Calibration &
proportions.

Placement guidance (§7) is locked until a valid calibration exists. This is not an upsell — without a scale reference, a bar-position recommendation would be an opinion dressed as a measurement, and Cence does not ship those.

Nothing derived from calibration decides a competition, seeds a bracket, or adjusts a rep standard. Movement standards apply identically to every athlete regardless of proportions.

You tell Cence which reference you are using before capture. The engine confirms it can locate an object of that shape in frame and rejects the session if it cannot — it does not guess which object you meant.

The reference must sit in roughly the same plane as the body being measured — held against the torso or placed at the feet, not on a wall several feet behind. An object further from the camera than the subject reads smaller and inflates every measurement taken from it.

Calibration captures a still A-pose — standing tall, arms at roughly 45°, palms forward, feet hip-width. Limb segments must be unflexed; a bent elbow shortens the forearm in the image and there is no way to recover the difference.

A front capture and a side capture, both with the reference in frame. Single-view calibration is accepted only at a widened tolerance, because one view cannot resolve limb foreshortening toward the camera.

Camera at chest height, 8–12 feet back, phone vertical and stable. Lens distortion at the frame edges is corrected using the device profile where one is available and flagged where it is not.

Loose clothing over the shoulder, elbow, hip or knee reduces landmark confidence. The engine reports which segments were affected rather than silently including them.

Ratios between body segments are available from any valid capture, because a ratio has no units to get wrong: forearm-to-upper-arm, femur-to-torso, wingspan-to-height, shoulder width relative to hip width, left-right symmetry.

These are the most useful outputs Cence produces and the most reliable. A long-femur athlete squats at different angles than a short-femur athlete at identical legitimate depth — the ratio explains it, and no unit is required to act on it.

With a valid reference, segment lengths are reported in units, each carrying the tolerance of the reference tier used. A Tier C calibration produces a forearm length stated as a range, not a figure.

Cence measures between tracked joint centres derived from surface geometry, not between anatomical landmarks located by palpation. These are close but not identical, and the difference varies between bodies. Every measurement is labelled as a joint-centre measurement so it is never mistaken for a clinical one.

Reference-tier error, landmark confidence, and view geometry each contribute. The reported tolerance on any measurement is the combination of all three, not the best of them.

Where the front and side captures produce segment lengths that disagree beyond tolerance, the engine does not average them. The session is marked unresolved and you are asked to recapture.

Any surface that displays a measurement displays its tolerance and reference tier alongside it — in the app, in an export, and in any shared card.

Without a valid calibration, placement feedback is relative only — it can tell you a position changed, not whether it is right for your build. A current calibration unlocks proportional targets and tracks them across sessions.

Every placement recommendation is checked on your following attempt. Guidance that does not produce the predicted joint angle is withdrawn, not repeated.

Placement is coaching. It never counts, discounts, or flags a rep, and it never appears in a competition verdict. Movement standards judge the movement; placement helps you train.