Revision History

Bring the RAG survey current through July 2026: add retrieval quality in practice (hybrid fusion, reranking, late interaction), iterative and agentic retrieval, graph and structured retrieval, and a new top-level section on the long-context question presenting both sides with the effective-versus-advertised context literature; extend evaluation with shared tasks, LLM-judge limits and contamination, and security with indirect prompt injection and access-control leakage

Correction: replace a volatile vendor catalog and universal tuning claims with a source-bounded account of RAG architecture, design, evaluation, failure modes, long-context tradeoffs, and security.

Added 6 contextual internal links

Auto-saved before update

Auto-saved before update

Pre-update snapshot before enhancement to 5726 words.

Auto-saved before bulk wikilink fix

Auto-saved before bulk wikilink fix

Auto-saved before high-confidence wikilink rewrite cleanup

Auto-saved before wikilink fix

Auto-saved before wikilink update

No summary provided