Skip to content

Acceptance Sampling Plan

Inspection procedure — instantiates Tolerance Band Management

Inspects a defined sample from a lot and accepts or rejects the whole batch on the result, buying a controlled confidence about conformance without inspecting everything.

An Acceptance Sampling Plan decides whether a batch is acceptable by inspecting only a subset of it. Its defining idea — the thing no sibling shares — is the deliberate trade of certainty for cost: rather than measure every unit (often impossible, destructive, or ruinously slow) or measure none, it draws a statistically chosen sample, counts how many fall outside the band, and accepts or rejects the whole lot against a fixed allowable-defect number. It is a lot-level accept/reject engine, tuned so that the risk of passing a bad batch and the risk of failing a good one are both known and chosen in advance.

Example

A contract manufacturer receives a shipment of 10,000 electronic connectors. Testing every one for pin retention would be slow and partly destructive, so incoming QC applies an acceptance sampling plan: for a lot this size at the agreed quality level, pull a sample of ≈200 connectors, and accept the shipment if no more than 3 fail, reject it if 4 or more do. The inspectors pull the sample per the plan's rule, test them, and get 2 failures — under the limit, so the lot is accepted and released to the line. Had 5 failed, the whole shipment would have been rejected and returned, not sorted. The plan never claims every connector is good; it says the sample gives an agreed confidence that the batch's defect rate is low enough to use, and it fixes exactly what "release" versus "return the lot" means so the decision isn't argued case by case.

How it works

Its distinguishing machinery is the sampling-and-decision rule, sized by risk. The plan sets a sample size and an acceptance number (max defectives to still accept) derived from lot size and a target quality level, so the plan's operating characteristic — its probability of accepting lots of each true quality — is chosen, not accidental.[1] Effort concentrates where it pays: bigger or higher-consequence lots draw larger samples; a supplier with a clean history may earn reduced inspection, a shaky one tightened inspection. The output is binary and lot-level — accept or reject the batch — which is what makes it a corrective decision (return/quarantine the whole lot) rather than a per-item gauge.

Tuning parameters

  • Acceptable quality level — the defect rate the plan is built to pass routinely. Looser passes more marginal lots cheaply; tighter rejects more but protects the line.
  • Sample size — how many units are drawn. Larger samples sharpen the accept/reject decision but cost more to inspect.
  • Acceptance number — the max defectives that still accepts the lot; the dial that sets producer's vs. consumer's risk.
  • Switching rule — normal / tightened / reduced inspection based on the supplier's recent record, concentrating effort by risk.
  • Sampling cadence — every lot, every nth lot, or event-triggered; matched to how fast quality can shift.

When it helps, and when it misleads

Its strength is economic: it delivers a known, chosen confidence about a whole population at a fraction of the cost of 100% inspection, with an accept/reject rule agreed by both sides so lot decisions don't turn into negotiations. Its limits are structural. Sampling is a filter on batches, not a per-unit guarantee — a passed lot still contains some defectives, so it is the wrong mechanism where a single escape is catastrophic (there, inspect directly). Its classic misuse is treating "lot accepted" as "every part good," or shrinking the sample to hit a throughput target until the plan can no longer see the defect rate it was built to catch. The guard is to size the sample from the actual consequence of a missed defect, and to route rejected lots and repeat rejections into review rather than re-sampling until one passes.

How it implements the components

  • risk_based_sampling_plan — its core: what to sample, how much, and how effort concentrates where deviation is most likely or most costly.
  • inspection_cadence — it sets when lots are checked (every batch, every nth, on trigger) and the switching between tightened and reduced inspection.
  • corrective_action_rule — the accept/reject-the-lot decision is the correction: a failing batch is returned, quarantined, or 100%-sorted, not passed.

It does NOT define the band a unit is measured against (that's Engineering Tolerance Specification) or the instrument that measures the sampled unit (that's Calibration Procedure and Go/No-Go Gauge); it decides which lots to check and what to do with a failing one.

  • Instantiates: Tolerance Band Management — this procedure enforces the band at the batch level when checking every unit is infeasible.
  • Consumes: Engineering Tolerance Specification supplies the per-unit band each sampled item is judged against.
  • Sibling mechanisms: Go/No-Go Gauge · Quality Control Limit · Engineering Tolerance Specification · Calibration Procedure · Statistical Process Control Chart · Exception Review Workflow · Clinical Reference Range · Grading Rubric · Policy Discretion Bounds · Service-Level Tolerance · Usability Tolerance Test

References

[1] The operating-characteristic (OC) curve of a sampling plan gives, for each true lot quality, the probability the plan accepts it. Standards such as ANSI/ASQ Z1.4 (formerly MIL-STD-105) tabulate sample sizes and acceptance numbers so the plan's producer's risk and consumer's risk are chosen deliberately rather than left to chance. ↩