A clean sample file can hide a dirty population. If your selections repeatedly land on low-risk, well-documented items while known problem vendors sit outside the frame, the design is comforting the team — not testing integrity.
Warning signs
- Population definitions change after selections are drawn
- High-risk strata are “too small to sample” every quarter
- Exceptions are cleared with oral explanations and no ticket trail
- Completeness of the population file is assumed, never tested
Reset by rewriting the population narrative first, then choosing method. Sometimes the honest answer is 100% testing for a week of postings or a vendor class. That is not failure; it is proportion.
Practice this reset in Sampling for Assurance.