AR measure apps

No - a phone AR measuring app is not an instrument for this. Its makers call the readings approximate and tune them for well-defined objects half a metre or more away, not a small, soft, curved target. The app builds a rough 3D map from camera and motion sensors, then reads the distance between two points on it.

How the estimate is built

An AR app tracks feature points across camera frames as the phone moves, using that motion to triangulate depth, then fits a mesh to what it can see. The whole approach depends on finding distinct, high-contrast features it can track frame to frame, and on surfaces that hold still and reflect light consistently while the phone moves around them.

Where the stated accuracy comes from

The platform makers are candid about the conditions. Apple's guide to its Measure app advises using it "on well-defined objects located 0.5 to 3 meters" from the phone, and adds in brackets that measurements are approximate. Google's ARCore depth documentation says depth is most accurate "half a meter to about five meters away", and that "surfaces with few or no features" produce imprecise depth. Neither publishes a figure that competes with the ±0.3 cm a careful manual reading with a rigid ruler achieves, as the instrument comparison sets out, and those caveats apply before the target gets any harder than a door frame.

Why the target here is the hard case

Skin is not a hard, flat, high-contrast surface - it is soft, curved, low-contrast, and moves. A curved surface gives the app fewer reliable tracking features and a less certain mesh; a soft one means the "surface" being mapped is not fixed the way a wall is; and the object is small enough that ordinary phone-to-subject distances push the app toward the near limit of what its depth sensing resolves well. Each of those factors pushes error in the same direction, away from the app's best-case conditions rather than toward them, so whatever error the app shows on a door frame is a starting point rather than a ceiling here.

Why manufacturer accuracy claims do not transfer here

App listings and hardware manufacturers often quote a headline accuracy figure - a small percentage error, or a fraction of a centimetre - and that figure is measured under the conditions the technology is designed for: bright light, a static hard surface, a reasonable working distance, and a target large enough to give the tracking algorithm plenty of features to lock onto. None of those conditions is guaranteed, or even likely, for a small, soft, curved target measured close up, often in indoor lighting that is dimmer and less even than a manufacturer's test lab. Quoting the headline figure as if it applied to this use case is the same mistake as quoting a ruler's millimetre markings as the accuracy of the whole measurement process - the instrument's best-case number and the achieved accuracy for a specific task are not the same claim.

What a repeat reading looks like

Because AR measurement rebuilds its scene estimate fresh each time, taking the same measurement twice in quick succession, without moving the phone in between, can still return two different numbers - a sign that the app's underlying spatial map, not just your placement of the two points, is unstable for this kind of target. That instability is itself informative: an instrument that cannot agree with its own repeat reading on the same static subject is not one to trust for a single figure, regardless of what any headline accuracy claim says.

What this has in common with a photograph

Like a single photograph, an AR estimate is a projection-based inference rather than direct contact between an instrument and the body - it shares the category of error, even though the underlying technique (depth sensing versus a flat 2D image) is different. Neither approaches the reliability of a rigid ruler pressed to bone, which needs no camera, no lighting, and no surface reconstruction at all. AR apps are just one entry in a wider category - a rundown of gadgets that cannot measure this reliably covers the others, and why each falls short for the same underlying reason. AI Penis covers a related but separate inference problem - an image model estimating size from a single photo rather than a phone rebuilding depth from motion - and its explanation of why a picture cannot substitute for contact measurement applies just as directly to a depth-sensed estimate.

Not a substitute for a rating, either

An AR app's number, however it was produced, is still an attempt at a length, which puts it in a different category from a subjective score. Something like Rate Cock is not attempting to recover a distance at all - it is producing a rating from an image, a different kind of number entirely, judged the way any photo-based score is, and a human's verdict on Rate Penis is an opinion rather than a measurement regardless of what sensor was or was not involved. For an actual figure in centimetres, a tape or a rigid ruler in direct contact with the body remains the instrument that works.