UCDP builds its record from publicly available sources — news agencies, reports from international and non-governmental organizations, and other documentation — which trained coders read and cross-check against a fixed set of rules. The aim is consistency across decades and regions, so that a fatality recorded in one place is counted on the same basis as one recorded somewhere else.
Each recorded event is geocoded: it is assigned to an administrative area and a date rather than to a precise location. A point on a map here represents the area a researcher could confidently attribute an event to, not an exact coordinate. This keeps the data honest about what the underlying reports actually support, and it is why several events can share the same place marker.
There is an unavoidable lag between something happening and it appearing as a confirmed record. Reports take time to surface, corroborate and code; figures are revised as better information arrives, and annual datasets are the most settled. For this reason the data is best read as a historical and statistical record that firms up over time, not as a live feed of current events.
Other datasets exist and are widely used — ACLED is a prominent example — and they make different choices about scope, sourcing and units, so their totals will not match UCDP’s. ConflictWatch uses UCDP because its definitions are stable, well documented and openly licensed for redistribution. Comparing across datasets means comparing methodologies first, not just headline numbers.