How we analyze your form.
What we measure, how we grade it, and — just as important — what we can't see from a side-view video. It's geometry, not a black box, and we'd rather tell you a metric is uncertain than guess.
The pipeline
1 · You film a clip
A 10–20 second side-view phone video — treadmill or open road. Full body in frame, camera roughly hip height, as steady as you can manage.
2 · Pose estimation
MediaPipe pose estimation tracks 33 body landmarks (nose, shoulders, hips, knees, ankles, …) in every single frame of your video.
3 · Geometry, not guesswork
From those landmarks we compute joint angles and positions across your stride or pedal cycles: lean angle, knee flexion, hip height, foot position at contact.
4 · Graded against published ranges
Each metric is compared to ranges from running biomechanics literature. The result is a grade per category plus the exact measured value — never just a score.
The grading scale
Every category gets a letter grade backed by the actual measured value. When the video doesn't give us enough confident frames to measure something, the grade is “-” — unknown means we'd rather not guess.
Good
Inside the healthy range. Nothing to fix here — keep doing what you're doing.
Fair
Slightly outside the healthy range. Worth attention, usually fixable with a drill or two.
Needs work
Clearly outside the range and likely costing you efficiency or loading tissue it shouldn't.
Poor
Far outside the range — the kind of pattern that's strongly associated with injury risk.
The 7 running metrics
Each card shows the healthy range we grade against and exactly how the measurement is taken.
Posture & Lean
3–12° forward leanHow forward (or backward) you lean from the ankles while running.
We measure the angle from your mid-hip to your nose against vertical, frame by frame. A healthy running lean sits between 3° and 12° forward. Negative values mean you're leaning back (braking with every step); >15° usually means you're hinged at the waist.
Deep diveOverstriding
<20% of frames landing ahead of hipsLanding with your foot ahead of your hips creates a braking force every step.
Each frame, we compare the horizontal position of your lead foot to your mid-hip at the moment of foot contact. If the foot lands more than a few centimeters ahead of the hip in more than 20% of frames, you're overstriding.
Deep diveVertical Oscillation
<6 cm bounce per strideHow much your body bounces up and down each step.
We track your hip position frame by frame and measure the difference between peak and trough across a stride cycle. The number is reported in centimeters at your running pace.
Deep diveHip Drop
<2% left-right asymmetryThe non-stance hip dropping below level — a sign of weak glute medius.
We measure the angle between your left and right hip across each stance phase. Any asymmetry above ~2% over a sustained portion of the video signals weak lateral hip stability.
Deep diveArm Swing
>85% symmetrySymmetric arm swing balances your stride and helps drive cadence.
We track wrist X-position over time for both arms and compute the symmetry percentage of their swing range. Above 85% is good; below 70% means there's a noticeable imbalance.
Deep diveKnee Mechanics
140–165° average knee flexionKnee flexion at stance — too straight means braking, too bent means quad-dominant.
We compute the angle formed by hip → knee → ankle across the whole video and average it. We also flag asymmetries between left and right knees.
Deep diveFoot Strike
No fixed target — context for overstridingHeel, midfoot, or forefoot — what part of your foot hits the ground first.
At each foot contact, we compare the relative vertical positions of your heel, ankle, and toe landmarks. Whichever is lowest at contact wins. We report the dominant strike type plus a breakdown across the video.
Deep diveBike fit metrics
The same pipeline applied to a side-view pedaling clip — graded against established bike-fit angle ranges.
Saddle Height
30–40° of knee bend at the bottom (knee extension angle)Your knee extension angle at the bottom of the stroke — the single most important fit measurement.
We track your near-side ankle to find the bottom of each pedal stroke, measure the hip-knee-ankle angle there, and report the extension angle (180° − that) as the median across strokes. The target is ~30-40° of bend.
Deep diveKnee at Top
~105–115° knee angle at the top of the strokeHow tightly your knee compresses at the top of the pedal stroke.
We find the top of each stroke from the ankle trajectory and measure the hip-knee-ankle angle there (median across strokes). ~105-115° is typical.
Deep diveHip Angle
~45–55° torso-to-thigh angle (endurance)How open the angle between your torso and thigh is — the balance of power, comfort, and aero.
We measure the shoulder-hip-knee angle and take the median across the stroke. ~45-55° suits most endurance positions.
Deep diveBack Angle
~40–55° back angle from horizontal (endurance)How flat or upright your back is relative to horizontal.
We measure the angle of the hip-to-shoulder line against horizontal (median across the clip). ~40-55° is a comfortable endurance range.
Deep diveElbow Angle
~150–165° elbow angle (slight bend)Whether your arms are softly bent (good) or locked out (harsh and twitchy).
We measure the shoulder-elbow-wrist angle (median across the clip). ~150-165° is a soft, slightly flexed arm.
Deep diveHip Rock
<2 cm of vertical pelvis movementHow much your pelvis rocks side to side (vertically) on the saddle — the classic too-high tell.
We track your near-side hip vertically across the stroke (detrended to ignore camera motion) and report the peak-to-peak movement in centimeters. Under ~2 cm is stable.
Deep diveFilming requirements — and what happens when they're not met
Side view, hip height, full body in frame. Every metric assumes the camera sees you from the side. A steady camera (or a treadmill clip) gives the cleanest landmark tracking.
We detect off-axis clips instead of silently grading them. The analyzer checks shoulder separation and depth spread across the landmarks; if the geometry says the camera was in front of or behind you, the report carries a warning banner and we tell you which numbers not to trust.
Low-confidence clips are flagged, not hidden. Too few clean strides, partial body tracking, or heavy occlusion lowers the capture confidence — the report says so and asks for a re-film rather than handing you confident-looking numbers.
What we don't claim
A single 2D side view has hard physical limits. We'd rather you know them:
- —No pronation assessment. Foot pronation/eversion happens in the frontal plane — it is physically invisible from the side. Apps that grade pronation from a side view are guessing; we don't.
- —Frontal-plane issues are inferred, not directly measured. Hip drop, for example, is estimated from left/right hip landmark asymmetry — useful signal, but a front/rear camera would measure it directly.
- —This is training guidance, not a clinical gait analysis. Grades come from published biomechanics ranges, not a medical assessment. If you're injured, see a physio — and bring your report along.
- —Video quality moves the numbers. Lighting, distance, frame rate, and camera shake all affect landmark tracking. The confidence signals exist precisely so you know when that happened.
See it on your own stride.
First analysis is free — film a side-view clip and get your report in minutes.