Acceptance criteria: turning a specification into an unambiguous yes or no

A limit is not a criterion until the method is fixed

A stated tolerance leaves several questions open: where on the feature the measurement is taken, how the part is held, at what temperature, with what class of instrument, and whether a form deviation counts against the size. Two honest inspectors can measure the same bore and disagree simply because one used a two-point gauge and the other a coordinate machine, and both readings are correct for what they measured. Criteria that state the limit and the method together remove that whole category of disagreement, and cost nothing extra to write while the specification is still being drafted.

Cosmetic requirements need physical references, not adjectives

Appearance is where written criteria break down fastest. Words such as free from significant marks depend entirely on the reader, and inspectors will apply them differently across shifts, let alone across a customer boundary. The workable solution is physical: an agreed acceptable sample and an agreed unacceptable sample, kept at the inspection point, signed and dated by both parties, with a defined viewing distance, orientation and lighting. It is cruder than a written definition and works far better. Refresh them periodically, because reference parts fade, corrode and get borrowed. Where a customer will not supply signed references, photograph what you have agreed and send the images with your interpretation attached.

Somebody has to own the interpretation, in writing

When a drawing is genuinely ambiguous, plants resolve it informally and the resolution lives in one person's memory. That works until the person leaves or the customer changes engineer, at which point production continues to a rule nobody can evidence. Every interpretation should be captured as a written note against the part, ideally acknowledged by the customer, and treated as part of the specification thereafter. This is unglamorous administration that pays for itself the first time a complaint arrives about a feature that has been made the same way for years.

The zone where neither answer is comfortable

Parts near a limit create the most expensive decisions in a factory, because measurement uncertainty means the true value may sit on either side. Two policies are defensible and both need stating in advance: accept only where the measured value is inside the limit by more than the uncertainty, which rejects some good parts, or accept anything measured within the limit, which passes some marginal ones. Choosing per-incident is what causes trouble, since the decision then tracks how badly the delivery is needed. Fix the rule by characteristic and record it with the criteria.

Criteria that quietly move over time

Standards drift in both directions without anyone deciding. Inspectors under delivery pressure become more lenient; a new inspector wanting to demonstrate rigour becomes stricter and the reject rate jumps. Both look like a process change to everyone downstream. Periodic comparison of several inspectors judging the same set of parts exposes the drift cheaply, and it is more useful than retraining because it shows where the criteria themselves are unclear. Where inspectors disagree with each other, the fault usually lies in the written definition rather than in the people applying it.

Frequently asked questions

How do we write acceptance criteria for something as subjective as surface appearance?

Stop trying to write it and produce physical references instead. Agree a part that represents the worst acceptable condition and one that is clearly unacceptable, sign both with the customer, and specify how they are to be viewed: distance, angle, lighting and whether magnification is permitted. Keep them at the inspection station rather than in an office. Written wording then only has to point at the references, which is a description two people can apply consistently.

What should we do when the drawing is ambiguous and the customer is slow to respond?

Record your interpretation, state the reasoning, send it to the customer and continue on that basis while making the assumption visible on your inspection documents. That converts a future disagreement about intent into a question of whether anyone objected when told. Do not resolve it verbally on the phone and leave no trace, which is how plants end up manufacturing for years to an understanding they cannot evidence when the engineer who agreed it has moved on.

Should measurement uncertainty be subtracted from the tolerance when accepting parts?

It depends on who carries the risk, which is a commercial choice rather than a technical one. Guarding the limit by the uncertainty means only parts proven inside are accepted, which protects the customer and rejects some conforming parts. Accepting anything measured inside favours the producer and lets marginal parts through. Either is defensible if agreed in advance and applied consistently; what causes disputes is switching between them according to how urgently the delivery is needed.