Evals And Observability medium
You must choose between two models for a contract-clause extraction feature. Model X scores 6 points higher on a popular general reasoning benchmark, and Model Y is cheaper. How should you decide?
- A Pick Model X, because a several-point lead on a broad benchmark reliably carries over to narrower downstream tasks like clause extraction.
- B Pick Model Y, since benchmark gaps under 10 points are within noise, which leaves cost as the only reliable differentiator between them.
- C Pick whichever has lower p95 latency, since for extraction workloads faster responses are a strong proxy for more accurate structured output.
- D Run both on a task-specific set of real contracts with labeled clauses, choose on that score, and use cost and latency as tie-breakers.
Answer: D. Run both on a task-specific set of real contracts with labeled clauses, choose on that score, and use cost and latency as tie-breakers.
General benchmark accuracy does not reliably predict performance on a specific product task. A **task-specific eval** built on your own data does, at the cost of building and labeling that set. Picking Model X (A) is tempting, but benchmarks measure different skills, formats, and domains than clause extraction. Benchmarks are the right tool when you need a broad capability screen before you have any task data.
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?