How it works — and what it doesn’t claim.
Any tool can put a number on a golf swing. Two questions decide whether that number means anything: compared to what, and which frame. This page answers both, then shows you where we stop.
How Timely worksFrom a phone clip to a number a coach will put their name on.
Seven stages. What matters is not that there are seven — it is that 4 of them are allowed to withhold something. A pipeline that can only produce answers will always produce one, whatever you feed it.
Grade the clip
The capture is judged before it is measured
Framing, light, focus, frame rate and whether tracking held on one person — all assessed before a single angle is read.
Can withhold
A clip that cannot support honest numbers is flagged, and can be blocked.
Read the body
Your body is tracked, on your device
Pose tracking runs in your browser and reads 33 body points in every frame. The video does not need to leave the device to be measured.
Square it up
Left and right handed become comparable
A left-handed swing is squared up so one set of reference ranges is valid for both. Only the angles are made comparable — your video is never flipped or altered.
Find the frames
The five checkpoints are located in the swing
Checkpoints are read from the motion of your body, not sampled on a timer, so the labels follow the swing instead of the clock.
Can withhold
Two positions in the coaching vocabulary are defined by the club shaft. Pose cannot see a club, so they are never reported.
Check the angle
The camera view decides what can be scored
Every measurement carries the view it needs to be reliable. Rotation wants down-the-line; weight wants face-on.
Can withhold
A measurement your angle cannot support is excluded from the score — not estimated, and not counted against you.
Compare
Measured against a real swing, not an ideal
Your pose at a checkpoint is compared to a curated tour reference at the same checkpoint and the same camera view — pose against pose, read the same way on both sides.
Verdict, then human
A band goes out; a coach decides when
The result is a band — tour range, on track, developing — with the score trailing it. Low-confidence data caps the score rather than inflating it.
Can withhold
Everything the AI writes is a draft until your coach has reviewed it and chosen to release it.
Compared to a real swing, not an idea of one.
The tempting way to score a swing is to write down an ideal — one target number per angle — and measure your distance from it. It reads well, and it is wrong twice: no golfer has ever had that swing, and the depth error every single-camera video carries gets scored as if it were a flaw in your golf.
So Swing Lab does not score you against an ideal. Your body at a checkpoint is compared to a curated tour reference at the same checkpoint, filmed from the same angle, read the same way. Put a genuine tour swing through it and it comes out looking like a tour swing — which is the least a scoring system should manage, and the one thing an abstract ideal can never guarantee.
- The same positionYour top is compared to a reference’s top. A checkpoint is never scored against a different checkpoint.
- The same camera viewA face-on swing is compared to a face-on reference. The depth error a single camera cannot avoid then sits on both sides of the comparison, where it largely cancels instead of being scored as your fault.
- The same reading, both sidesReference swings were read by the same pose tracking that reads yours. A systematic quirk of the tracker is shared by both poses rather than charged to one.
- A range, not a single frameWhere enough verified tour swings exist for a measurement, you are judged against the spread that group actually occupies. Tour players do not all set up identically, and a system that pretends they do is wrong about golf.
- What your angle can’t support is droppedA measurement your camera view cannot make reliably is excluded from the score. Not estimated, and not counted against you.
- A tracking glitch is not a swing faultReadings that look like sensor error are withheld and reported as withheld, rather than scored as something you did.
The result is a band first. The number trails it.
A band is what the comparison can honestly support. Leading with a decimal would imply a precision that a phone camera and a human body do not have between them.
Inside the range tour swings occupy at that position.
Outside it, but not far enough to be the thing to fix first.
Furthest from the reference — named plainly rather than softened.
Finding the right frame is most of the problem.
Measure a perfect takeaway on a frame from the top of the backswing and you get a confident, detailed, entirely wrong result — and nothing on screen looks wrong. That is the most dangerous failure in a system like this, so it gets the most care.
Frames are not sampled on a timer, and checkpoints are not guessed from where they usually fall. The swing is first located inside the clip — however long you left the camera running before and after — and that window is then re-read at higher resolution. The checkpoints come from the motion of your body itself: where the hands begin to move, where they reach their highest point, where they reach the bottom of their arc, where the motion settles.
Because the trace follows both hands and reads speed without a direction, a left-handed swing traces exactly like a right-handed one. Handedness is squared up once, in the angles only — your video is never flipped, cropped or altered.
Impact is approximate. The bottom of the hand arc sits at or just after true contact, and body pose cannot see a club, so we do not claim a frame-exact impact and you should not read one into it.
