Film a swing on your phone. We find the eight positions a coach would talk through, measure the joint angles at each one, and compare them against published biomechanics research — and we tell you plainly which of them we could not actually see.
Early build · invited testers only
The trail arm and leg change as they pass behind the body. That is the system reporting what it can and cannot see — not an animation effect.
The problem
Consumer swing apps present a confident number for everything — swing plane, clubhead speed, hip rotation — with no citation and no indication of whether the camera could see the joint being measured. A phone camera genuinely cannot measure some of what they report. A joint hidden behind the body is inferred by the pose model, not observed, and the inference is confident and wrong.
30°
A physically impossible elbow angle
Measured at impact from landmarks the model had not actually seen. Caught by our reliability gates; a system without them would have scored it.
45–60°
Clubhead arc between frames at 30fps
Which is why we decline to report clubhead speed from ordinary phone video, and say so, while competitors report it.
74cm
Motion blur on a 1cm shaft
At a typical shutter, a 100mph clubhead smears most of a metre. There is no shaft left in the frame to detect.
How it works
Any phone, down-the-line or face-on. No sensors, no wearables, nothing to attach to the club.
Address, Toe-up, Mid-backswing, Top, Mid-downswing, Impact, Mid-follow-through and Finish — detected automatically, then measured.
Joint angles at each position, each graded against published research where one exists, with everything we could not see marked as such.
What is different
Not a promise in the marketing copy — a constraint enforced in the code.
Every measurement is checked for whether the joint was actually observed rather than inferred, whether the reading holds still across neighbouring frames, and whether the limb is angled too far toward the camera for a two-dimensional angle to mean anything. Fail any one and it is reported to you with a reason, never scored.
A measurement is only graded against a specific published figure, and the loader rejects any check marked scoreable without a verified source. Where no reference exists the reading is shown and left ungraded rather than given an invented threshold.
Your swings are scored against your own best, not against a tour professional with different proportions and mobility. The dashboard unlocks at three swings, because with fewer any difference is as likely to be how you filmed it as how you swung.
Where we are
Since the same claim applies to this page: here is the real state of it.
8
Swing positions detected
Automatically, from one camera.
4 of 9
Checks with verified sources
The other five are displayed but never scored, because their references are coaching consensus rather than a followable figure.
3
Reliability gates live
A fourth, checking anatomical consistency, is built but disabled until it earns its place.
Not yet
Club tracking
Three attempts have failed visual verification. It stays off and the club is clearly labelled as hand-inferred.
Upload one swing and you will get the eight positions, the joint angles we could measure, and a straight answer about the ones we could not.