Acceptance sampling was designed for a manufacturing world with defect rates orders of magnitude higher than today's. The mathematics has not changed, but the conditions it assumed have, and the result is that a passing sampling plan now hides precisely the failures that matter most.
The arithmetic of a rare defect
A sampling plan is a statement about the probability of detecting a defect rate, not about the presence of defects. At the rates a well-run modern line achieves, the defects that escape are rare enough that a sample of reasonable size is likely to contain none of them — while the lot still contains several.
Why the escapes are the expensive ones
Systematic defects, the ones caused by a tool wearing or a fixture drifting, produce clusters. Random sampling is well suited to uniformly distributed defects and poorly suited to clustered ones, so the defect mode with the highest cost per escape is the one sampling is least likely to catch.
- A tool drifting produces a run of affected parts, then stops when the tool is changed.
- A fixture loosening affects one cavity or one station, not the population.
- A material lot change affects a contiguous block of production.
In each case the defective parts are adjacent in time. A sampling plan drawn across the lot has a good chance of stepping over the cluster entirely.
What 100% inspection changes
It is not primarily about catching more defects, though it does. It is that inspecting every part turns quality data into a time series. A cluster becomes visible as a cluster, which means the tool change or the fixture problem is detectable while it is happening rather than at the next audit.
Written from deployment experience across multiple sites. Patterns are described without identifying any customer — we are under confidentiality with the operators involved.
Talk to an engineer about this