A financial services company runs a customer-support assistant on a Bedrock Knowledge Base backed by Amazon OpenSearch Serverless. Support agents complain that when they ask questions about very niche products, the assistant frequently returns confident answers built from chunks that are only loosely related to the question, producing hallucinated guidance. When the same query has no truly relevant documents, the developers want the system to return no supporting passages rather than the closest-but-irrelevant ones. Which change to the retrieval configuration best addresses this requirement while keeping the managed RetrieveAndGenerate flow?