Reference · Glossary
Re-ranking
A second pass that **re-scores retrieved chunks** with a smarter (often slower) model so the best passages rise to the top before generation.
When to use
RAG quality is "almost right" — retrieval returns relevant docs but in wrong order, or noisy neighbors slip in.
When not to
Tiny doc sets where vector search already returns perfect top-3 results.
Example
1. Vector search returns 20 chunks
2. Cross-encoder re-ranker scores each against the question
3. Top 5 re-ranked chunks go into the prompt