Evals And Observability medium
You compare two prompt variants with an LLM judge that assigns absolute 1 to 5 scores. Prompt 1 averages 4.1 and Prompt 2 averages 4.2, and individual case scores flip between runs. What is the better way to decide between them?
- A Move to a 1 to 100 scale so the judge has enough resolution to express the small quality differences between the two prompt variants.
- B Average absolute scores over five judge runs per case, and if both still round to about 4, treat the variants as equivalent and ship either.
- C Have the judge pick the better output per case, randomizing order, and report win rate; relative judgments are more stable than absolute ones.
- D Drop the judge and compare mean output length and latency instead, since these are deterministic and correlate with perceived answer quality.
Answer: C. Have the judge pick the better output per case, randomizing order, and report win rate; relative judgments are more stable than absolute ones.
**Pairwise comparison** is more reliable than absolute scoring because judges, like humans, are better at saying which of two answers is better than at placing one answer on a fixed scale. The tradeoff is that you get a ranking of the variants but no absolute quality level. A finer scale (A) adds false precision and usually more noise. Averaging repeated absolute scores (B) reduces variance but can still hide a consistent preference that a head-to-head comparison would reveal.
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?