Nonconforming work under ISO/IEC 17025 (clause 7.10)
An out-of-tolerance reference, a failed data-quality check or a wrong method must be controlled, evaluated for impact on results issued, and recorded.
Clause 7.10 requires a procedure for nonconforming work — any aspect of laboratory activity or result that does not conform to the laboratory's own procedures or the customer's agreed requirements. The procedure must define responsibilities, evaluate the significance of the nonconformity, decide on its acceptability, and inform the customer and recall work where necessary.
Temperature-work examples
- A reference thermometer found out of tolerance at recalibration: every certificate issued since its last good calibration is assessed.
- A mapping dataset with a flat-lined sensor: the sensor is excluded, the impact on hot/cold-spot conclusions evaluated, the report amended.
- A logger calibrated at the wrong points for its use: the certificate is withdrawn and reissued.
ValiTrac's data-quality states (REVIEW REQUIRED, CRITICAL) and validation failures are designed to become nonconforming-work records rather than silent corrections.
The five things the procedure must define
Clause 7.10.1 is unusually specific about what the procedure has to cover, and a procedure missing any of these is a finding on its own terms rather than a judgement call.
| Requirement | In practice |
|---|---|
| Responsibilities and authorities | Named roles: who may halt work, who evaluates, who authorises resumption |
| Actions based on risk levels | A graded response — a resolution error on one certificate is not a failed reference standard |
| Evaluation of significance | An impact assessment, including on results already issued |
| Decision on acceptability | A recorded decision, with the reasoning, not just an outcome |
| Notification and recall where needed | Who tells the customer, in what timeframe, and how certificates are withdrawn |
The look-back, which is where the work is
A reference thermometer found 0.08 °C out at recalibration is not one problem; it is every certificate issued since the last good calibration. The look-back has to establish when the drift plausibly began — intermediate check data is what makes this answerable rather than a guess — and then, for each affected certificate, whether the error was large enough to change a reported value beyond its stated uncertainty or to flip a conformity statement.
Most of them will be unaffected, and saying so with a calculation is the point. The ones that are affected have to be reissued and the customers told. A look-back that concludes no impact without showing the arithmetic is the version that gets challenged.
Worked examples from temperature work
- Flat-lined sensor in a mapping study: exclude it, record the exclusion and its range, reassess whether the remaining sensors still satisfy the study design, and amend the report. If the excluded sensor was the hot spot, the study no longer supports its own conclusion.
- Logger calibrated at the wrong points: the certificate does not cover the range of use. Recall it, recalibrate, and assess whether the customer relied on it in the uncovered range.
- Conformity statement issued without a decision rule: the results are sound, the statement is not. Reissue the certificate with the rule named, referencing the version it replaces.
- Calibration performed with an expired reference certificate: traceability is broken for the period, regardless of whether the reference turns out to have been in tolerance.
Where 7.10 stops and 8.7 starts
Clause 7.10 is containment: stop it, assess it, fix this instance, tell whoever needs telling. Clause 8.7 is corrective action: work out why it happened and stop it recurring, then check the action worked. Clause 7.10.2 makes the link explicit — where the evaluation indicates the nonconforming work could recur, the corrective-action procedure applies. Recording nonconformities and never raising a corrective action produces a log that grows and a laboratory that does not improve.
The healthy signal is a nonconforming-work log with entries in it. An empty log means the detection mechanisms are not working, and assessors read it that way.
Frequently asked questions
- Is a data-quality warning a nonconformity?
- It is a candidate. The evaluation of significance decides whether it affects results; the decision and its rationale are the record.
References
- [1]ISO/IEC 17025:2017 — General requirements for the competence of testing and calibration laboratories
General technical guidance written against the cited sources. It is not regulatory or legal advice and does not replace the applicable standard, guideline or a qualified reviewer's judgement.
Have a question on this topic?
Ask ValiTrac AI and see the evidence and calculation behind the answer.
Ask ValiTrac AIFree calculator: Certificate checker
Paste the text of a calibration certificate and screen it against ISO/IEC 17025:2017 clause 7.8: uncertainty with k, traceability, as-found and as-left, the decision rule behind a pass, and every other required element, with the matched text shown for each.
Open the calculatorCalibration on ValiTrac
Support calibration workflows, uncertainty budgets and certificate generation.
Open the calibration worksheetRelated articles
As-found vs as-left
Why both sets of readings matter, and what to do when the as-found data are out of tolerance.
Data-quality checks before analysing mapping data
Statistics computed on bad data are precise and wrong. Check the data first.