Evals And Observability easy
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?
- A Score cases that humans already labeled and confirm the judge agrees with them roughly 80% to 90% of the time.
- B Run the judge on a public benchmark and confirm its own task accuracy is high enough to qualify as a reliable grader.
- C Give the judge a short instruction like 'rate quality' and let it pick its own criteria so you do not bias it.
- D Point the judge at the newest available model alias and let it auto-update so grading quality improves over time.
Answer: A. Score cases that humans already labeled and confirm the judge agrees with them roughly 80% to 90% of the time.
A judge saves human grading cost only if it matches human labels, so calibrate it against a labeled set; 80% to 90% agreement is usually usable. A high benchmark score is tempting evidence, but solving tasks well is not the same as agreeing with your graders on your rubric. The judge also needs an explicit rubric and a pinned model version, because an auto-updating judge shifts scores between runs.
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?
- 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?
- 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?