Empowerment Parity Across Retrieval Architectures
Despite large differences in comprehensiveness and directness, vector search, graph-based, and ontology-driven retrieval achieved comparable scores on “empowerment” (the ability to enable users to make informed judgments.
The differences weren’t statistically significant (p > 0.05).
This suggests that more comprehensive answers don’t necessarily produce better-informed users. A direct, specific answer from vector search may be as actionable as a thorough synthesis from structured retrieval.
The implication: comprehensiveness and diversity metrics may overvalue breadth relative to the user’s actual decision-making needs. Getting the right answer quickly might matter more than getting all possible context.