Retrieval-Augmented Generation is often described as a way to reduce hallucination, but retrieval alone does not guarantee grounded answers. A system can retrieve relevant documents and still answer with unsupported claims, mix sources incorrectly, or ignore the evidence when the model's prior knowledge feels more fluent.
Reliable RAG needs a stricter loop: retrieve candidate evidence, judge whether the evidence actually supports the question, constrain the model to answer from those sources, and evaluate the final response against the cited material. This turns retrieval from a context injection trick into an evidence workflow.
The harder design question is not only "did we retrieve something?" It is "should the system answer at all, and can it show why this answer is justified?"