Methodology
How we put crisis data together
This page explains, in plain language, how our content pipeline works — and, just as importantly, what it cannot guarantee today.
The pipeline
From raw research to published page, every crisis passes through six stages.
Discovery
An AI-assisted research pipeline (using the Perplexity model) looks for emerging and ongoing crises in recent reporting. It is instructed to prioritize under-covered regions and slow-burn crises over headline-chasing.
Moderation
Each candidate is moderated automatically against editorial criteria: it must be verifiable from credible reporting, geographically specific, and significant in scale. Candidates that duplicate an existing entry are merged instead of published twice.
Enrichment
Entries are enriched with structured context — affected population, trend, recent developments, and a plain-language summary — plus a representative image where one can be responsibly sourced.
Image validation
Images are checked by a vision model for relevance and appropriateness. Poor matches are automatically retried with better search terms — though a stock photo can never fully represent a real crisis, and we treat it as illustrative only.
Publish
Entries are published with their source citations and the site is rebuilt so every crisis gets a permanent, indexable page.
Weekly refresh
A scheduled job re-evaluates stale entries every week, refreshing data as new information becomes available and retiring issues that are no longer active.
Fairness over sensationalism
Crisis coverage has a systematic bias: wars and famines get documented, while slow-burn disasters — democratic backsliding, economic collapse, chronic pollution, migrant deaths — often go unrecorded. Our discovery pipeline is explicitly built to counter this.
A region × category coverage matrix is analyzed before every discovery run, and the most underserved regions are targeted first. A severe crisis in an uncovered region is preferred over a marginally worse one in a region we already document well. The goal is balanced global coverage, not a highlights reel.
Honest limitations
We would rather tell you what our data cannot do than imply it is more reliable than it is.
What you should know
- • This is a young, independent project run by a small team. Content is drafted with AI assistance and moderated automatically — it is not yet independently verified by a human before publication.
- • We do not yet re-verify each claim against its cited source. Sources are attached to every entry so you can check them yourself, and we treat the cited outlets' reporting as the authority.
- • Affected-population figures are estimates from reporting agencies and should be treated as orders of magnitude, not exact counts.
- • Featured images are stock photography chosen to be topical and appropriate — they are illustrative, not documentary.
- • If you spot an error, please tell us — we correct mistakes quickly and publicly.
Questions?
The FAQ covers the most common questions about how we work.