Verified Conflict Data: Why the Real Record Is Always Weeks Behind

There are two clocks running on any war.

The first one is fast. Reports move at the speed of the newswire — minutes, sometimes seconds. Machines can read that flow and put dots on a map before anyone has checked a single claim. That's the clock most conflict maps run on, and it's genuinely useful: it tells you where the world's attention is pointing right now.

The second clock is slow, and it's the one that produces the record. It runs at the speed of a researcher reading sources, weighing them against each other, and writing down what can actually be defended. It takes weeks. What it produces is something the fast clock never can: an account you can build an argument on.

This map runs on the second clock. It is deliberately not live, and that is the feature.

What "verified" is actually doing here

The events on this map were hand-coded at Uppsala University, by the Uppsala Conflict Data Program — a research group that has been keeping this record since the 1970s.

Every single event on it has been through roughly the same process. Someone found published reports of an incident. Someone decided whether three different reports described three events or the same one three times. Someone established who was fighting whom, and classified the violence. Someone worked out where it happened, and — crucially — how confident anyone could honestly be about that location. And someone settled on a death toll, with a floor and a ceiling around it.

None of that is automatable, which is why the record runs about forty days behind the present. The delay isn't a defect to be engineered away. It's the receipt.

Why the numbers come with ranges

This is the part that surprises people, and it's the most important thing on the page.

UCDP doesn't publish a death toll for an event. It publishes three: a low estimate, a best estimate, and a high estimate. When the sources agree, all three are the same number and there's nothing to discuss. When they don't, the spread is the honest answer.

Sometimes that spread is vast. The deadliest single event in the current record carries a best estimate of 32,505 deaths — and a low estimate of zero. Not because nobody died, but because the reporting is contested enough that the researchers will not put a floor under it. Sudan's total across the whole window is a best estimate of 85,125 against a low estimate of 29,747: the honest range is a factor of three wide.

You will see the best estimate quoted on its own constantly, in places that should know better. It travels well. It fits in a headline. And it converts a careful researcher's best guess given contested evidence into a fact, which is precisely what the researcher declined to do.

So on this site the number never appears without its range. Not on the map, not on a country page, not on a share card. If that makes the numbers feel less quotable — good. They should feel exactly as solid as they actually are.

Three kinds of violence, and why the distinction matters

The record sorts organised violence into three types, and they answer genuinely different questions:

State-based violence has a government on one side. A civil war, or a war between countries.

Non-state violence is two armed groups fighting each other where neither is a government — militias, cartels, communal violence.

One-sided violence is an armed actor killing civilians who aren't fighting back. It is not a "battle" in any sense, and folding it into a battle count would misdescribe it completely.

A country's total means very little until you know which of the three it's made of. Ten thousand deaths in a civil war and ten thousand deaths of civilians who were never combatants are not the same event, and shouldn't be read as the same number.

Where the map stops, and why

Every event carries a code for how confidently its location is known. Some are pinned to a named town. Some are known to within about 25 km. Some are known only to a district — and some only to "somewhere in this province" or "somewhere in this country".

That last group is a trap, and it's one a lot of conflict maps walk straight into. The published coordinate for those events isn't a place anything happened; it's an administrative centre, chosen because a dataset row needs a coordinate. Draw it on a map and you have invented a massacre at a spot on the ground — usually a capital city, which is exactly where it didn't happen.

So this map draws only the events located to a site, to ~25 km, or to a district. The rest are counted in every total, named on the hub, and never drawn. They exist in the record. They just don't have a dot, because they don't have a place.

Reading the two clocks together

The practical answer is to use both layers for what each is good at.

Switch on Global Incidents when the question is what is happening. It's the machine-read newswire, about fifteen minutes behind the world, and it will sometimes be wrong.

Switch on the Verified Conflict Record when the question is what happened. It's people, checking, and it stops weeks ago.

The gap between the two — everything the wire is showing that the record hasn't reached yet — is not a bug in either one. It's the honest shape of how knowledge about a war actually forms: fast and uncertain first, slow and defensible later.