Hi Elasticsearch team,

With the addition of vector search, Elasticsearch is now a common backend for RAG systems. Developers who wire RAG on top of Elasticsearch often run into subtle failures that are hard to categorize.

I maintain WFGY RAG 16 Problem Map, an open-source diagnostic map for RAG and LLM pipelines.

Repo (MIT):
https://github.com/onestardao/WFGY

Main reference page:
https://github.com/onestardao/WFGY/tree/main/ProblemMap/README.md

WFGY provides:

• A taxonomy of 16 RAG failure types
• A one-page Global Debug Card image
• A triage prompt that helps any strong LLM classify failing runs and suggest structural fixes

This map is already integrated or cited in several RAG frameworks and research projects:

• RAGFlow
• LlamaIndex
• ToolUniverse (Harvard MIMS Lab)
• Rankify (University of Innsbruck)
• Multimodal RAG Survey (QCRI LLM Lab)
• And curated lists such as Awesome LLM Apps and Awesome Data Science – academic

Proposal:

Add a small RAG failure guide for Elasticsearch vector search that uses the WFGY 16-problem map. For example:

1. A doc that:

   • Describes common failure situations for RAG on Elasticsearch (mapping choices, vector fields, retrieval vs ranking).
   • Maps them to the 16 WFGY categories.
   • Links to the Global Debug Card so users can drive a diagnosis with an LLM.

2. An example that:

   • Sets up a simple vector based RAG application on Elasticsearch.
   • Introduces misconfigurations and records failing queries plus retrieved docs.
   • Uses the WFGY triage prompt to label each failure and walk through fixes.

This would provide Elasticsearch users with a practical diagnostic lens for RAG, and complement the existing vector search documentation.

If this sounds interesting I can draft a first version of the guide as a PR.