Reading a Posture Score: What 92 Actually Means
形正之度
The number is a similarity measure against a reference frame, not a judgement of your practice. What it can see, and the four things it cannot.
The number is a similarity measure against a reference frame. It is not a judgement of your practice, and knowing what it can and cannot see is what makes it useful rather than distracting.
What the number is
The pose estimation model returns coordinates for around thirty-three body landmarks. From those, a set of derived values is computed — the angle at the knee, the height of one shoulder relative to the other, the angle of the spine against vertical, the position of the wrists relative to the torso.
Those values are compared against reference values for the movement you are performing, captured from the demonstration. The score expresses how closely they match, weighted so that the elements the movement actually cares about count for more.
- 形xíng
- Form, outer shapeWhat the score measures. The external geometry.
- 意yì
- Intent, attentionWhat it cannot measure, and what the practice largely consists of.
- 形正氣順xíng zhèng qì shùn
- Form correct, breath smoothThe traditional pairing. The score can assess the first half only.
What 92 means
It means your joint geometry is close to the reference for that movement. It does not mean your practice was 92 per cent good.
| Range | Reasonable reading |
|---|---|
| 90–100 | Geometry closely matches the reference. Nothing gross is wrong. |
| 75–89 | Broadly correct with one or two identifiable deviations. Read the cue. |
| 60–74 | A real alignment problem, or the camera has a poor view of you. |
| Below 60 | Usually framing — you are partly out of shot, badly lit, or side-on. Check that before assuming your form collapsed. |
The four things it cannot see
- Rotation. A single camera sees a two-dimensional projection. Rotation toward or away from the lens is the axis it estimates worst — which is unfortunate, because knee tracking is exactly a rotation problem.
- Tension distribution. Two hands in identical positions can be doing entirely different things. Lotus Palm is specified by its tension pattern, and tension has no silhouette.
- Breath. Nothing in the pipeline observes breathing. You can hold your breath through an entire movement and score well — see Five Beginner Mistakes.
- Attention. The largest gap. A perfectly executed sequence performed on autopilot is geometrically indistinguishable from an attentive one.
The crown suspension illustrates the limit precisely. The model can tell that your spine is vertical. It cannot tell whether you achieved that by hanging from the crown or by pushing up from below — and those are opposite instructions with opposite consequences.
Why cues matter more than scores
A number tells you that something is off. A cue tells you what to do. The design bias throughout is toward the second: "drop the shoulders two centimetres" is actionable in a way that "84" is not.
This is also why guidance is written as instruction rather than error. Colour-coded red-means-wrong feedback trains people to chase the indicator, which produces stiff, self-conscious movement — the opposite of what is wanted. Cues are held on screen long enough to read, and adjustment guidance uses a neutral gold rather than an alarm colour.
Getting a usable reading
- 01Frame the whole body
Head to feet in shot, with a little margin. Cropped feet make every leg measurement unreliable.
- 02Camera at chest height
Low angles distort the spine and leg angles badly. A phone on the floor gives consistently poor readings.
- 03Face the camera for most movements
Front-on gives the best estimate for the majority of the curriculum. A few movements read better from the side.
- 04Even lighting, contrasting background
Backlighting from a window is the most common cause of nonsense results. Avoid clothing that matches the wall.
- 05Move at practice speed
The model works on individual frames. Moving unnaturally slowly to chase a score changes the movement into something else.
The honest summary
Pose scoring is a decent substitute for a mirror and a poor substitute for a teacher. It catches the gross errors that people cannot see in themselves — a locked knee, a lifted shoulder, a spine off vertical — and those account for a meaningful share of what goes wrong in solo practice.
What it cannot do is put a hand on your shoulder and say here, feel this. Anyone claiming an app closes that gap is overselling. Treat it as an attentive mirror: useful, limited, and worth switching off once it has told you what you needed.