Partners, payers, and clinicians don't adopt marketing — they adopt a method they can check. So we publish ours: every metric, every threshold, every formula, every citation, and every limitation behind the gait number. It's built on Apple's on-device mobility data and a decade of peer-reviewed gait science — and it already runs, transparently, on a person's own phone. This is a methods card, not a brochure. Read it, disagree with it, improve it.
A serious AI lab ships a model card — capabilities, evaluations, and limitations stated plainly — because a system you can inspect is one you can trust, build on, and regulate. A serious measurement company owes the same. The gait number is a clinical signal; if we can't show our work, no clinician should act on it and no partner should attach their hardware to it. So the method is the product's front door, not its trade secret. The engine that runs it stays ours; the method is open for anyone to audit.
A shoe, insole, or exoskeleton team plugs into a documented method, not a black box — their regulatory and clinical staff can read exactly what the number means before they commit a roadmap to it.
Reimbursement and clinical adoption run on defensibility. Published thresholds with citations are the difference between "an app says so" and a measurement a utilization committee will stand behind.
Rigor is the moat. Anyone can draw a line on a chart; few will state the meaningful-change threshold, cite the study behind it, and name where the method breaks. That discipline is what compounds.
The method is sensor-agnostic by design: it consumes standard mobility metrics, whatever produced them. The reference implementation reads Apple Health — the mobility metrics an iPhone already computes on-device from its motion coprocessor, with no wearable required. The same fields can arrive from a clinical insole, a camera-pose pipeline, or a partner device.
| Metric | Apple Health identifier | Unit | What it captures |
|---|---|---|---|
| Walking speed | walkingSpeed | m/s | The primary signal — gait speed, the 6th vital sign. |
| Walking steadiness | appleWalkingSteadiness | % (OK / Low / Very Low) | Apple's validated fall-risk classification from iPhone mobility data. |
| Step length | walkingStepLength | cm | Stride geometry; shortens early in decline. |
| Walking asymmetry | walkingAsymmetryPercentage | % | Left/right timing imbalance — a limp signature. |
| Double-support time | walkingDoubleSupportPercentage | % | Share of the gait cycle on both feet; rises with instability. |
| Stair speeds | stairAscentSpeed / stairDescentSpeed | m/s | Load-bearing challenge; sensitive to lower-limb weakness. |
Walking speed is the anchor because it carries the most evidence. The rest sharpen the picture: two people can share a gait speed while one is asymmetric and spending more of each stride double-supported — the earlier warning.
Bands are not chosen for looks. Each cut point traces to the gait-speed literature — the same numbers used in geriatric assessment and physical therapy.
Because a raw number means little without context, the method also reports a gait age — the age at which the measured speed equals the population-average comfortable speed — interpolated from Bohannon's age-and-sex reference values (1997). "Your gait moves like the average 72-year-old" lands where "1.02 m/s" does not.
A single reading is noise. The signal is the trajectory — and a change only matters if it clears the threshold research says is meaningful. Here is the entire computation, no hand-waving.
Day-level median discards a single sprint-for-the-bus or a phone-in-a-bag artifact; the 14-day mean across those clean daily values is a stable "where you are now." Robust first, smooth second.
These aren't round numbers — they are the small and substantial meaningful-change values established by Perera et al. (2006), the thresholds physical therapy uses to decide a change is real, not measurement drift.
The anchors below are load-bearing — each one is why a specific number in this method is what it is.
Pooled analysis of 9 cohorts, 34,485 older adults: gait speed predicted survival as strongly as age and sex, with each 0.1 m/s increment tracking consistently better survival. Sets the survival bands and the ~1.0 m/s hinge.
Established the small (~0.05 m/s) and substantial (~0.10 m/s) meaningful-change values for gait speed. Sets the Watch / Slipping decline flags.
The clinical case for treating gait speed as a routine vital sign, with interpretive ranges for function and risk. Frames the whole method — and the brand.
Age- and sex-stratified normative comfortable gait speeds. Powers the gait-age interpolation and the community-strong band.
The national screen–assess–intervene framework for fall risk, with gait and mobility as core components. Anchors the screening posture and the elevated-range cut.
The published methodology and validation basis for iPhone mobility metrics and Walking Steadiness. Grounds the input data and the steadiness override.
The most important part of a methods card is the part that admits where the method stops. Stated plainly, because a partner's regulatory team will find these anyway — better they find them here.
The input metrics carry Apple's own validation for iPhone-derived gait parameters. The method's own thresholds are drawn directly from the cited literature rather than fit to a private dataset — so its behavior is auditable from public sources.
The claim still to be earned is that a detected decline predicts a future fall in a monitored population. That is a prospective outcome study — named here as the honest next step, not asserted as done. The measurement layer generates that evidence as a by-product of use.
The method runs where you can watch it: entirely in your own browser, on your own phone's exported health data. Nothing is uploaded — the computation happens on-device and the result never leaves it. Point your own Apple Health export at it and check the number against the formula above.
The identical thresholds and slope logic apply whether the input is Apple Health, a phone-camera pose pipeline, a 3-point AirPods + iPhone + Watch capture, or a partner insole. That invariance is precisely what makes this a method and not a product feature — validate it once, reuse it everywhere.
This is Method v0.9 (July 2026). Threshold or formula changes are versioned so any result can be traced to the exact method that produced it — the audit trail a payer or regulator expects.
A better shoe, a smart insole, an exoskeleton, a balance sensor — each produces a raw stream that, on its own, doesn't bill, doesn't screen, and doesn't carry a defensible clinical claim. Feed it into this method and it becomes a gait number, a cited risk band, a decline flag, and a monitorable trend — with the science already published for your regulatory and clinical teams to read before they sign anything.
You keep your brand and your hardware. We supply the measurement layer, the published method, and the clinical rail — the half most hardware teams choose not to build. If you build things that measure how people move, the method is open for you to check today, and the conversation is open too.
Start a partner conversation → Read the platform spec