Vector RAG Directness Win Rate: 92%

In head-to-head evaluation against ontology-based retrieval, simple vector semantic search won 92% of comparisons on the “directness” metric (p < 0.001).

Directness measured how specifically and directly the response addressed the question asked.

The same vector approach lost 88% on comprehensiveness and 86% on diversity.

This asymmetry suggests vector search remains the better choice for targeted, single-answer questions, despite its weaknesses on synthesis tasks.

Related: 07-atom—directness-comprehensiveness-tradeoff, 07-molecule—retrieval-architecture-spectrum