The call nobody in a lab wants

The short answer. When a reference standard is found out of tolerance, ISO/IEC 17025 treats the results it may have affected as nonconforming work: the lab looks at the effect on results issued since the standard was last known good and, when it matters, tells the customers (clauses 6.4.9 and 7.10.1 of ISO/IEC 17025:2017). That search, called reverse traceability, narrows in three steps: which certificates used the standard, which readings used the faulty part, such as one block in a gauge block set, and which pass or fail decisions could change once the part's measured error is applied under each customer's decision rule. Only customers in the last group need a personal notice; the rest need to know their results stand.

Your gauge block set goes out for its own calibration and comes back with two blocks out of tolerance. The set leaves service, and that part is easy. The hard part starts next: every certificate your lab issued with that set since it was last known good may now be in question.

ISO/IEC 17025 asks a lab to treat this as nonconforming work. When equipment is found defective, you look at the effect on earlier results, and when it matters, you tell the customers (clauses 6.4.9 and 7.10.1). The search behind that is called reverse traceability: starting from one standard and working out to every result it touched.

In a lab that keeps its jobs in spreadsheets and Word certificates, it usually means searching every past job for one serial number, then reading each one to see what the standard was used for.

Last reviewed: November 11, 2026

Three questions, in order

Reverse traceability is easier when you split it into three questions, each narrowing the last.

  1. Which certificates used the standard? Everything issued between the standard's last good calibration and the day it was found out of tolerance. This is the widest list, and in a busy lab it can run to hundreds.
  2. Which readings used the faulty part? A gauge block set is many standards in one box. If only two blocks are out, a certificate that used only the others is not affected. The same is true of a multi-range standard where only one range drifted.
  3. Which decisions could change? For each reading that used a faulty part, apply the part's measured error and ask whether the pass or fail decision on the certificate would come out differently under that customer's decision rule. Most won't. The few that would are the customers who need a personal notice.

The first two questions are lookups. The third is arithmetic, and it is where the lab's judgment and uncertainty budgets come in.

A worked example

Here is an illustrative case, with every figure made up.

A 25 mm block in the set was certified at its nominal length. At its recalibration it measured 0.4 µm short. Since its last good calibration, it was used on many micrometer calibrations.

Take one of them. The micrometer's tolerance at 25 mm is ±2 µm. On the block, it read 25.0018 mm, so the certificate recorded an error of +1.8 µm, a pass.

Now apply the block's real length. If the block was actually 24.9996 mm, the micrometer's true error was 25.0018 minus 24.9996, or +2.2 µm. That is outside ±2 µm, so under a simple acceptance rule the pass becomes a fail.

Another micrometer checked on the same block read +0.6 µm. Its true error was +1.0 µm, still well inside tolerance, and that result stands.

Same block, same error, two different outcomes. That is why the third question matters: it separates the handful of customers who need to act from the many who only need to know their results stand. The example ignores measurement uncertainty and guarded bands to keep the arithmetic plain; a real assessment uses the lab's own budget and each customer's agreed rule.

How to do it better, even in a spreadsheet

A spreadsheet can answer the first two questions quickly if it records the right things at the time of the calibration.

  1. One row per reading, not per certificate. Record the certificate number, the date, the customer, the instrument, the test point, the reading and the decision.
  2. Name the standard and the part on every row. The serial number of the set is not enough; record which block, or which range, was used for that reading. Without this column, question 2 means rereading every job.
  3. Keep the standard's history in its own sheet: each calibration date, its result and the date it was known good.
  4. Filter, then calculate. Filter the readings by the standard's serial and the date range, then by the faulty parts. Add a column that applies the part's error to each reading and compares the result with the tolerance.
  5. Log what you did. Keep the list, the decision for each and the date each customer was told, as the nonconforming work record.

The limits: nothing makes anyone fill in the part column at the bench, the decision column is only as right as the formula behind it, and the customer letters are still written one by one.

How a system does it

The demo for Tessaline Metrology, a fictional calibration lab, works this case through. Its accreditation body is fictional too.

The quality manager records the gauge block set's calibration results, with two blocks out. The set leaves service at once, and the bench refuses it from then on. Because every reading in the system records the blocks it used, the narrowing is arithmetic rather than a search: hundreds of certificates touched since the set was last known good, fewer with a reading on a bad block, and 6 where a decision could change once the block's measured error is applied.

Then the AI helper drafts, from the facts the system worked out and never from its own guesses: the impact assessment for the nonconforming work record, a personal notice to each customer where a decision could change, and one notice for the customers whose results stand. The AI never decides which certificates are affected, or how. The quality manager edits and approves every notice before it is sent, and customers read and acknowledge it in their portal.

Results imported from procedure automation, such as MET/CAL, pass the same checks as readings typed at the bench, so a set taken out of service is refused there too. The system applies the lab's uncertainty budgets and scope; it does not work them out, and whether the lab complies is for the lab, its quality manager and its assessor to decide.

What it costs, and the demo

The calibration lab page describes the rest of the system: the bench, release checks, recall and the customer portal. The live demo lets you record the failed gauge block set yourself and watch hundreds of certificates narrow to the 6 where a decision could change. Sign in with one click as the quality manager, a technician, the front desk or a customer, with no account to create; the data resets every night, so you can't break anything.

A typical build costs $8,000 to $25,000 at a fixed price, and hosting and support start at $350 a month.

Questions

What is reverse traceability in a calibration lab?

Reverse traceability is the search that starts from one reference standard found out of tolerance and works out to every result it touched since it was last known good. ISO/IEC 17025 asks a lab to treat the affected work as nonconforming and to tell customers when the effect matters.

Does every customer whose certificate used a failed standard need a personal notice?

Not necessarily. A personal notice is needed where applying the faulty part's measured error could change the pass or fail decision under that customer's decision rule; customers whose results stand need to know that. A real assessment uses the lab's own uncertainty budget and each customer's agreed rule.

What should calibration records capture to make reverse traceability fast?

One row per reading, with the certificate number, date, customer, instrument, test point, reading and decision, plus the specific part of the standard used, such as which block or which range. The serial number of the set alone is not enough, because then finding the readings on the faulty part means rereading every job.