Methodology

1. Question. Your question is read for its subject, places, time period and the kind of answer that fits (a number, a comparison, a map, a trend).
2. Tool search. Atlas searches the Humaps source registry and tool catalogue for sources that can answer it — statistical APIs, map-layer services, document repositories and live feeds.
3. Live calls to official sources. The selected sources are queried at the moment you ask. Figures come from those responses, not from the language model's training data.
4. Typed artifacts. Results are returned as structured artifacts — charts, tables, maps, timelines, reports — each carrying its sources, method and caveats alongside the data.

When several sources can answer the same question, Atlas prefers the one closest to where the data is produced:

• National first: the national statistics office or another official national source for the country in question.
• Then regional: EU and other regional bodies (for example Eurostat) where they harmonise national figures or fill gaps.
• Then global: international institutions such as the World Bank, IMF, OECD and UN agencies, which are best for cross-country comparison but can lag national releases or use different definitions.

A comparison across countries may deliberately use one harmonised international source so that definitions match; the artifact says which choice was made.

• Country coverage in Humaps (in the app and in the MCP connector) describes the routing candidates we know for a place: national, regional, global and local sources Atlas can try.
• It is not a guarantee of completeness. A listed source may not publish the specific dataset, year or breakdown you need, and an endpoint being reachable does not prove the dataset exists. Verify the selected source and dataset before relying on a claim.
• The full registry is browsable in the source registry.

• Every figure is attributed to the publisher and dataset it came from, with a link.
• Two dates matter and are kept apart: the reference period the value describes (for example “2023”) and the retrieval time when Atlas fetched it.
• Units, definitions and known breaks in a series are carried as caveats on the artifact rather than smoothed over in the text.

• Each artifact lists its source, the method or query used and the retrieval time, so the same request can be repeated against the original source.
• Source data can be revised after retrieval. A repeated query may return newer figures; the retrieval time tells you which release an artifact reflects.
• Shared links are snapshots: the data in a shared artifact does not change later.

• The live map combines feeds of very different reliability — official alert and earthquake feeds, Earth-observation signals, curated conflict data, and events extracted from news, video or social sources.
• Each event carries a quality label based on its source and method class (for example “Official structured” or a reporting-derived signal), plus a caveat describing what it can and cannot show.
• That label is not a fact-check of the individual event. Treat reporting- and social-derived events as signals to corroborate, not as confirmed incidents.

• Atlas uses a language model to plan, call tools and write the answer. It can misread a question, choose a less suitable source or summarise a result incorrectly.
• Always check important figures against the linked source before you publish or act on them. The source link, not the summary, is the authority.
• Sources can be unavailable, rate-limited or slow; when that happens Atlas should say so rather than fill the gap, and the answer may be incomplete.

What Humaps stores about your questions and account, and for how long, is described in the privacy policy. More about the project is on the about page.