Clusters
A cluster groups events that appear to describe the same developing situation or a closely related set of developments.
Clusters prevent the user from mistaking repeated coverage for multiple independent events and make cross-source context easier to inspect.
How clustering works
Section titled “How clustering works”The system uses a stable, idempotent grouping process. Important inputs include:
- canonical entity;
- source family and source diversity;
- event timing;
- normalized content and event type;
- relationship to existing active clusters;
- corroboration and similarity signals.
Persistent derived clusters generally require more than one related event. Entity-less or isolated observations are treated cautiously. In real time, the interface can temporarily represent a new event while the backend waits for corroboration or the next derivation cycle.
Cluster fields
Section titled “Cluster fields”A cluster can expose:
- title;
- entities;
- member event count;
- first and last event times;
- source breakdown;
- provider or source diversity;
- maximum importance;
- average or aggregate confidence;
- recent event identifiers.
Cluster importance
Section titled “Cluster importance”Cluster importance does not simply add every member score. It preserves the strongest material event and can increase when corroboration, velocity, or breadth justifies it. Repetition from weak sources should not make a minor event dominant.
Cluster confidence
Section titled “Cluster confidence”Aggregate confidence reflects the quality and agreement of the evidence. A cluster supported by independent primary, market, and on-chain evidence is different from one built from several related social posts.
Cluster lifecycle
Section titled “Cluster lifecycle”A cluster can:
- begin with a new event;
- gain related member events;
- expand across source families;
- increase or decrease in relevance;
- contribute to an emerging narrative;
- become the subject of a report or Deep Research run;
- stop receiving evidence and become historical context.
Where clusters appear
Section titled “Where clusters appear”Live Pulse and Social Radar provide cluster views and member-event inspectors. Event Explorer links an event to its cluster. Deep Research can use a cluster as a research scope. Alerts and automatic watchlist research can respond to significant new clusters.