Evals And Observability easy
Your RAG assistant's answer-quality score dropped from 82 to 70 after a change. You only measure final answer quality. Why is that a problem for debugging?
- A Final answer quality is an online metric, so it cannot be measured offline against a fixed golden dataset.
- B Answer quality should be measured only through latency and token cost, which directly reflect retrieval health.
- C A 12-point drop is within normal judge noise, so separate metrics would not change how you would debug it.
- D You cannot tell whether retrieval returned the wrong documents or generation misused the right ones.
Answer: D. You cannot tell whether retrieval returned the wrong documents or generation misused the right ones.
Separate retrieval metrics (such as recall of relevant chunks) from generation metrics (such as faithfulness) so you can localize a regression; the cost is labeling which documents are relevant for each case. Writing off a 12-point drop as noise is tempting, but run-to-run variance is usually a point or two. A single end-to-end score is enough only when you just need a ship/no-ship signal and do not need to diagnose the cause.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- In one release, a RAG assistant got a new embedding model and a new chunking strategy, and the end-to-end judge score dropped 6 points. The team only tracks the end-to-end score and has spent a week tweaking the generation prompt with no recovery. What is the best next step?
- Over two months, your team iterated through 40 prompt versions, each tuned against the same 150-case golden set. The offline score rose from 72 to 91. Online task completion and retry rate have not changed. What is the most likely problem, and what is the fix?
- A prompt change produced an impressive answer in a live demo for leadership. The PM wants to ship it today. What should you do?
- You plan to replace human graders with an LLM judge for your answer-quality eval. Before trusting its scores in a CI gate, what should you do first?
- After you swap to a smaller model, p95 latency falls from 900 ms to 350 ms and the ops dashboard is all green. Leadership asks whether answer quality is still fine. What is the right response?
- Your CI gate fails any prompt change that drops the score 3 or more points on a 100-point rubric over 60 cases. Production generates at temperature 0.7. Running the eval twice on the same commit gives scores that differ by up to 4 points, and engineers have started rerunning failed gates until they pass. How do you fix the gate?