Acceptance Sampling for Parts Orders: Passing a Sample Is Not a Guarantee for Every Part

An order can contain nonconforming parts even when the inspected sample is entirely conforming. Random selection does not guarantee that a small sample includes every condition present in the lot. Acceptance sampling uses an agreed decision rule to manage that uncertainty; it does not remove the uncertainty by calling the result a pass.

That distinction is easy to lose in a receiving record. A sample of undercarriage fasteners may pass a specified thread check, while a summary says only “order accepted.” The shorter statement hides what was checked, how the sample was selected and which decision the result supports. The value of sampling lies in making that bounded decision deliberately, with its risks understood, rather than treating a few satisfactory parts as a miniature guarantee of the entire shipment.

Define the lot, the unit and the characteristic

The lot is the population to which the sampling decision will apply. It must have a defined boundary: for example, identified items of one part and revision supplied under a particular controlled lot identity. A shipment, pallet or purchase order can contain more than one such population. Sharing a delivery date does not establish that every item belongs in the same sampling frame.

Keep supplier identity and the actual lot designation linked to the items being considered. Mixed revisions, unrecorded substitutions or separated stock with lost identity can make it unclear which units the sample represents. Sampling calculations cannot recover a population boundary that was never preserved.

The inspection unit matters too. It might be one bolt, one nut or one defined assembly. A carton containing several assemblies is a packaging unit, not automatically the unit counted by the inspection plan. If a record says “five checked,” it should be possible to tell whether that means five individual items, five packages or five measured features.

Then name the characteristic. Thread acceptability, a specified dimension, identity marking and visible surface condition are different inspection questions. A result for one does not automatically answer the others. In a bolts-and-nuts order, the fact that a sampled thread satisfies its matching gauge rule does not establish material grade, heat treatment or suitability for an unidentified joint.

Nonconforming items and individual nonconformities are different counts

One item can have several nonconformities. A fastener might have both an incorrect marking and a dimensional discrepancy, while another has only the marking discrepancy. A plan counting nonconforming units and a plan counting separate nonconformities would not necessarily produce the same count from those observations.

Use the counting convention specified by the actual plan. Do not decide after seeing the sample which convention gives the more favourable result. The report should identify both the observations and the count used for the decision, so that the result can be reconstructed without guessing how multiple findings on one item were handled.

Why a clean sample can come from a lot containing nonconforming parts

The NIST/SEMATECH description of acceptance sampling starts with a random sample used to support an accept-or-reject decision about a lot. It also distinguishes that purpose from estimating the lot's quality precisely. The inspection observes the selected units; the lot decision necessarily extends beyond those directly observed units.

A simple fictional population makes the limitation concrete. Imagine 100 items, of which two do not conform to one defined characteristic. The full population is known only for this mathematical illustration. A randomly selected sample of ten could include neither nonconforming item, one of them or both. A second independently selected sample could produce a different observed count even though the underlying population has not changed.

For the first draw of ten without replacement, the probability of selecting only conforming items in this invented population is:

(98/100) × (97/99) × … × (89/91) ≈ 0.809

The approximately 80.9% result is not an inspection recommendation. It shows why finding no nonconforming units in that sample is compatible with their presence elsewhere in the lot. The calculation assumes exactly the invented population above, equal random selection and a perfectly effective check of the specified characteristic. It cannot be used to estimate the defect rate of a real order whose full condition is unknown.

An actual lot-acceptance decision also needs the agreed sampling plan and its decision rule. The example does not supply a sample size, acceptance number or AQL for load-carrying hardware. Those choices belong to the applicable quality requirements and responsible technical authority.

Random selection is part of the method

Picking the easiest ten items from the top of one open carton may be convenient, but it need not give every unit in the defined lot a chance of selection. If different packages or production groups have different conditions, a convenient sample can systematically miss part of the population.

The sampling method must match the agreed plan and physical organization of the lot. This does not require inventing a universal method for every delivery. It requires preserving how the actual sample was selected, which population it came from and whether any selection limitation affects the decision.

Replacing an inconvenient sampled item with a better-looking one changes the evidence. So does combining results from unlike parts or revisions after inspection. When the intended plan cannot be followed, the deviation needs to remain visible to the person responsible for disposition rather than being hidden behind the final count.

AQL is not a promise about the contents of one delivery

AQL is a parameter used within a specified acceptance-sampling system. It should not be read as a guarantee that the particular accepted lot contains no more than that percentage of nonconforming items. Nor is it permission for a supplier to place a known number of bad parts in a box.

Sampling systems balance risks in both directions. A lot from an acceptable process can produce an unfavourable sample and be rejected; an unsatisfactory lot can produce a favourable sample and be accepted. The probabilities depend on the plan and the actual quality level. A short “PASS” entry does not make either risk disappear.

IFS's acceptance-sampling documentation describes the relevant concepts within its named software version, including lot decisions and the role of AQL. Its supported standards and implementation limits should not be treated as a current universal inspection rule for every parts order. A software default is not evidence that the same setting is appropriate for an undercarriage application.

The consequence of the characteristic matters. A cosmetic variation and a feature affecting the intended load path may need very different controls. Material verification, critical dimensions and functional requirements can require evidence beyond an appearance sample. The responsible inspection plan should make those distinctions before the order is assessed.

Increasing inspection does not expand what a method can see. Even a thread check on every unit would remain a thread check; it would not become a material test. Likewise, 100% inspection is not a logical guarantee of perfection if the method cannot detect the relevant condition or if inspection errors occur. The appropriate combination of methods matters alongside the number of units inspected.

Keep the observation and the disposition separate

A useful record distinguishes three statements. First, it states what was observed on the identified sample. Second, it records the lot decision under the specified plan. Third, it records any authorized disposition affecting the material. These statements may be related, but they are not synonyms.

For example, “no nonconforming units found for the inspected dimension in the selected sample” is an observation. “Lot accepted under the identified plan” is the resulting decision when that plan supports it. “Released for installation on this machine” would require further authority and applicability evidence; it does not follow automatically from the first two statements.

Preserve any observed nonconforming units and their identities within the applicable quality controls. A lot-level acceptance result should not erase a known individual finding. Equally, a rejected sampling result should retain the actual observations so the responsible person can determine the permitted next action under the governing requirements.

In the invented 100-item population, the two nonconforming items still exist after a clean sample is drawn. The approximately 80.9% chance of that outcome explains why the sample result and the condition of every item are different statements. An agreed plan determines what lot decision may follow from the sample; any nonconforming item actually found retains its own inspection finding and required disposition.

References

• NIST/SEMATECH: what acceptance sampling is intended to decide
• IFS Cloud 25R2: acceptance-sampling concepts and implementation scope