Evals And Observability easy
Your app collects thumbs-up and thumbs-down on answers. Thousands of thumbs-down accumulate each month, but nobody looks at them. What is the most valuable use of this data?
- A Review a sample of thumbs-down cases, label the failures, and add representative ones to the offline eval set.
- B Report the monthly thumbs-down rate on the dashboard as the single quality number, with no further analysis.
- C Fine-tune the model directly on every thumbs-down output as a negative example, skipping any human review step.
- D Delete the feedback after 30 days, since online signals lag too much to inform day-to-day prompt changes.
Answer: A. Review a sample of thumbs-down cases, label the failures, and add representative ones to the offline eval set.
Turning sampled user failures into eval cases costs review time, but it makes every future change get tested against real failure modes. Training on raw thumbs-down is tempting because it is automatic, but the feedback is noisy: users downvote for tone, latency, or correct but unwelcome answers. The aggregate rate is worth tracking as a trend, but it cannot tell you what to fix.
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?