Most people who work with horses can see when something is wrong. A trainer who watches the same horse every day will notice a subtle hollowing of the back, a head that carries a fraction higher than usual, a stride that shortens on one diagonal before the rider reports any change in feel. That intuition is real and it is valuable. It is also, at the level of detail that matters clinically, unreliable.

The human eye operates at roughly thirty frames per second of conscious perception and has no stable memory of what a horse looked like six months ago. A developing asymmetry can grow for weeks before it exceeds the threshold at which it becomes visible to an observer who is not looking for it specifically. By then, in many cases, the underlying tissue has already changed.

Biomechanics is the measurement layer that sits underneath intuition. This post walks through what biomechanics actually tracks, why video is enough to do it well, and what changes when you reference every reading against the horse's own history rather than against a textbook average.

What biomechanics measures

At its core, equine biomechanics describes four things:

Phase structure. How long each limb spends in stance (on the ground, bearing load) versus swing (in the air, moving forward). A healthy trot has tightly symmetric stance-swing ratios across diagonals. Asymmetry of even a few percentage points, sustained over many strides, is diagnostically interesting.
Joint kinematics. The angles through which carpal, fetlock, tarsal, and coffin joints flex and extend during each gait phase. Range of motion is individual — some horses simply move through smaller arcs than others — but the shape of the angular curve across a stride is stable for a given horse once mature.
Axial posture. Head carriage, neck curvature, and pelvic tilt through the stride. The head is an especially sensitive indicator: a horse offloading a front limb will lower its head as the sound limb contacts the ground and lift it slightly as the compromised limb contacts. This "head nod" is one of the oldest lameness signs, and it is almost always present before visible limb dysfunction.
Rhythm and beat. Particularly for the five-gait horse. Tölt is a four-beat lateral gait; when a horse slips toward pace, the beat separation between the same-side limbs shortens. When it slips toward trot, the diagonal pairing tightens. Both drifts are measurable in milliseconds before a rider or judge can hear them.

These four dimensions produce, per video, a few hundred distinct numbers. Most of them are not individually interesting. The value is in how they change over time, for this specific horse.

Why video is enough

The dominant alternative to video is wearable sensing — IMUs strapped to the horse, pressure plates on the arena floor, markers on the skin for high-frame-rate optical capture. These systems produce excellent data. They also require equipment, setup time, calibration, and cooperation from the horse. They do not scale to every session.

Video scales. A phone camera captures enough signal for the analysis above. The hard work is in the computer-vision model: detecting the horse against background, localising joint positions to sub-pixel accuracy across thousands of frames, and translating pixel motion into biomechanically meaningful quantities.

What makes this tractable now — and did not, three years ago — is two things working together. The models are better: a modern keypoint-detection network trained on a million-plus frames per week across breeds, surfaces, and lighting conditions generalises cleanly from Icelandic turf to artificial dressage sand to a thoroughbred breezing on dirt. And the compute is cheaper: a short clip is analysed in under two minutes on commodity hardware, not in a batch job overnight.

The practical consequence is that biomechanics can now be a routine check, not a special-occasion assessment. A clip after every ride produces a trajectory, not a snapshot.

The longitudinal baseline

The most important idea in how Alendis treats biomechanics is that population averages are the wrong reference.

A textbook tells you that a healthy horse at trot has approximately symmetric stance times across diagonals. Fine. But your horse may have been born with a three-percent asymmetry that has no clinical meaning for it. Measured against the textbook, it looks suspicious. Measured against itself over eighteen months of consistent three-percent asymmetry, it looks healthy — and any new drift becomes visible against its own stable floor, not against a population noise band that hides the signal.

This is the longitudinal baseline. Every horse builds, over multiple uploads, a statistical fingerprint of how it moves when it is well. New readings are compared against that fingerprint. The first upload establishes the baseline; subsequent uploads refine it; changes are flagged as deviations from the horse's own history.

There are two practical consequences of this design that matter for anyone using the platform.

First, the baseline improves with use. The first upload gives you a provisional reading. The fifth gives you a reliable one. The twentieth gives you sensitivity to drifts that were previously inside the noise band. This is the opposite of the pattern in many platforms where the first experience is the most impressive and subsequent experiences are incremental. Biomechanics is cumulative.

Second, a horse's baseline is worth protecting. Uploading a corrupted or atypical session — a lame horse under medication, a horse in an unusual arena, a rider who is compensating for something — shifts the baseline in ways that reduce future sensitivity. Good hygiene, which the platform enforces by asking about context on upload, is: clean sessions for baseline, everything else tagged and held out of the reference set.

What the measurements cannot tell you

Biomechanics measures movement, not cause. Alendis can show you that your horse is offloading its right fore by four percent more than baseline; it cannot tell you whether that is a suspensory issue, a trimming imbalance, a shoe coming loose, or a stone wedged in the frog. That is a veterinary question, and the right next step when the platform surfaces a finding is a conversation with your vet — with the biomechanics report attached.

This framing matters. Alendis is not trying to replace veterinary diagnosis. It is trying to make the diagnostic conversation start earlier, on firmer ground, with objective evidence that a change has happened. The vet's judgement is where cause attribution lives. The platform's job is to make the symptom visible before it becomes obvious.

Where to go from here

If you are a rider, the simplest path is a clip after your next session. Record thirty seconds of trot on a straight line, with the horse moving directly toward or away from the camera. Upload it. Read what comes back. Do it again after your next ride.

If you are a trainer with multiple horses, the pattern extends: build baselines for every horse in the barn, check them weekly, watch for drifts. The cross-horse view shows you which horses are stable, which are improving, and which deserve a closer look.

If you are a veterinarian, the platform is available as a clinical adjunct — longitudinal objective data to bring into differentials, a second view alongside physical exam and imaging. Reports export as shareable links; raw metrics are available to practices running research programmes.

Biomechanics is not new. What is new is that it fits in your pocket, produces data every session rather than every six months, and references that data against the one benchmark that actually matters: the horse itself.