protologue

Corrective RAG

Also called CRAG.

Corrective RAG (CRAG) adds a lightweight evaluator that grades retrieved documents for relevance and, depending on the grade, uses them, discards them in favor of web search, or combines both before generation.

Description

It targets the failure where poor retrieval silently produces a confident wrong answer. Retrieved documents are also decomposed into smaller strips and filtered so that only relevant parts reach the generator. The method plugs into existing RAG pipelines without retraining the generator.

Sources

  1. Yan et al. (2024). Corrective Retrieval Augmented Generation.

Cite this entry

Protologue. (2026). Corrective RAG. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0135). https://protologue.com/t/corrective-rag/

BibTeX
@misc{protologue_corrective_rag,
  title = {Corrective RAG},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0135},
  url = {https://protologue.com/t/corrective-rag/}
}

Markdown JSON