Sample approval before mass production: part 16

Key takeaways

1. Size pilot verification sets so they can actually expose the defect rates you cannot tolerate, following the reasoning behind acceptance sampling methods^[1].
2. Define acceptance numbers for the verification set before it is drawn, so results are judged against prior limits rather than negotiated afterward.
3. Require root-cause analysis with dated corrective actions for every significant approval finding, not just rework of the reviewed units.
4. Run approval findings through a plan-do-check-act routine so corrections become process changes, the operating rhythm of recognized quality management practice^[2].
5. Re-verify corrections on fresh output before releasing bulk, because an unverified corrective claim is only a promise.

Sample approval often fails in a quiet way: the round ends with a list of comments, the factory says everything is understood, bulk starts, and the same issues reappear at final inspection. The comments were treated as remarks about the sample rather than as symptoms of the process. Without a corrective routine, approval feedback evaporates the moment production pressure begins, and the buyer pays for the same lesson twice.

A second weakness is statistical. Approval teams review a handful of units and declare the design proven, but small sets carry wide uncertainty: a serious underlying defect rate can easily hide in a tiny verification. Understanding, even qualitatively, how sample size relates to detection power changes how buyers scope the approval round. It replaces the question that feels thorough with the question that is thorough.

Part sixteen of this series addresses both weaknesses together. It explains how acceptance reasoning applies to approval verification, how to structure corrective action so it survives production pressure, and how to confirm the approval still describes reality once bulk runs. The approach needs no specialized software, only written limits and followed-up records.

Treat approval evidence like acceptance evidence

Every sample reviewed during approval is a small acceptance decision: accept the design, accept the process setup, or require more work. Framing it this way brings a useful discipline, which is deciding the acceptance rule before seeing the results. When the factory knows that more than a defined number of defects in the verification set means another round, preparation improves and arguments about whether the findings matter shrink to almost nothing.

The statistical background of formal acceptance procedures explains why this prior commitment matters and how sample size shapes confidence^[1]. Buyers do not need the mathematics on site; they need its consequences written into the approval plan. Several importers ask UTS to witness the verification review so the count, the findings, and the verdict are recorded independently. A witnessed verdict ends the debate about what was actually seen during approval.

Size verification so it can find trouble

A verification set of five units can only catch defects so frequent that almost any draw would show them; moderate defect rates slip through routinely. Expanding the set to a few dozen units transforms detection power for the same inspection effort scaled modestly. The practical lesson is to agree the verification quantity with the factory before the pilot runs, sized against the defect rate the buyer cannot tolerate rather than against convenience.

Stratify the verification across the sources of variation you care about: different cavities, shifts, material batches, and line positions. A verification set drawn from a single good hour proves that hour was good. Spread the draw across the pilot run and the evidence starts describing the process instead of an anecdote. Record where each verification unit came from so clustering, if it appears, points at its cause.

Agree the acceptance rule in advance

Write the rule as a plain sentence before verification begins: this many units checked, this defect classification applied, this number allowed. Both sides sign it as part of the approval plan. Advance agreement converts the verdict from a judgment call into the execution of a contract, which factories accept far more readily than after-the-fact strictness.

Include the re-verification rule in the same sentence: what happens on failure, how large the next set is, and who bears the cost of repeats caused by factory-side issues. Agreed consequences make the first verification round the serious one instead of a rehearsal.

Route findings through corrective discipline

Post this table in the approval meeting and require every finding to take the right-hand path. The left-hand column is what factories default to under schedule pressure; the approval plan must explicitly forbid it.

Make correction survive production pressure

Corrective action only works when it follows a closed loop: find the cause, change the process, verify the change on fresh output, and record each step with dates and names. This plan-do-check-act rhythm is the operating core of recognized quality management practice, built precisely to convert one-off fixes into lasting process control^[2]. Ask the factory to present corrections in this format rather than as verbal assurances, and reject any corrective report that describes only rework of the reviewed units.

Schedule the verification of corrections before bulk release, not during final inspection. A mid-production check that confirms the fixes on real output costs little and leaves time for a second correction cycle if needed. Corrections first verified at final inspection arrive too late to help the shipment they were meant to protect.

Frequently asked questions