You already know
the exercises.
This tells you if
they're working.

The measurement

Two markers cost you years. We measure three more anyway.

Postural Age is your age plus the years each deviation adds. These are the thresholds the app scores against, not a summary of them. The unscored markers have no thresholds in this table, either because nobody has published one or because the measurement cannot support one, so we left those cells empty rather than fill them with a number we made up.

On the three we measure and do not score. We measure sagittal alignment on every scan and score it on none. The measurement itself holds up: it is the hip angle used by SAPO, a postural assessment tool used in research, where five blinded raters agreed on it to an ICC of 0.93 on the right side and 0.86 on the left (Ferreira 2010, 22 adults, 88 photographs). Nobody has published a normal range for it in healthy adults. Measuring something consistently is not the same as knowing what a good reading looks like, and we would rather leave the cell empty than invent a threshold to fill it. So we compute the angle, store it, and show it to you next to the two that count. It adds nothing to your Postural Age. One thing to hold against that 0.93: it came out of a lab, from markers a clinician placed by hand on bony landmarks. A phone guessing at the same joints from a photograph has the harder job.

Shoulder symmetry joined it on 17 August 2026, and a clinician made that call too. Asked whether left-to-right shoulder height should carry up to three years, the physiotherapist reviewing our method answered that he would measure and display it but keep it out of the score, "unless there's a clear clinical or empirical basis for penalizing a specific degree of asymmetry." Two things we already knew argued the same way. The paper we had been citing measures rounded shoulders as present or absent by plumb line, not height difference, so it never matched what we compute. And a lower right shoulder is reported in roughly 68% of people, which means we had been charging most of our users for the population norm. The measurement stays on your breakdown. It costs nothing.

On the forward neck tilt angle. Until 1 September 2026 this row read "craniovertebral angle", with a normal threshold of 40° derived as app angle = 90° − clinical CVA for the C7-to-ear vector. The app has never measured that vector. Apple's Vision framework gives us a "neck" point that is the arithmetic midpoint of the two shoulders, not C7, so what we actually compute is the ear against the shoulder line. Every production reading we have sat between 1.9° and 23.0°, far below 40°, which means the largest marker on this table — six of the available years — had never fired for anyone. A threshold cannot be repaired by moving it when the vector underneath is measuring something else.

What we compute is itself a published construct: the forward neck tilt angle, tragus to acromion against the vertical. It needs no conversion, because it is already defined from the vertical. The severity bands come from Moon 2024 (n=145, aged 10 to 19): normal 0–12°, level 1 12–25°, level 2 25–40°, level 3 at or above 40°. Our severe bound takes level 3. Two things you should know about that paper before trusting the bands: they are the authors' own and cite no external source, and the radiographs validated a diagnostic model combining this angle with eight other parameters, not these four boundaries. The paper calls itself preliminary in its own title.

The no-penalty boundary of 20° is not Moon's. Moon's level 1 begins at 12°. Reviewing the re-anchored marker on 4 September 2026, the physiotherapist assessing our method set it at 20° instead, "so that you can address a larger population without much error", and we told him what that costs before he confirmed it. Across eleven production readings, a 12° boundary charges six of them a mean of 0.72 years; his 20° charges one of them a mean of 0.08, and the worst reading we have seen pays 0.90 of a possible 6. He was choosing specificity over sensitivity deliberately, and he kept the number after seeing the arithmetic. We would rather report a marker that rarely fires than one that charges healthy people for standing normally.

Caveats we owe you: the sample is adolescent, the measurements were taken with a depth camera rather than photogrammetry, Vision gives us an ear centroid rather than the tragus and a shoulder centre rather than the acromion, and the construct cannot see a forward head where the head and shoulders have drifted forward together. That is weaker anchoring than a validated CVA cutoff would be. It is, however, anchoring for the vector we actually measure, which the old 40°/55° pair never had.

On the forward shoulder angle. The same convention flip applies here. The published Shoulder Angle runs off the horizontal from the same neck-to-shoulder vector we use, so app angle = 90° − clinical. Ruivo 2014 measured 275 people and put the protraction cutoff at a clinical 52°, which arrives here as 38°. We published 20° until 29 July 2026, and that was wrong in an expensive direction: 20° on our scale is a clinical 70°, about 2.2 standard deviations above Ruivo's mean of 51.4° ± 8.5. Someone with an average shoulder angle was collecting roughly 2.7 of the 5 available years for standing normally.

We then went past the published cutoff, and a clinician sent us there. Reviewing our thresholds on 17 August 2026, the physiotherapist pointed out that 52° penalises around 40% of people who have no symptoms at all, and asked us to loosen it. He named 43.5° on our scale. Against Raine and Twomey, who found a mean of 53.7° across 160 asymptomatic people aged 17 to 83, that takes the share of symptom-free adults picking up a penalty from 42.1% to 19.8%. In Ruivo's 15-to-17-year-olds it goes from 52.8% to 28.2%. We report both because he asked for the age groups separately.

So 43.5° is no longer a transform of anyone's published cutoff. It is a clinician's number, chosen against a target rate of false alarms, and we would rather say that plainly than dress it up in a citation. One caveat we owe you: Raine and Twomey publish a mean but not a standard deviation, so the adult figures above borrow Ruivo's. The adolescent figures do not. The severe cutoff has not moved and is still ours rather than anyone's published figure.

None of that matters any more, and the reason is worth stating plainly. On 1 September 2026 we stopped scoring this marker altogether. Correcting the far end of the vector — we had been measuring to the point of the shoulder rather than to the middle of the upper arm — exposed the near end as the real problem. The neck point Vision gives us is the midpoint of the two shoulders, and that point travels forward with the shoulder girdle itself. A vector anchored to the thing it is supposed to be measuring against cannot detect rounded shoulders at any cutoff. The physiotherapist reviewing our method agreed on 4 September, in his words, that we "can remove it as app cannot properly measure it", and signed the amendment on 5 September. We still measure the angle and still show it to you. It costs nothing, and no threshold edit would have fixed it.

Angles come from normalised image coordinates, so they are calibrated for what a phone camera can see rather than read off a clinical goniometer. Postural will only give you a score when the markers it managed to measure carry the majority of the available years, so forward head on its own is enough at six of eight, and head level on its own is not at two of eight. Below that it asks for a retake instead of filling in the gaps. The three unscored markers carry no years and so count toward nothing here, and when a scan rests on one marker rather than both the app says so under your number.