Design an unbiased prevalence estimator for a rare, activity-based harm

Measurement & estimationHard

Problem. You own measurement for a severe harm vertical where violations are rare (well under 1% of activity) and the "unit" is an account's behavior over a day, not a single piece of content. Human labeling is expensive. Design a system that produces a trustworthy daily prevalence number.

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.

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