Intermediate Checks: Verifying Instruments Between Calibrations

Intermediate checks gather evidence about performance between calibrations. They aim to detect relevant changes before further measurements are compromised, but their usefulness depends on what they can observe. A PASS box repeated daily does not, by itself, demonstrate control across the instrument’s entire range.

Start with a concrete question: which fault could affect our work, and which check would detect it? Reference, configuration, frequency, limits, records and actions follow from that question. The check becomes part of the metrological programme, with declared coverage and recognisable limitations.

1. Start with the change to be detected

Link the check to quantity, range, function and use: zero shift, sensitivity change, poorer repeatability, incorrect channel or reference instability. A power-on test detects some functional faults but not necessarily a metrological error. Define which change would matter to the laboratory’s decision.

EU GMP Chapter 3, §3.41, requires calibration and checks at defined intervals using appropriate methods and records. It does not prescribe a universal daily frequency. ILAC-G24/OIML D10:2022 links intermediate checks to maintained performance and programme review; it does not make any arbitrary check interchangeable with full calibration.

Consider detection timing too: a check only at the end of a week may leave more activities requiring review than one before critical use. Balance risk, the possible speed of change, check capability and practicality, while respecting established applicable requirements.

2. Choose references, points and configurations

The reference must be stable for its use, identified and adequately characterised. Check assigned value, uncertainty, conditions, storage and checks on the reference itself. A contaminated weight or degraded solution can trigger an alarm without instrument failure; investigate this possibility rather than assuming it to dismiss an unfavourable result.

Assess independence: a reference assigned solely by the instrument being checked can make the comparison circular. Two devices may share a standard, software coefficient or environmental influence. Good agreement does not exclude a common error. Document these dependencies and introduce alternative evidence where risk requires it.

Select points representing relevant failure modes. Checking zero does not demonstrate sensitivity; checking mid-range does not exclude end-range faults. The number of points depends on model and use. The NIST example for controlling linear calibration is not a general “three points are enough” rule for every system.

3. Original matrix: risk, check and coverage

The responsible person assigns frequency, triggering events, performer, reviewer and records to each row. This matrix is a design structure, not a mandatory test list for every instrument.

4. Set limits, frequencies and triggers

Separate acceptance limits related to use, preventive thresholds and statistical control limits. A statistically stable series may be centred on an unacceptable value; a statistical signal may require investigation before tolerance is exceeded. Do not automatically widen thresholds because data have become more dispersed.

Define treatment of comparison uncertainty, what constitutes a doubtful outcome and when use must stop. Frequency may depend on time, usage or events: transport, impact, maintenance, configuration changes or suspected reference deterioration. Greater frequency does not fix a check unable to detect the fault.

Assess whether the check itself changes the system. If it involves cleaning or adjustment, preserve the earlier condition where needed and separate the stages. Post-intervention data do not replace the data intended to monitor stability.

5. Record values and interpret trends

Retain instrument and reference identities, point, configuration, date, operator, conditions, original values, processing, criterion and outcome. PASS alone removes information useful for recognising a trend. Also identify missed checks and their handling: a gap in the series is not a demonstrated period of conformity.

Compare like with like. Changes in reference, method, configuration or adjustment may interrupt comparability and must be annotated without deleting history. Graphs and control charts need an appropriate model and sufficient data; do not turn a few points into a certain prediction of the next deviation.

A doubtful or unfavourable result activates the defined procedure: preserve evidence, assess reference and conditions, suspend affected uses where necessary and involve technical and quality personnel. Repeats must test a hypothesis, not continue until the first PASS. Return to use and assessment of earlier results remain separate decisions.

6. Simulated case: the usual point misses the fault

A balance is routinely checked with a 100 g reference. Today’s comparison gives error e = indication − reference of +0.2 mg and expanded uncertainty U = 0.3 mg, with k = 2. The fictional internal limit is ±1.0 mg. The illustrative rule accepts when |e| + U ≤ 1.0 mg, declares nonconformity when |e| − U > 1.0 mg and treats other cases as inconclusive.

At 100 g, 0.2 + 0.3 = 0.5 mg: the point meets the criterion. A later test with a 10 g reference gives e = +2.0 mg and U = 0.3 mg. Since 2.0 − 0.3 = 1.7 mg, that point fails the requirement under the rule. Masses, errors and limits are invented; they do not define pharmacopoeial requirements or minimum weight.

The favourable 100 g history neither excludes the 10 g fault nor establishes its onset. The laboratory assesses affected uses, confirms the finding and expands the check with technical justification. No blanket correction is applied to historical data on the basis of two points.

Review must establish whether the check detects relevant faults and whether its frequency is adequate. Results inform calibration interval review without automatic deferral. For investigation, see out-of-tolerance instruments; return to the Calibration and Metrology hub.

7. Sources and applicability limits

Checked on 1 October 2026. The matrix, case and numerical rule are original and illustrative.

1. EU GMP, Chapter 3: full text, effective 1 March 2015, §§3.41 and 3.44.
2. ILAC-G24/OIML D10:2022: full text, §§4.9–4.10 and 6.5; guidance stating limitations of partial checks.
3. NIST/SEMATECH, §2.3.5 and §2.3.7: complete pages consulted, no visible revision date; specific examples of check standards and linear calibration, not universal GMP frequencies.
4. ISO/IEC 17025:2017: official catalogue, confirmed in 2023; full standard not consulted. Check applicable requirements in the controlled copy.