorter stride as evidence of greater stamina — when some of the effect is actually the surface.
The answer isn’t to correct every horse back to a theoretical good-ground baseline.
It’s to use the right model for the conditions in which the stride was produced.
Where sufficient data exists at a specific track, models are calibrated to account for that venue’s geometry — gradients, bends, camber and configuration. Where the track is neutral, or where the going itself is the more important influence on the stride signal, models calibrated to the going condition provide the better context.
Two competing environmental influences can shape a stride profile: track geometry and going. The prediction approach needs to account for which matters more.
On a course with pronounced geometry like Epsom, the track is the dominant influence. On a neutral course on Soft ground, the going matters more. The right model for each situation reads the stride in the right context.
Predictions are most reliable from stride data captured on Good to Firm through Good to Soft ground. Soft ground can still be modelled, but the risk of distortion increases as the surface becomes more testing. Predictions from Soft conditions should be interpreted with greater care, particularly where there is no corroborating stride data from better ground.
The model needs to understand the conditions that produced the stride before it can interpret what the stride says about the horse.
Related Articles
Epsom Derby 2026 Predictions – Full stride analysis of current Derby contenders
Racehorse Stride Biomechanics Explained – How stride predicts optimal distance
Epsom Oaks 2026 Predictions – Stride analysis for the fillies’ Classic
The 7-Furlong Breakpoint – Why some horses develop speed while others develop stamina