About & methodology

How it works

1. On a schedule (currently daily, possibly every few hours), public data is fetched from independent sources (see below).
2. The raw data is stored, then condensed into a short structured brief.
3. One AI model call turns that brief into the headline, the per-category assessments and a list of key events — each event cites the source line it is based on and is linked to the original report.
4. The result is stored and shown here, together with the places each event happened in (searchable), the weather of major world cities and a map of recent earthquakes.

Data sources

• Earthquakes: USGS, EMSC
• Weather: Open-Meteo, NOAA National Hurricane Center
• Conflict: GDELT, UN News, International Crisis Group, US State Dept and UK FCDO travel advisories
• Health: WHO Disease Outbreak News, ECDC
• Cyber: CISA Known Exploited Vulnerabilities
• Disasters: GDACS

Each digest shows which of these responded on the day. Full list with links to each provider: Sources. A category shows “No data” when none of its sources did — it is never silently treated as calm.

How to read the numbers

• Risk (0–100): a judgement of current conditions in the category — Low below 25, Moderate below 50, Elevated below 75, Critical above.
• “% chance in 7 days”: the model’s rough estimate that a notable event happens within a week. It is not a forecast or a calibrated probability.
• Overall: simply the highest risk among the categories that had data.
• Accuracy: after 7 days each “% chance” is checked against what was reported — see Forecast accuracy.
• Trend: the same risk level across previous digests — see Trends.

Limitations

• The summaries are AI-generated from a limited set of public feeds; they can be incomplete or wrong. Always follow the links to the original reports.
• Coverage is uneven: some regions and topics are reported far more than others, and that shapes the picture.
• This is an information aid, not an emergency or safety service. For urgent risk, use your local authorities.