How our swing analysis works, and where it stops
Filmed on a phone, a swing analysis can measure some things precisely, estimate others, and see nothing at all of the rest. This page states which is which, with the counts and thresholds the system actually uses, so you can check the output instead of trusting it.
What the system measures
Each figure below is a count, not a claim. They are what the analysis produces on every swing it accepts.
- 26
- named faults detected — 10 from a down-the-line view, 16 from face-on
- 21
- body and club key points tracked through the swing
- 8
- swing phases the motion is cut into
- 17
- scored dimensions, plus one weighted overall score out of 10
The eight phases are the ones a teacher would name, and every fault is reported with the exact frame range where it appears — so you can scrub to it rather than take the label on trust.
Address · TakeAway · BackSwing · Top · DownSwing · Impact · FollowThrough · Finish
What it refuses to guess
Some measurements need a camera angle you may not have filmed. Kinematic sequence and X-factor need the face-on view; head rise and spine angle need down-the-line. When the view cannot support a measurement, the system returns it as not applicable and removes it from the overall score, rather than producing a plausible number. That is why two analyses of the same swing filmed differently do not report the same fields.
A system that always fills every field is not more capable than one that sometimes declines — it is less honest about the same limits.
The thresholds we publish
A threshold you can read is a threshold you can argue with. These are the ones the analysis applies:
- X-factor: flagged as insufficient below 25° and excessive above 60°.
- Tempo: measured against the roughly 3:1 backswing-to-downswing ratio observed on tour.
- Severity: every fault is returned as high, medium or low, with its start, end and detection frames.
The error rate we accept
Ball-flight prediction — slice, fade, straight, draw, hook — is right about 75 % of the time. We publish that figure because the alternative is implying it is 100 %, which no single-camera system achieves. Timing data such as tempo and phase duration is reliable to a frame or two. Body rotations are reasonably reliable. Clubface angle, angle of attack and anything about the ball are estimated from limited visual information and should be read as indicative, not measured.
What we do not publish
The internal scoring weights, matching windows and coefficients stay private. The counts and thresholds above are what you need to judge an output; the implementation is what makes it work, and it is ours. Anything on this page that you cannot verify against your own analysis is something we should not have written.