RAG EngineeringdebuggingIntermediate

The document exists but is never retrieved

Symptoms

  • A user asks about the Q3 refund policy. The exact policy is in the knowledge base, but the agent answers "I could not find that."
  • The retriever returns five chunks, none of them from the refund policy document.
  • Copying a sentence from the document into the query still fails to surface it.
query: "what is the Q3 2024 refund window for enterprise plans?"
retriever: top_k=5, search=vector, similarity=cosine
filter: { tenant_id: "acme", doc_date: { $gte: "2024-10-01" } }
returned docs: [ handbook#12, handbook#41, billing-faq#3, billing-faq#7, onboarding#2 ]
target doc: refund-policy-2024#4  (doc_date: "2024-07-01", tenant_id: "acme")

Investigate

Inspect areas in any order (0/6 inspected). When you think you know the root cause, commit to it.

Metadata filter
Chunking
Embedding model
Search type
Top-k
Reranking