Apple Health Walking Steadiness: samples, events, and context

Collection conditions are part of the data

Apple documents that Walking Steadiness is automatically recorded on iPhone 8 or later when the phone is carried near the waist, such as in a pocket, and the person walks steadily on flat ground. Height must be set in the Health app. If wheelchair status is enabled, the iPhone does not record Walking Steadiness samples.

These are not incidental notes. They define when the estimate is intended to be produced. An analysis should preserve whether the phone was carried in the expected position, whether the height profile was available, and whether wheelchair status or another setup choice changes the collection path.

• Supported device context: iPhone 8 or later.
• Placement context: near the waist during steady, flat-ground walking.
• Profile context: height is required; wheelchair status changes whether samples are recorded.

The quantity sample is a periodic estimate

The appleWalkingSteadiness HealthKit type uses percentage units and represents a discrete value between 0.0 and 1.0. Apple says the system creates a sample every seven days, although the interval can be longer when there is not enough mobility data to calculate a result.

A weekly sample is not a continuous monitor. It summarizes the data available for the estimate date and should be compared with other samples only after checking the collection interval, source device, and surrounding mobility context. Do not fill gaps with a guessed score or treat the next sample as proof that the interval was stable.

Classification and notification are different records

HealthKit provides APIs to map a percentage to an OK, Low, or Very Low classification. It also defines appleWalkingSteadinessEvent as a category sample for an incident where the person received a reduced steadiness score. The category event is read-only and does not replace the underlying percentage sample.

A notification event therefore answers a narrower question: a system-defined event was recorded. It does not contain a video of the walk, explain why a value changed, or prove that a fall will occur. Keep the event kind, timestamp, and source attached when summarizing repeated notifications.

• Quantity: a percentage estimate with a sample date.
• Classification: the label derived from that percentage.
• Event: a read-only record of a reduced-score notification state.

Missing samples and context change the question

A missing Walking Steadiness sample can reflect insufficient mobility data, a different phone-carrying pattern, a changed Health profile, a supported-device change, or a setting that prevents collection. It is not evidence that gait was normal or abnormal during the missing period.

Other mobility quantities—walking speed, step length, asymmetry, or double-support percentage—answer related but different questions. Combining them into one mobility score without declaring the method can make a trend look more precise than the source data supports.

A synthetic mobility timeline

The following values are invented and contain no real mobility data.

• Jan 1 — an iPhone sample was recorded after the phone was carried near the waist during flat-ground walking.
• Jan 8 — no sample was created because the available mobility data was insufficient.
• Jan 15 — a percentage sample was classified as Low and a notification event was recorded.
• Defensible summary: one classified sample and one notification event exist; the missing week is unknown and no future outcome is established.

What an AI analysis should preserve

Before asking a model to summarize Walking Steadiness, include the sample date and interval, percentage unit, classification, event kind, source device, iOS version when available, height-profile status, phone-placement context, wheelchair setting, selected date range, missing intervals, and extraction time.

The model can describe the recorded sequence, compare declared windows, and list collection gaps or setup changes. It should not infer a gait disorder, predict a fall, recommend exercises as treatment, invent a population normal range, or turn a notification into a clinical conclusion.

MCP may be an optional third-party interface for a minimized, user-approved summary. It is not an Apple protocol, does not grant HealthKit permission, and should not transmit a full mobility history when a date-limited aggregate answers the question.

Privacy and clinical boundaries

Mobility records, notifications, timestamps, and phone-location context can reveal daily routines and vulnerability. Request only the fields needed for the stated question. Prefer a selected date range and source-labeled summaries when raw samples are unnecessary, and disclose external model processing, retention, deletion, and revocation behavior.

This article explains a data model and interpretation method. It does not diagnose a gait or neurological condition, estimate an individual fall probability, prescribe exercises, or replace an appropriate clinician or physiotherapist. A concerning symptom or notification deserves professional context rather than an AI-only conclusion.