Definitions before comparisons.

Country context and measured use.

The context map combines 14 World Bank indicators into directional human-pressure, deployment-condition and bridge-gap scores. Country adoption estimates, individual surveys and enterprise surveys are joined by country code and shown alongside the context. They do not enter those scores. Scoped productivity studies remain separate from country comparisons.

Two clocks on every view.

The edition is a frozen evidence package. Reviews run quarterly, beginning 1 October 2026; edition keys retain the last month of the completed quarter (September, December, March or June). The August 2026 launch edition remains available. The observation period tells you when the activity was measured. “Comparable edition month” includes only matching full calendar-month observations. “Latest available evidence” retains quarterly estimates and annual survey waves with their actual periods. Data is known to this edition only by its stated cutoff; no historical publication date is invented.

One comparison policy.

Country ranks require the same metric version, measurement profile, unit and native observation period. Scenarios, restricted values, unreliable observations and observations marked with differing definitions are excluded. A latest-evidence table may contain older country records, but these receive no rank in the newest cohort. Country and platform audiences are never summed to create a unique category audience.

Microsoft diffusion is an adjusted estimate for people aged 15–64. Eurostat individual surveys cover ages 16–74, including people offline. Enterprise surveys count enterprises. These denominators cannot be swapped or pooled. Microsoft estimates may use regional adjustments in thin-telemetry markets; the atlas does not independently allocate regional values to countries.

Attention and benefit need stronger inputs.

Monthly app hours = the sum of each day’s active entities × foreground minutes per active entity, divided by 60. The shortcut using average DAU works only with weighted minutes, matching scope and full coverage. Multiplying monthly active users by active-day minutes and the number of days is invalid.

Net hours saved = quality-equivalent baseline time minus instruction, interaction, review, correction, escalation and allocated setup time. Keep negative results. Cost per success includes the costs of every failed and successful attempt; it is undefined with no successful outcomes. GMV, provider revenue, capacity released, realized savings and GDP are separate accounts.

Revision and refresh rules.

Each edition freezes its observations, definitions, source metadata and file hashes. Corrections receive a new revision; the original remains accessible. No monthly change is computed from carried-forward annual data. Source breaks interrupt trends. Public records must pass source, geography, date, value and reuse checks before publication. Quarterly review is scheduled for 1 January, 1 April, 1 July and 1 October at 08:00 Europe/Zurich, starting 1 October 2026; source releases arrive on their own calendars.

Initial staleness policy: monthly records older than 45 days, quarterly records older than 135 days and annual records older than 18 months are stale, measured from their observation period. A stale record can remain useful context with its date visible. Unknown sample sizes and intervals remain unknown.