A detected sequence that is not physically possible for a golf swing is rejected rather than displayed — including the slow swings that a simpler check would throw away with it.
Five checkpoints, found the same way every time.
Coaching's P-system. Your body is measured at these exact frames — the same five whether the swing is yours, your coach's, or the tour reference it is compared against.
P1
Address
Setup, read before the club moves — stance, posture, spine angle.
P2
Takeaway
The first move away — arm structure and early rotation.
P4
Top
The transition — shoulder turn, hip turn and X-Factor at their peak.
P7
Impact
The moment that matters — body angles and weight as the club returns.
P10
Finish
The swing's receipt — balance and full rotation onto the lead side.
Scans using the ML event model add P3, P5 and P9. P6 and P8 are defined by the club shaft — body pose can't see it, so Swing Lab never reports them.
The clip is graded before it is measured.
Analysis of a bad clip does not look bad. It looks exactly like analysis of a good one, which is why the capture is judged first and separately — and why a poor grade can stop the run rather than decorate it.
Framing
How much of the frame your body fills. Film from too far back and every joint position gets imprecise at once.
Light
Brightness and contrast together. A silhouette against a bright sky is a shape with no joints inside it.
Focus
Judged from the sharpest frame in the clip, not the average one. A golf swing motion-blurs by nature, so grading the average would fail every real swing ever filmed.
Frame rate
When the true frame rate cannot be read from the file, it is left unknown rather than assumed. A guessed frame rate makes timing look precise when it is not.
One body, start to finish
If tracking jumps to somebody else mid-swing — a partner walking behind you, a reflection — that shows up as a break in the trace, and the clip is graded down for it.
Enough readable frames
Too few clear frames and the capture is graded unknown rather than scored. An unknown grade is a real outcome here, not a failure state.
The grade lands as good, limited, poor or unknown. Unknown is not an error — it is the system declining to grade a clip it cannot read.
What the output does and doesn’t tell you.
Five things worth knowing before you read a score — including the two places we deliberately show you less than we could.
What the numbers are made of
Every measurement comes from your body posture and motion, estimated frame by frame from the video: joint angles and positions — spine angle, stance width, rotation, arm structure, weight position — read at each of the five checkpoints.
Nothing on this page is measured from equipment or from the result of the shot.
How scores work
A phase score is a 1–10 assessment against fixed positional ranges for that checkpoint. Scores are clamped to that scale, and when the underlying data is low-confidence the score is capped — we would rather under-claim than inflate. A coach can override a score with their own judgment; that is the point of having a coach in the loop.
You don’t have to tell us how you filmed it
The camera view is worked out from the swing itself, from two independent signals that have to agree. When they disagree, the view is recorded as unknown — and unknown is treated cautiously rather than confidently, because a wrongly confident view would hide a metric your coach can actually see, or show one they cannot.
Where a metric is not supported by your view, you get an explicit “n/a from this view” rather than a fabricated number.
Consistency is a band, not a score
When you record several swings in one session from the same camera, we report how consistent a checkpoint is as a qualitative band — for example “fairly consistent” — based on how much that body angle varies across the swings. It is never reduced to a 1–10 number, and we never compare across different sessions or camera setups, where the numbers would not be comparable.
A coach reviews before you see feedback
AI output starts as a draft. Your coach reviews it, can edit the written feedback, and decides when it is shared with you. The provenance of the measurements is kept intact even when a coach edits the words around them.
What we don’t measure.
Body pose can’t see the club or the ball, so Swing Lab does not analyze or claim anything about ball flight or shot shape, club path, the clubface, swing plane or shaft position, or wrist hinge and lag. If you ever see language like that in your feedback, treat it as a bug — those claims are deliberately kept out of the system.
The analysis is also not deterministic: the same video won’t always produce byte-identical wording, and the scores are estimates, not measurements from instrumentation. Swing Lab is a structured second opinion to speed your coach up — not a launch monitor, and not a substitute for a coach’s eye.
Shot Tracer, a separate tool, draws the path of the ball onto your video. That line is a picture of the flight, not a measurement: it follows the ball while the camera can see it, predicts the rest, and reports no numbers about the shot.
For how we handle your video, audio and motion data, see our Privacy Policy.
The rest of the picture
How the whole product fits together.
Scored against the tour. Reviewed by your coach.
Timely brings booking, swing analysis, lesson notes and progress together — so coaches spend their time coaching, and players always know what to work on next.
