The Directness-Comprehensiveness Tradeoff

Vector search and structured retrieval optimize for fundamentally different response qualities.

In comparative evaluation, vector RAG won 92% of comparisons on “directness” (giving specific, targeted answers to questions. But it lost 88% of comparisons on “comprehensiveness” (covering the full scope of what a question touches.

Structured approaches (knowledge graphs, ontologies) flip this pattern: they excel at synthesizing information across sources but often bury the specific answer inside broader context.

This isn’t a matter of one being better. It’s a matter of query shape. “What’s the setting for X?” wants directness. “What factors affect Y across our documentation?” wants comprehensiveness.

The implication: retrieval architecture should match query type, not be chosen once and applied universally.

Related: 07-molecule—retrieval-architecture-spectrum, 07-molecule—vectors-vs-graphs