Decision rules: turning uncertainty into pass or fail

Every measurement result carries an uncertainty, so the moment you write “PASS” or “FAIL” against a tolerance you are making a judgement about that uncertainty. A decision rule is the agreed way you make that call — and since 2017, ISO/IEC 17025 (§7.8.6) has required you to have one, apply it, and report it whenever you state conformity.

What a decision rule is

A decision rule ties three things together: the specification (tolerance) limit, your measured result, and the expanded uncertainty U of that result. It states how the uncertainty is taken into account when you decide whether an item conforms. Clause 7.8.6.1 is explicit: when you make a statement of conformity, you shall document the decision rule employed, take the level of risk into account, and apply it. Clause 7.8.6.2 adds that the report must make clear which results the statement applies to, against which specification, and by which rule.

The problem uncertainty creates

Take a gauge block with a tolerance of ±5 µm. You measure an error of −4.8 µm with an expanded uncertainty U = ±0.6 µm (k = 2). The reading sits inside the tolerance — but the true value could lie anywhere from −5.4 to −4.2 µm, and part of that band is out of tolerance. Do you pass it? A decision rule is simply the agreement that answers that question the same way every time, instead of leaving it to whoever signs the certificate.

Simple acceptance (shared risk)

The plainest rule is simple acceptance, or “shared risk”: compare the measured value directly to the tolerance limit and ignore the uncertainty band (guard band w = 0). It is easy and widely used, but it means a result sitting right on the limit has roughly a 50% chance of being wrongly accepted. Under simple acceptance the customer agrees to carry that risk — which is exactly why 17025 wants the rule stated, not silently assumed.

Simple acceptance decides at the tolerance limit. A result exactly on the limit is a coin toss — accepted half the time it is actually out.

Guard bands: building in a margin

To cut that risk you pull the acceptance limit inside the tolerance by a guard band. The common choice is a guard band equal to the expanded uncertainty:

Acceptance limit = Tolerance limit − U  (guard band w = U)

Now you only pass a result if it is inside the tolerance by at least its own uncertainty — guarded acceptance. You can guard the reject side the same way. This is the approach ILAC G8 and JCGM 106 describe, and it sorts every result into zones: clearly conforming, clearly non-conforming, and a guard-band zone in between where a confident binary statement isn’t possible.

The zones, and what you report

• Guarded acceptance — result inside the acceptance limit: PASS with confidence.
• Guarded rejection — result beyond the tolerance by more than U: FAIL with confidence.
• Guard-band zone — result within U of the limit: report conditional (“non-binary”) language, or let the rule you agreed decide which way it falls.

A binary rule forces every result to PASS or FAIL; a non-binary rule allows “conforms / does not conform / conditional” so the genuine doubt near a limit stays visible instead of being hidden by a confident-looking tick.

Risk, in one number

Behind guard bands sits the probability of false acceptance (PFA) — the chance of passing an item that is truly out of tolerance. Tightening the guard band lowers PFA but rejects more good product; loosening it does the opposite. JCGM 106 frames this as the balance of consumer’s and producer’s risk; ILAC G8 turns it into practical guard-band choices. You don’t have to compute PFA for every result, but the rule you pick is a decision about how much of that risk you and your customer will carry.

Agree it up front, print it on the certificate

Because the decision rule decides who carries the risk, 17025 expects it to be agreed with the customer before the work — not chosen after the numbers are in. Then each certificate should state the rule used and, per result or characteristic, the statement of conformity it produced. A bare “Pass” with no rule and no uncertainty is exactly what an assessor will question first.

Where it fits

A decision rule is where measurement uncertainty stops being an academic figure and starts doing work: it converts your corrected result and its expanded uncertainty into a defensible pass or fail. Keep the tolerance, the corrected result, the uncertainty and the rule together in one place, and the verdict on the certificate carries its own justification.

More from the Cali blog: browse all calibration & metrology guides →