Methodology
GlobeAlert does not write news. It collects open sources, groups and scores them, and puts them on a map. This page explains every step, what the numbers mean and where the limits are, so you can judge what you see on the desk.
The threat scale
Every story and event carries one level. The colour is used everywhere: feed, map, country reports, alerts.
Routine reporting with no safety or security impact: business, science, culture, sport.
Notable but contained: policy, diplomacy, local incidents without casualties.
Developing situations with potential impact: protests, cyber incidents, moderate natural events, political crises.
Significant harm or escalation: strikes, attacks with casualties, major disasters, infrastructure outages, market shocks.
Wars, invasions, mass-casualty events, major escalations between states, large-scale disasters. Critical stories drive push alerts and the Critical Desk.
Categories
A story gets exactly one category. Conflict covers armed violence between organised groups or states; Military covers forces, exercises and procurement without active combat; Disaster covers natural hazards; Cyber covers attacks on systems and infrastructure. When two apply, the one with the higher safety impact wins.
From source to desk
1. Collection
Thousands of public RSS and API sources are polled continuously. Every item keeps its publisher, original URL, publication time and language. Geolocated events (earthquakes, wildfires, floods, storms, conflict incidents, crime reports, cyber incidents, launches) arrive from dedicated open feeds with coordinates.
2. Deduplication and cross-checking
Stories about the same event from different publishers are grouped. The number of independent sources reporting it is stored as the source count and shown on the story. A single-source story is labelled as such.
3. Classification
A language model assigns each story one category and a threat level from 0 to 4, using severity, scale, escalation potential and source agreement. Events from sensor feeds are scored from their own magnitudes (for example earthquake magnitude, wildfire radiative power, storm category).
4. Summaries
Summaries are machine-generated from the publisher text, kept short and always shown under the original headline with a link to the source. They can be wrong. The source is the record; the summary is an index card.
5. Country and theater scoring
Country reports and theater status weight the last seven days of events by threat level and produce an index from 0 to 100 and a trend (rising, stable, falling) by comparing the last 24 hours with the 24 before. The same weights drive the map legend.
6. Review and correction
Misclassifications and wrong summaries reported through the contact form are checked by a person and corrected on the page. Publishers can ask for a story to be removed at any time.
Limits you should know
- Classification is automatic and can be wrong, especially in the first minutes of a breaking story. Levels are revised as more sources report.
- Source feeds have their own delays and gaps. A quiet map can mean a quiet world or a quiet feed.
- Geolocation of news stories is at country or region level; only sensor events carry precise coordinates.
- Scheduled events (for example rocket launches) are listed with their planned time and can appear ahead of today.
- GlobeAlert is not an emergency, safety or official government service. See the terms of use.
Questions about a specific story or score: contact the desk. Who we are: about GlobeAlert.