What we measure — and what we refuse to print

Most form apps are graded on what they claim. This one is worth more for what it declines to claim: if a single side-on camera cannot see it, we do not put a number on it.
Get your scorecard
Running: thirteen metrics, six categories
Cadence and rhythm. Landing and braking. Posture and alignment. Running economy. Drive and extension. Stability and symmetry. One headline score is the mean of those six categories, so a category never gets outvoted just because it holds fewer rows.
Seven more movements, each with its own categories
Wall ball, sled push, ski erg, rowing, back squat, overhead press and pull-up. A squat has no cadence and a runner has no depth, so every discipline gets categories that actually belong to it rather than borrowing running's six and leaving them half empty.
Form under fatigue
This is the section that justifies filming a whole set instead of three clean reps. The set is split into thirds and each third is scored against your own best third — not against a textbook target, because absolute quality is what the rest of the card already measures. Durability is a separate question: does the pattern survive. Under nine reps we return nothing rather than guess, because 'your technique broke at rep eight' is the most falsifiable claim in the app and a wrong one sends you to break a set you could have held.
Why bands and not decimals
Cadence and trunk lean are reliable from one 2D camera. Overstride and left/right asymmetry are angle-sensitive and noisy. Printing a decimal on a noisy measurement is how an athlete stops trusting the whole card the first time it contradicts itself.
Measured
• Joint angles — trunk lean, knee bend, hip extension, elbow angle, hinge depth
• Timing — cadence, step-interval spread, rep rate, consistency, left/right timing difference
• Ranges — squat depth against parallel, lockout, catch and finish positions, depth fade across a set
• Centimetre metrics — overstride, bounce, step length — for running and sled push, from your height
Not measured, and why
• Bar path, ball height, load, watts, split, calories — the camera sees no equipment, ever. The pull-up is the one exception, because your hands are on the bar
• Back rounding — no thoracic or lumbar landmark exists on either pose provider, so a clean hinge and a rounded one look identical. This is why the deadlift is absent
• Hip drop, knee valgus, arm crossover — frontal-plane, and side-on footage cannot separate alignment from depth
• Foot-strike pattern and ground-contact time — strike type is not graded on purpose (landing position matters more than which part lands first), heel landmarks are blank on Apple Vision, and contact is roughly two frames at our sampling rate
• Ground reaction force, power, or injury probability — not derivable from pose at any frame rate. Printing them would be fabrication, not estimation
Questions
Why not just estimate the things you cannot see?
Because an estimate presented as a measurement is a lie with a decimal point on it. An athlete who trains toward a fabricated number wastes a training block.
Will the missing metrics ever arrive?
Some need a second camera angle (frontal-plane alignment), some need a higher frame rate (Olympic lifts), and some need hardware the phone does not have (force). Those are different products, not thresholds we can tune.
Are the thresholds validated?
They come from mainstream coaching practice and are tested against synthetic athletes, which proves the geometry is recovered correctly. Calibration against filmed sessions is ongoing, and the scorecard is deliberately banded rather than precise for exactly this reason.
Keep your results
The free check scores one clip and forgets it. An account saves every analysis, writes the 4-week plan around your worst band, and shows whether the number actually moved when you re-film. Five analyses a month, free.
Create your free account
Download on the App Store
Or get the iPhone app — the same analysis, filmed live on your phone.