A cluster is a region in vector space where semantically similar texts are grouped. Content about the same topic occupies nearby positions; content about different topics occupies distant positions. Retrieval finds the content nearest to a query’s position in this space.
Understanding clusters explains why mixed-topic content retrieves poorly for everything. A passage that splits its meaning between two topics lands between two clusters — close to neither one. A passage that stays within one cluster retrieves consistently for all queries that land in that region. The structural instruction — one topic per section — is a geometric instruction: stay inside one cluster.
