RAG & Retrieval · Ground answers in evidence
Faithfulness vs answer relevance vs context precision?
Short answer
Faithfulness asks whether claims are supported by context; answer relevance asks whether the answer addresses the question; context precision asks how much retrieved context was useful. They fail independently.
A faithful answer can still be irrelevant if it accurately describes the wrong passage. A relevant answer can be unfaithful if it answers the question from memory. Context precision measures the retrieval side: how much of what you packed was actually useful. Reporting one number hides these distinctions.
- High faithfulness, low relevance: the model used the evidence but answered a different question.
- High relevance, low faithfulness: the answer sounds right but is unsupported.
- High recall, low context precision: the right passage arrived with a lot of noise.
- Low recall makes both generation metrics unreliable, because the evidence is missing.
Follow-ups you should expect
Interviewers use these to check whether the first answer was understanding or memorisation.
- 1Which metric would you look at first after a bad answer?
- 2How would you measure faithfulness without a human label?
- 3Why can context precision be high while recall is low?
Say it in the interview
They measure different failures. Faithfulness is whether the claims are supported by the context. Relevance is whether the answer addresses the question. Context precision is how much of the retrieved context was actually useful. An answer can be faithful but irrelevant, or relevant but unsupported. I diagnose in order: recall first, then precision, then faithfulness, then relevance, because a downstream metric on top of missing evidence is misleading.
Use this as a model, not a script. Saying it in your own words is what makes it survive follow-up pressure.
Related questions
Reading is step one
Now say it out loud.
Understanding an answer and delivering it under pressure are different skills. Rehearse this topic in a live RAG & Retrieval mock interview and get scored feedback on depth, clarity and confidence.