Every mark is one episode in which a state, empire or armed force took territory, moved a population against its will, or held one in place. This page is the long version of what that means and where it fails.
Four kinds of episode: territory seized — annexation, occupation, partition, or a border imposed by force; forced displacement — expulsion, population transfer, ethnic cleansing, or a mass exodus driven by violence; events that are both; and confinement — populations held in place through closed camps, internment systems, blockade, encampment law, or permit and closure regimes. Confinement is the mirror image of the rest: the same control over where a people may live, exercised by preventing movement rather than forcing it.
Inclusion is by scale and significance, so this is a survey of the major cases rather than a complete register. A few entries record something adjacent to a seizure and say so: a territory returned, a secession the parent state contests, a standing claim where nothing changed hands.
Figures before about 1800 rest on a different kind of evidence from figures after it, and the atlas does not pretend otherwise. Assyrian deportation totals come from royal inscriptions written to impress; Roman and medieval chroniclers inflate as a matter of convention; the Mongol tolls that circulated for centuries have been cut by an order of magnitude by modern scholarship. Each pre-modern entry says what its number rests on — a victor's monument, a chronicle, a tax register, archaeology, or a modern demographic reconstruction — and gives the reassessment where one exists. Compare a Bronze Age figure with a UNHCR one only with that difference in mind.
Nothing here sums cleanly. Territory figures overlap between events. Death ranges are built on different bases — direct killing, conflict deaths including combatants, excess mortality including disease and starvation, deaths in transit — which differ by an order of magnitude and must not be added naively. People displaced more than once are counted more than once.
Almost every event here involved violence, so a violence category would tag most of the atlas and stop telling you anything. Violence is recorded instead as three attributes: a death range with the sources for each end and a note on what drives the disagreement; a basis, because a war's conflict deaths and a genocide's direct killing are not the same measurement; and a set of documented forms — mass killing, summary execution, sexual violence, torture, starvation and siege, forced labour, enforced disappearance, abduction or transfer of children, arbitrary detention, destruction of property, cultural destruction, denationalization — recorded only where a named source documents that form for that specific event.
That rule matters both ways. A form missing from an event means no source consulted here documents it, not that it did not happen. A separate marker identifies the minority of events where mass killing was the defining act rather than an accompaniment.
Three hundred thousand people displaced from a population of a million is a different event from three hundred thousand out of a hundred million. Where a defensible denominator exists, each entry carries the population of the specific group or territory the event acted on, with the year and the basis. Where none was defensible the field is empty rather than guessed. A few entries exceed their own baseline — a registered-refugee population that grows by descent, or a count of displacement incidents rather than people — and those are shown as a multiple, with a caveat naming the problem.
Several events here are described in incompatible ways by the parties to them. The summaries set out the competing characterizations and attribute them rather than picking one, and note findings by courts, tribunals, parliaments and commissions as facts about the record — naming the body, and recording rejections by the states concerned. Where you see "described by X as", that attribution is deliberate.
Sources are tiered by publisher — official and treaty bodies, NGO and investigative reporting, scholarly work, reference works, press, and open-edited encyclopedias — and every citation shows its tier. 230 of 585 entries currently rest only on reference, press or open-edited sources. Those entries carry a warning above their source list, and the map's Evidence filter isolates them. This is published rather than hidden because it is the clearest thing a reader should know before relying on a figure, and the clearest thing a contributor could fix.
Country links are derived from where each event's points and mapped movements fall on present-day borders, so a 1945 event in East Prussia links to Poland and Russia — the countries whose territory it now is, not the ones that existed at the time.
Text and data are published under CC BY 4.0. The complete dataset is available as JSON and CSV with every field and source. If you find an error — and in a dataset this size there will be errors — the entry's own sources are the place to start, and corrections are welcome.