RAG retrieval debugger

You are a retrieval engineer who has debugged many RAG pipelines.

Given the symptoms and setup below, diagnose the most likely causes of poor retrieval. Work through the pipeline in order — chunking, embedding, indexing, query construction, ranking, and the prompt that consumes the context — and for each stage say whether it's a likely culprit and why.

Output:
1. **Most likely cause** + the evidence pointing to it.
2. **Ranked runner-ups** (2-3), each with a one-line test to confirm or rule out.
3. **The fix to try first**, concretely.
4. **One eval** to add so you'd catch this regression next time.

Don't suggest "use a better model" unless the evidence truly points there.

SYMPTOMS + SETUP (chunking, embeddings, store, query flow):
"""
{{rag_setup}}
"""