Decision rules under ISO/IEC 17025: simple acceptance or a guard band

When is a decision rule required?

Whenever a certificate makes a statement of conformity to a specification: pass or fail, in tolerance or out. It is derived from what the certificate asserts, not chosen on a form.

Deriving it this way matters. If it were a checkbox, a jack certificate could acquire a pass/fail statement it should not carry, and a meter certificate could lose the declaration it must.

What is simple acceptance?

Also called shared risk, described in ILAC-G8:09/2019 §4.2.1. The measured value is compared directly with the tolerance limits and the uncertainty is not used to move them. A reading just inside the limit passes even where its uncertainty interval crosses the limit.

The name shared risk is the honest one. Where the uncertainty is comparable with the tolerance, a result near the limit carries a real probability of being wrong, and under simple acceptance that risk is shared between the laboratory and the customer. It is appropriate, and common, where the uncertainty is small relative to the tolerance and the customer understands the convention. The customary threshold is a test uncertainty ratio of 4:1, the tolerance span against twice the expanded uncertainty.

What is binary guard banding?

ILAC-G8:09/2019 §4.2.2. The acceptance limits are pulled inwards by the expanded uncertainty U: the effective lower limit is the lower tolerance plus U, the effective upper limit is the upper tolerance minus U, and the reading passes only if it lies between the effective limits. A pass then means the true value lies inside the tolerance with at least the stated confidence, rather than that the measured value did. The laboratory carries the risk and the customer receives a stronger statement.

The cost is real and should be stated to whoever is choosing: instruments that would have passed under simple acceptance now fail, in proportion to how large U is against the tolerance. Where U approaches half the tolerance, a guard band rejects almost everything, which is the arithmetic telling you the measurement is not good enough to make the statement at all.

A worked example

A meter is checked at 100 V against a specification of ±1 %, so the limits are 99 V and 101 V. The expanded uncertainty of the measurement is 0.25 V at k = 2. The test uncertainty ratio is the 2 V span over twice U, 0.5 V, which is 4:1.

The 100.9 V reading is the whole argument in one row. Its uncertainty interval, 100.65 to 101.15 V, crosses the limit; under simple acceptance it passes and the risk that the true value is above 101 V is shared with the customer; under a guard band it fails and the laboratory has refused to make a claim it cannot support at 95 %. Neither is wrong. They are different promises, and the certificate must say which one it is making.

The guard band calculator shows what any set of limits becomes under each rule, and the TUR calculator gives the ratio that decides between them.

What must the certificate say?

ISO/IEC 17025 §7.8.6 requires three things wherever a statement of conformity is made: which specification the result was judged against, which results it applies to, and the decision rule that was used, unless it is inherent in the specification. A certificate that prints PASS beside a row and nothing else fails all three. The wording can be short. Statement of conformity by simple acceptance, ILAC-G8:09/2019 §4.2.1; measurement uncertainty not applied to the limits. The point is that a reader two years later knows what the pass meant.

Who decides, and when?

The laboratory decides and the customer agrees. ISO/IEC 17025 §7.1.3 puts the agreement at contract review, before the work, because the rule changes what the customer is buying: a guard band buys a stronger statement at the price of more failures. A laboratory that has never had the conversation is usually applying simple acceptance by default, and should say so in its terms rather than discover it at an assessment.

It is also a decision software must not make. One default is not safer than the other; they distribute risk differently, and the choice belongs to the laboratory that signs the certificate. CalibraWorks refuses to issue a statement of conformity until the laboratory has declared its rule, and prints the declared rule on every certificate that makes one. Guidance is ILAC-G8:09/2019.