Evaluation of measurement uncertainty (clause 7.6)

Clause 7.6 requires laboratories to identify the contributions to measurement uncertainty and take all significant contributions into account using appropriate methods of analysis. A laboratory performing calibrations, including of its own equipment, must evaluate the measurement uncertainty for all calibrations.

What 'appropriate methods' means in practice

• The GUM's law of propagation for analytical budgets.
• Monte Carlo (GUM Supplement 1) where the model is non-linear or the distributions are far from normal.
• Documented budgets that can be reproduced and audited.

ILAC-P14 adds the accreditation-body policy: the CMC listed in the scope defines the floor, and the reported uncertainty on a certificate must include the contributions of the customer's device, so it is normally larger than the CMC.

CMC, and why your certificate says a bigger number

The calibration and measurement capability in an accredited scope is the smallest uncertainty a laboratory can achieve on a near-ideal artefact under near-ideal conditions. It is a floor, and it describes the laboratory. The uncertainty on your certificate describes a measurement of your instrument, and it has to include contributions your instrument brings with it — its resolution, its repeatability, its hysteresis, how it sits in the bath. Those are frequently larger than everything the laboratory contributes.

ILAC-P14 is the policy behind this: the reported uncertainty must not be smaller than the CMC, and it normally exceeds it. A certificate quoting exactly the CMC for every instrument has probably not included the customer's device at all, which is worth asking about.

Appropriate methods of analysis

Clause 7.6 requires significant contributions to be taken into account using appropriate methods, without prescribing one. Three are accepted, and the choice is a technical judgement to be recorded.

• The GUM law of propagation of uncertainty: an analytical budget with sensitivity coefficients, combined in quadrature. Right for almost all temperature work, where the model is close to linear.
• Monte Carlo propagation of distributions, from GUM Supplement 1 (JCGM 101). Right where the model is strongly non-linear, where an input is far from normal, or where one dominant input is rectangular and the usual normal-coverage assumption for k = 2 no longer holds.
• Evaluation from method-validation data or interlaboratory results, where a full analytical model is impractical. More common in testing than in calibration, and it still has to be documented well enough to reproduce.

Testing laboratories are not exempt

The obligation is graded rather than lifted. A calibration laboratory must evaluate the uncertainty of every calibration, including calibrations of its own equipment. A testing laboratory must identify the contributions and evaluate the uncertainty, though where a well-recognised test method sets limits on the values of the major sources and specifies the form of reporting, following that method is accepted as meeting the clause. Identifying the contributions is not optional in either case.

The budget itself is a technical record under clause 7.5: it has to be reproducible by someone else, which means the sources, the values, the distributions, the divisors and the sensitivity coefficients are all written down — not just the final number.