How the map thinks
Every few hours, World Chaos Map reads the news from around the world, works out what actually happened and where, and turns it into a score from 0 to 100 for each country. This page explains how, in plain words.
The map scores what the news says happened, not how much a country is in the news. A hundred articles about a diplomatic row count for less than one report of an airstrike. Each story is judged on the hard evidence in it:
| Evidence | Examples |
|---|---|
| Deaths and injuries | "3 killed", "dozens injured" |
| Attacks | Airstrikes, shelling, bombings |
| Unrest | Riots, clashes with police, mass arrests, large protests |
| Destruction | Looting, arson |
| Emergencies | Curfews, martial law, coups, internet shutdowns |
| Disasters and outbreaks | Earthquakes, floods, cyclones, epidemics |
Each piece of evidence gives the story a harm value between 0 and 1. Casualty counts are on a logarithmic scale: going from 1 to 10 deaths matters about as much as going from 100 to 1,000, so one huge figure cannot swamp everything else. When a story has several kinds of evidence (an airstrike that also kills people), they reinforce each other the way independent probabilities do, without simply adding up.
The type of event then sets how much that harm matters, from most to least: armed conflict, civil unrest, disasters, health and political crises, crime, and cyber attacks. Harm is recalculated from the stored text on every run, so when the reading rules improve, the whole history is rescored the same way.
Deaths count most. Without a reported death, riots, curfews and mass arrests stay below the level of deadly violence, and a day with no reported deaths cannot reach the top of the scale, however much it is covered.
Plenty of news uses alarming words without anything happening today. Every story is classified as a hard incident, analysis or background, and the map turns these right down:
- History and anniversaries: "10 years on", memorials, documentaries
- Festivals and re-enactments: traditional events that sound violent but are not
- Opinion and commentary: analysis pieces, op-eds, "experts warn"
- Words without events: diplomatic statements, condemnations, threats, "braces for"
- Unconfirmed alarming claims: a coup, chemical weapons or a nuclear strike reported by a single outlet counts for less until a second independent outlet reports it
- Reactions: a leader responding to an event ("X slams Y over the riots") is not the event itself, even when the article repeats what happened
- Other noise: weapons tests, court cases, ordinary accidents
An event counts for the country where it happened, not the country behind it. "Russian strike on Kyiv" counts for Ukraine. Place names that could mean two places, like Georgia the US state and Georgia the country, are handled so events do not land in the wrong country. Place names are read in English, French, Spanish and Portuguese, with or without accents, and "Congo" on its own means the Democratic Republic of the Congo, as it does in most news.
- More than 400 working news feeds across every region, plus official data: earthquakes (USGS), disasters (GDACS), disease outbreaks (WHO), humanitarian reports (ReliefWeb) and recorded deaths in organised violence (UCDP). We test the feeds regularly, and a person reviews every change to the list before it goes live. The full list is on the Sources page.
- Global event data in 100+ languages, so countries with little English-language coverage are not invisible.
- Reliable sources count fully. Each source builds a track record over time. Sources that often fail, repeat others or rarely report real events count for less.
- Each event is counted once. Duplicates are caught at several levels: the same link, near-identical text (fingerprinting), and the same event told in different words or languages (grouped by type, country, date and the people and places named). More independent sources make us more confident it happened, not more alarmed.
- The worst events set the day. Each day gets an intensity from its most serious stories. Lots of minor news cannot add up to more than one day of airstrikes.
- Recent days count most. The score looks at the last 30 days and gives the most weight to the last week.
- Wars stay visible. Deadly armed conflict is remembered for a few weeks, so a war does not fade off the map just because coverage thins out.
- Sudden spikes stand out. Each country is compared with its own normal. When a usually quiet country suddenly erupts, the score jumps, even before the wider world notices.
- Global attention is a small extra signal, never the main driver.
- Everything ends up on one 0 to 100 scale, so countries can be compared directly.
For readers who want the mechanics. The exact tuning values are not published.
| Stage | What it does |
|---|---|
| Story weight | Harm, multiplied by how sure we are the event happened in that country, how many independent sources confirm it, and how widely it is covered. Coverage has diminishing returns, so being in the news a lot adds little. |
| Daily intensity | A saturating curve over the day's stories, heavily weighted to the most serious ones. The second-worst story adds less than the worst, the third less again, so volume cannot buy a high score. A day with no reported deaths is capped below the top. |
| Level | A recency-weighted average of the last 30 days of daily intensity, with exponential decay so last week matters most. Days with no news are filled in rather than skipped, so quiet countries are not judged on a few loud days. |
| Recorded deaths | Deaths in organised violence recorded by the Uppsala Conflict Data Program (UCDP) set a minimum score, the same way for every country, so a war with little news coverage cannot look calm. |
| Conflict memory | Deadly armed violence is remembered for a few weeks and fades slowly. The memory follows both the worst single day and the worst few days together, so a war reported a little at a time stays on the map as well as one in the headlines. A single attack raises a score, but only killing on several days holds it at war level. |
| Surge | Compares the last week with the country's own baseline. A big jump above its normal lifts the score, which is how new crises appear before they become global headlines. |
| Global signal | The share of worldwide event reporting about the country, as a small independent nudge on top. It only applies to countries with at least one news story that week. |
| Final scale | An S-shaped (logistic) curve maps everything to 0 to 100, so the middle of the scale separates countries well and the extremes are hard to reach. |
| Smoothing | Rises are applied in full. Falls are capped per day and scaled to the time since the last update, so recovery is gradual. |
- No country lists. There are no special rules, discounts or caps for any country. The same news about two countries gives the same score, and an automated test checks this on every change.
- Coverage does not equal chaos. Countries that get a lot of media attention are not scored higher just for being talked about.
- Scores can be wrong. They depend on what gets reported. Some places are under-reported, which the Confidence page explains.
- Rules, not an AI model. Scores come from fixed, tested rules applied to the news. No AI model decides a score. We use AI tools to help write code and documentation, and a person reviews every change.
Scores rise as soon as something serious is reported: a war breaking out shows immediately. They fall gradually, by a limited amount each day, so a country does not look calm the moment the headlines move on. Recovery takes days, not hours.
The single number at the top of the map. It is the chaos level the average person lives in: every country's score, weighted by how many people live there. A plain average would let dozens of small, calm countries hide a crisis affecting hundreds of millions of people.
- Same 0 to 100 scale and levels as countries, so 42 means Moderate
- "On last week" compares it with the same calculation using each country's score from 7 days ago
- Countries without a score yet are left out
A score describes recent reported events. It is not a prediction, a risk rating, travel advice or a judgement of a country or its people. For travel, always check your government's official advice.
See it live
Tap any country on the map to see the stories behind its score.