An out-of-tolerance device: the real question is how far back you go

The calibration report is in and the device is out of tolerance. The decision about the device itself is usually quick: adjust it, send it for repair, or take it out of service. The real work starts after that, and it is the part most often skipped — what was measured with this device before today?

The question is uncomfortable because the answer reaches into the past. Not asking it does not make it go away; it only means it will be asked during an audit or by a customer.

The suspect period is a definition, not a guess

You do not know when the device drifted out of tolerance. The only thing you know is this: at the last valid calibration it conformed, and today it does not. The suspect period is exactly the span between those two dates.

That span is not open to negotiation; it falls out of the records. The device’s previous valid calibration date (or, if it was never calibrated, the date it was acquired) is the start, and the date of this measurement is the end. In BellaKYS, recording a non-conforming result opens the impact assessment together with that span — trying to recall the period by hand biases you towards shortening it every single time.

What is inside that period is not the calibration record’s business

Here it is worth being blunt: the system gives you the suspect period, not the list of products measured with that device during it. That information does not live in the calibration record but in your production traceability — which device, at which station, on which batch, under which control plan.

So the impact assessment is not a calculation but a decision record: you supply the answer, the system keeps the answer and the reasoning. Where traceability is weak, the question resolves to “the whole period is suspect” and that is where the cost shows up. No calibration software can claim to remove that cost; it can only make the question impossible to skip.

“Nothing affected” is an answer too, and it has to be written

Most assessments end with “product conformity not affected” — the deviation is negligible next to the acceptance criterion, or no product was measured with that device in the period at all. That is a perfectly legitimate outcome. What is not legitimate is leaving the field blank.

A blank assessment is, to whoever reads it a year later, the question “was this considered or not?” and it has no answer. One sentence of reasoning — “deviation 0.02 mm against a tolerance of ±0.15 mm; product conformity not affected” — makes the record defensible.

The device does not stay on the floor while you decide

There is no case for keeping an out-of-tolerance device in use; in practice it stays, because withdrawing it is something a person has to actively do. That is why withdrawal should hang on the result itself: a non-conforming result puts the device in quarantine, and it cannot return to service while the impact assessment is still open.

The same applies to conditionally acceptable results. “Usable within a restricted range” is not a decision unless the condition of use is written down.

An overdue device raises the same question

A device past its calibration due date is “probably still fine”. The problem is precisely the word probably: you cannot demonstrate it is fine, so you cannot demonstrate the measurement made with it either. From an auditor’s point of view there is no difference.

This is why due-date tracking has to be a lock rather than a reminder: an overdue device should drop out of the usable list, not turn amber on it.

If product conformity was affected, the assessment alone is not enough

If the assessment says “yes, product was affected”, this stopped being a calibration matter and became a nonconformity: has anything shipped, is customer notification required, is a recall in scope. All of that belongs to corrective action, and an impact assessment closed without one leaves the answer half finished.

In BellaKYS that door is shut: if the assessment is marked as affecting product conformity, it cannot be approved without a corrective action linked to it. The corrective action is linked as derived from the calibration record, and its number is issued in the same form as every other record — DOF-2026-0001 — and does not change afterwards.

---

What makes a calibration programme defensible in an audit is not that devices were calibrated on time; that much is expected. What makes it defensible is what you did on the day a device came back out of tolerance. If there is no record of that day, the programme is only keeping a calendar.

Related pieces

5 August 2026 Why do corrective actions stay open? If your open CAPA count keeps climbing, the cause is usually not a lazy team but a step in the loop that nobody owns. 6 August 2026 Why does the supplier scorecard change nobody's decision? The scorecard is produced, filed, and produced again next quarter. If no decision hangs on the score, the measurement stops being a measurement. 28 July 2026 Why does audit preparation take weeks? If the records are already in the system, preparation should be short. When it is not, the problem is not the records but the missing link between a record and a clause.

Short answers

How far back do you go when a device is found out of tolerance?

The suspect period is not an estimate but a definition: the time since the last valid calibration. Because the moment the device started drifting is unknown, the boundary is drawn there.

Does the system identify the affected products itself?

It does not. The system gives you the suspect period; which products were affected in that period is a judgement made by someone who knows the device–measurement–product link.

Should "no products affected" be recorded?

Yes. Most assessments end with "no product affected" — that is also an answer, and without its reasoning written down it cannot be defended in an audit.

Is an overdue device the same problem?

It raises the same question. An overdue device is "probably still accurate" — that is exactly the problem; since accuracy is unproven, the same suspect-period assessment is required.