Inspection Sampling Plan

Inspecting every item can be slow, costly or destructive, so a sampling plan checks a defined subset and uses its results to decide what to do with the lot, accepting that the choice carries a risk of error. The plan starts with a clearly defined lot, because mixing products from different suppliers or processes can hide problems.

A fictional buyer receiving parts from two production days treats them as separate lots rather than combining them just to reduce tests. A simple attributes plan specifies sample size n and acceptance number c, and if defects exceed c the lot is rejected under that plan; NIST presents this (n,c) approach.

A fictional factory receiving 1,000 parts randomly selects the agreed sample and checks dimensions, with a plan that checks 50 items and allows at most one defined defect, so two defects trigger rejection under the pre-agreed rule. The figures are illustrative, not a recommended plan.

Selection should avoid bias, because choosing only boxes at the front can miss damage elsewhere, as when a fictional warehouse sampled only top cartons for months and defects in lower pallets went unnoticed. Random or otherwise specified selection methods matter.

It is also essential to define what counts as a defect, since cosmetic marks, functional failures and safety hazards have different consequences, so a fictional electronics order allows a small number of minor finish issues but none of a critical electrical fault. The acceptable quality level, lot size and inspection severity can affect standard tables, so do not pick an n and c pair by intuition when a contract or standard governs, and seek quality expertise for high-risk goods.

A fictional procurement officer who finds a generic online sampling table checks the agreed standard, inspection level and product risk before using it. Producer risk is the chance of rejecting a sufficiently good lot and consumer risk is the chance of accepting an unacceptably poor one; no finite sample eliminates both, so a fictional supplier and buyer choose a plan that reflects their agreed risks.

Acceptance sampling is not process control, because it makes a lot-level decision from limited evidence while preventing defects requires attention to production and suppliers; a fictional buyer that rejects repeated lots sees the supplier fix a machine setting rather than rely only on more end-of-line inspection. A passed lot may still contain defective units, since the plan permits a statistical decision, not a warranty for each item, and a later faulty unit is handled under the customer and supplier terms.

Some tests destroy the sampled items, so the plan should state whether tested units can be sold and how replacements are counted, instead of silently returning damaged samples to stock, as a fictional food lab adjusts the saleable quantity after consuming several packages. Record the lot ID, selected units, test method, inspector, results and decision, and keep calibration or lab records where relevant, because a sampling plan is a documented decision rule under uncertainty that is sound only when lot identity, selection, defect definitions and risk tolerance are explicit.

Rejection does not always mean immediate disposal, since contracts may allow sorting, rework or resampling under rules agreed beforehand, such as 100% sorting and reinspection of a corrected lot rather than drawing new samples until one passes. Sampling frequency can change with supplier performance or regulation, but reduced inspection should be earned by evidence, and critical safety goods may need tests on every unit that a generic commercial plan cannot override.