GLOBAL CONFLICT INTELLIGENCE
Where the world is on fire, updated every three hours.
The Conflict Map turns thousands of news-derived events into a country-by-country picture of armed conflict, with honest limits on what the data can and cannot tell you.
THE PIPELINE
From raw news events to a shaded map
Five automated steps run on every cycle.
Pull the latest events
The bot downloads the most recent window of global events from GDELT 2.0, an open dataset that refreshes every 15 minutes with no usage restrictions.
Keep only physical conflict
Only CAMEO root codes 18 Assault, 19 Fight and 20 Mass violence are kept. Everything else (diplomacy, protests, statements) is dropped.
Add fatality data when available
When an ACLED connection is configured, its recorded events are merged in. ACLED is the only layer that carries fatality estimates.
De-duplicate and store
Events are normalised and de-duplicated by source and ID so the same incident is not counted twice across runs.
Recompute country statistics
Rolling 7, 30 and 90-day counts, an intensity score for each country and a global summary row are refreshed. The map reads these, not the raw feed.
KEEPING IT A WAR MAP
Three filters between the feed and the map
News-event codes for 'assault' and 'fight' also cover ordinary crime and policing. These filters keep that noise off.
READING THE SHADING
Intensity is relative, not absolute
A country scores 100 only because it has the most recorded events right now. When another country overtakes it, scores shift for everyone. A country is flagged active once it passes 5 events in 30 days.
BE CLEAR-EYED
What the map shows, and what it never will
It gives you
- Where qualifying events happened
- How frequently, over 7, 30 and 90 days
- Fatality estimates, where ACLED data is connected
- A relative intensity ranking between countries
- A separate verified-intel layer that is curated by hand
It never invents
- Wounded counts
- Property-damage dollar values
- Territory-control claims
- Anything marked verified that a person has not checked
- Precision beyond what the underlying news data supports
HOW THE DATA IS COLLECTED
Open data, processed by rules
Unlike the other features, the Conflict Map doesn't use a language model to write anything. It's a data pipeline with fixed rules.
WHERE THE DATA COMES FROM
GDELT 2.0
An open database of news-derived events, updated every 15 minutes. The main source for locations and frequency.
ACLED (optional)
Recorded conflict events with fatality estimates, merged in when connected.
Verified intel layer
A separate table curated by a person for details automated sources can't provide reliably.
WHAT THE ANALYST DOES, STEP BY STEP
- 1
Download the latest window
Every 3 hours the latest window of GDELT events is pulled.
- 2
Keep conflict codes
Only Assault (18), Fight (19) and Mass violence (20) events remain.
- 3
Require an armed actor
At least one side must be a military, rebel, insurgent, separatist or other armed group.
- 4
Remove noise
Countries where those codes mostly reflect domestic crime and policing are excluded, and duplicates are dropped.
- 5
Recompute the statistics
7, 30 and 90-day counts, intensity scores and the global summary are refreshed for the map.
No invented numbers
Because the map is rule-based, it can only report what the sources contain. Wounded counts, damage values and territory control are never estimated.
Limitations
- The source is news coverage. Under-reported regions can look calmer than they are, and heavily covered ones can look worse.
- Automated location and actor tagging can be wrong. A handful of events may be mis-assigned or duplicated.
- Event counts measure how often incidents are reported, not how severe they are.
- The map is context for risk awareness, not a trading signal. See the Risk Disclosure.
QUESTIONS
Conflict Map, answered
A world map of recent armed-conflict activity. Countries are shaded by how many conflict events were recorded in the last 30 days, and a corner panel summarises the global picture.
Event data comes from GDELT 2.0, an open global news-event database updated every 15 minutes. If configured, ACLED data is added, which includes fatality estimates. A separate, manually curated layer holds verified intelligence.
The collection job runs every 3 hours. Each run pulls the latest window of events, refreshes country statistics and updates the summary panel.
Only where an event source provides fatality estimates, such as ACLED. The map deliberately does not invent wounded counts, damage values or territory-control claims, because no free automated source provides these reliably worldwide.
Events coded as Assault, Fight or Mass violence in the CAMEO event taxonomy, where at least one side is tagged as a military, rebel, insurgent, separatist or other armed actor.
Countries where those event codes mostly capture domestic crime and policing rather than armed conflict (currently the US, France and the UK) are excluded to keep ordinary incidents off a war map.
It is a relative measure from 0 to 100: a country's 30-day event count divided by the highest 30-day count of any country, multiplied by 100. It ranks countries against each other; it is not a measure of severity or casualties.
A country is flagged active when it has recorded at least 5 qualifying events in the last 30 days.
It is context, not a signal. News-derived data can be wrong, late or duplicated. Always read the Risk Disclosure.
No. The map is built by an automated, rule-based pipeline on open data. No language model writes or estimates anything on it, which is why it can't invent figures.
The Global Database of Events, Language and Tone: an open dataset that codes events reported in world news, updated every 15 minutes.
The Armed Conflict Location & Event Data project, a research dataset of recorded conflict events that includes fatality estimates. It is used only when connected.
A separate layer that is curated by a person rather than collected automatically, used for details no automated source provides reliably.