Auditing a Citation System That 'Already Has Citations'

Your RAG assistant already shows numbered citations under every answer, and the team considers hallucination "solved" because of it. A customer reports an answer that cited [2] for a claim, and when they clicked through, the linked article said nothing resembling the claim.

1. Explain precisely why "we show citations" did not solve hallucination here, and name the two distinct things a citation system needs to guarantee.
2. Design the minimal check that would have caught this specific failure before the answer reached the customer, and say what it costs.
3. What's the next layer up, and why is it meaningfully more expensive than the check in (2)?