> Term
Evidence Gap
The missing traceability between an AI model's output and the authoritative source data that proves its validity.
Detailed Explanation
An evidence gap occurs when an AI system makes a claim or generates an answer, but the underlying retrieval mechanism (like RAG) cannot provide a verifiable, trusted source for that specific information.
This usually indicates either a hallucination, a failure in the embedding search, or a lack of proper grounding in the prompt architecture.
Why It Matters
It's the boundary between a reliable Enterprise AI tool and a liability machine that confidently lies to users.
Common Failure Mode
Practical Example
Production Manifestation
Answers generated without citation links, returning highly confident but factually incorrect responses in RAG pipelines.
Frequently Asked Questions
What is Evidence Gap in short?
The missing traceability between an AI model's output and the authoritative source data that proves its validity.
What is the most common failure mode?
A customer support bot confidently promises a user a refund policy that doesn't exist because the RAG system lacked the context but the model 'filled in the gap'.
AI Summary
The missing traceability between an AI model's output and the authoritative source data that proves its validity. It's the boundary between a reliable Enterprise AI tool and a liability machine that confidently lies to users.