Uniform, PPS, or ML-assisted sampling — and what does it cost you?

Measurement & estimationHard

Problem. You can spend a fixed daily label budget. A teammate wants plain uniform sampling "so the math is simple." When is impression-weighted or ML-assisted probability sampling actually better, and what's the hidden cost of leaning on it too hard?

Before you reveal: say your answer out loud, as if you were in the real interview — get your reasoning across clearly first. There is no single correct answer: reading what the interviewer is really after and defending your own thinking is what makes an answer strong.

← Design an unbiased prevalence estimator for a rare, activity-based harmYour LLM labeler isn't perfect — how does that bias the number, and how do you fix it? →