Methods
How every number is made
A number without its method is a rumor. This page is the method: what each measurement is, what counts as a day, how sources are merged, and what is never shown.
Measurements
| Measurement | Daily value | Limitation shown beside every chart |
|---|---|---|
| Steps count per day | Daily cumulative count from the phone's health store using one documented source-merging query, validated against the Health app's own total. | Device-recorded steps. Not all movement, and not proof the phone was carried all day. |
| Walking and running distance meters per day | Daily cumulative distance under the same date and source policy as steps. | Device-estimated. Phones and watches estimate distance differently. |
| Resting heart rate beats per minute | Daily mean of readable resting-heart-rate estimates. Requires a watch. Optional. | A summary of device estimates, not a clinical measurement. |
Steps and resting heart rate describe movement, not health. The map never shows a diagnosis, a biological age, a risk score, or a life expectancy.
What counts as a day
Three different states
A readable value, an explicit zero, and nothing readable are stored as three different things. A missing day is never filled in as zero. A zero is never treated as proof someone sat still all day.
One reporting day
Each contributor picks a reporting time zone once. Days are calendar days in that zone, including the 23-hour and 25-hour days at daylight-saving transitions. No travel or location data is used to decide this.
Phone plus watch
Raw samples from a phone and a watch are never added together. A statistics query that merges sources produces the daily total, and it is validated against the total the Health app displays.
Manual entries
Where a manually typed value can be identified it is excluded from passive-sensing summaries. Where provenance cannot be established, the limitation is stated rather than claiming sensor verification.
Late edits
Samples that arrive or are deleted later cause the affected day to be recomputed. Each recomputation is a new revision. Old results are never silently rewritten.
Retries are not observations
Each day for each contributor and measurement has one key. Replaying an upload cannot create a second observation or raise a count.
Counters
| Counter | Definition |
|---|---|
| Contributors | Distinct people with accepted consent and at least one valid day. Installs are not contributors. |
| Contributed in the last 30 days | Distinct contributors with at least one valid measurement-day in the window. |
| Contributor-day | One person on one date with at least one valid measurement. Counted once regardless of how many measurements. |
| Person-years on the map | Contributor-days divided by 365.25. Does not imply continuous wear or continuous participation. |
| Sync freshness | Time since the last successful contribution, not the last attempt. |
The baseline layer and the sketch
While the live cohort is small, some views are drawn over a clearly labeled baseline built from published population data. The baseline and the live layer are never blended into one number. When the live cohort supports a view on its own, the baseline is removed from it.
Published step distributions come from different instruments: hip-worn accelerometers in national surveys, phones in the large published datasets, and a mix of phones and watches on this map. They disagree by thousands of steps a day. Any baseline shown here names its instrument and its year.
The age curve on the home page is cross-sectional: the median of people currently at each age. It is not a forecast for any individual. People who stayed active are over-represented at older ages, so a person's real slope is usually steeper than the curve. Within-person slopes appear only after a contributor has been on the map for twelve months, because seasons alone swing steps by thousands a day and a twelve-week slope is noise.
In this demo the baseline and sketch values are fictional placeholders. The production Methods page will cite the actual sources.
What is never shown
- Any cell with fewer than 50 distinct contributors. This is a conservative starting rule, not a guarantee of anonymity; some combinations stay unpublished above it.
- Individual timelines, contributor lookup, or free-form subgroup filters.
- Frequent micro-updates that could reveal one person arriving or leaving. Releases are fixed and dated.
- Causal claims. Associations are described as associations. An activity dip is never interpreted as a medical event.
Successive releases are reviewed for differencing and linkage across totals, related tables, and deletions, not only for the per-cell minimum.
It is a map, not a census
People who install an app to contribute their steps are not the population. The map does not claim to be. It says: here is what the people on the map are doing. Device access, region, and unanswered profile questions are shown as part of the data rather than hidden. Weighting may be added later; it cannot say anything about people who were never here.