Query-Time Filtering

Query-time filtering evaluates each candidate result against the requester's entitlements at the moment a question is asked. It is the most common pattern because it can be added to an existing index without reprocessing content, and because it appears to produce the correct outcome. Its weakness is that correctness depends on the filter covering every query formulation, similarity threshold and retrieval mode the system will ever support — a coverage obligation that grows each time a new interface, agent or integration is added. Filtering also cannot undo embedding: the semantic signal of restricted content remains in the vector space and can influence ranking even when the record itself is withheld.

Centralpoint treats query-time filtering as a second line rather than the primary one. Audience and role evaluation still occurs at request time, but it operates over an index from which restricted material was already excluded during ingestion, so a filter failure degrades to showing less rather than to exposing more.