Evals And Observability easy
A new prompt improved your offline eval score by 6 points. After a week in an A/B test, task completion and retry rates show no difference. Which signal should drive the decision?
- A The offline score, because it is measured on a curated golden set and is less noisy than real user behavior.
- B The online metrics, which reflect real users; then find out why the offline set missed what users value.
- C Neither; the A/B test ran too long, so restart it and stop as soon as one variant appears to be ahead.
- D The judge's verdict, since an LLM judge that saw both prompts gives a more objective view than user behavior.
Answer: B. The online metrics, which reflect real users; then find out why the offline set missed what users value.
Online metrics are the ground truth for product quality but arrive slowly. Offline evals are fast and repeatable but only approximate what users care about. Trusting the cleaner offline number is tempting, but a gap like this means the eval set is out of step with users and should be updated. Offline scores should lead only when online traffic is too small to give a meaningful signal.
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?