About
Global Crisis Monitor is a personal, artistic project. It was built in a period of growing conflict — a time when wars that once felt distant became part of everyday conversation, appearing in feeds and notifications alongside everything else.
There is something disorienting about that: a bombing in a city you can name, a ceasefire that collapsed overnight, a famine declared — and then, scrolling past it, an advertisement. The architecture of attention flattens everything into the same urgency and the same forgettability. This project is a refusal of that flattening. It is an attempt to make the weight visible: to collect the signals, hold them together on a single surface, and sit with what the picture looks like.
It is not an answer. It is not even a good map. It is the act of not looking away — run as software, updated continuously, rendered in the dark.
The monitor aggregates RSS feeds from over 80 international news outlets and research organisations — from wire services and quality press to specialist defence journals, regional outlets, and cybersecurity publications. Articles are fetched continuously, processed by an AI model, and grouped into unified events. Each event is geolocated on a map, assigned an estimated impact score (0–100), and stored alongside the key entities — people, locations, organisations — that appear in the coverage.
How the feed works
The monitor operates on a 48-hour rolling window. Events from the last 24 hours are shown as live; events from the previous 24 hours appear in the archive. Nothing older is displayed — the map shows what the feeds are saying now, not a historical record.
Flash events — those with an impact score of 80 or above that were reported within the last six hours — receive visual priority with pulsing indicators on the map and feed. These represent the highest-severity, most recent signals the system has detected.
Impact score
Every event is assigned an impact score from 0 to 100 by the AI model, based on estimated humanitarian severity, military implications, and geopolitical significance. Scores are classified as: Low (0–39), Elevated (40–59), High (60–79), and Critical (80–100). The Global Threat Index displayed in the header is the arithmetic mean of all live event scores.
Entity context
Clicking on any event in the feed opens an Intelligence Brief with enriched context for the people, organisations, and locations mentioned. This data is drawn from Wikipedia, Wikidata, REST Countries, World Bank Open Data, and ReliefWeb — all free and open APIs. Country cards show population, area, GDP per capita, military expenditure, and active humanitarian disasters. Person and organisation cards show founding dates, membership, and biographical extracts.
Other data layers
The dashboard includes a weather widget showing conditions at the selected event location, and a Doomsday Clock indicator sourced live from the Bulletin of the Atomic Scientists — both serving as ambient context for the events displayed. The left panel tracks the most frequently mentioned personalities, locations, and organisations across all visible events, making cross-event patterns visible at a glance.
Source credibility
Sources are not treated equally. Each feed carries a credibility weight between 0 and 1. Wire services, UN bodies, and established research institutes sit near the top (0.90–0.95). Regional outlets and specialist blogs sit lower. State-affiliated sources (Xinhua, TASS) are included for completeness but carry the lowest weights (0.60–0.65). The full list is on the Sources page.
Limitations
This is not a newsroom. It is not a verification service. The pipeline is fully automated: an AI reads the headlines, clusters what seems related, invents a short synthesis, and scores the severity. The model will hallucinate, conflate, or miss context. Some conflicts are systematically underreported and will appear as quiet zones on the map despite ongoing violence. The aesthetic of precision — the coordinates, the percentages, the amber glow — should not be mistaken for accuracy.
What you are looking at is a meta-analysis constructed from public information. It reflects what over 80 sources chose to publish in a given window, filtered through an AI with its own biases and gaps. It is a representation, not a record. It cannot and should not be used as a journalistic source, cited as evidence, or treated as ground truth.
The impulse behind it is something closer to cartography as witness. Maps have always been partial — shaped by who drew them, what they chose to include, what they left out. This is no different. But the act of drawing, of aggregating, of refusing to look away, still feels like it means something.
Built with Next.js, Prisma, Leaflet, and OpenAI GPT-4o-mini for event synthesis. Context panels draw on Wikipedia, Wikidata, REST Countries, World Bank, and ReliefWeb — all free and open APIs. Weather data from Open-Meteo. Doomsday Clock from the Bulletin of the Atomic Scientists.