Validation
Software validation pack
An accredited laboratory has to validate the software it uses before relying on its results (ISO/IEC 17025 clause 7.11.2). This pack gives you the evidence for Metrology Bench: 695 published test vectors with cited sources, a way to run them in your own environment, and a record to fill in.
The pack is evidence for your validation, not a validation done for you. Your lab decides the intended use, the acceptance criteria and whether the results meet them. Results are calculated values, not certified results.
Run the vectors in this browser
Checks every published test vector against the library this site’s calculators run, in this browser. It takes a few seconds. Nothing you see here is sent anywhere.
A validation in four steps
- Write down what you use Metrology Bench for: which calculators, which ranges, which results end up on a certificate or in a decision.
- Pick the test vectors that cover that use (each calculator page lists its own), and add cases of your own where your use goes beyond them.
- Run them where you work: in the browser your technicians use (above), or with the command below if you use the library or the MCP server.
- Fill in the validation record, attach the report, and have it approved. Repeat when the version changes; the changelog says whether any result moved.
Run it on a computer
With Node.js 20 or later. Exits with code 1 if any vector fails, so it can gate an installation check.
npx -p @metrologybench/core@0.6.0 metrologybench-validate --out report.md
Add --json for every result in full, or --vectors file.json to run your own cases in the same format.
What the report shows, and what it doesn’t
- Which library version ran, in which runtime, against which vector file (by SHA-256), and when.
- For every vector: the expected value, its tolerance, what the library returned at full precision, and pass or fail. Inputs outside an equation’s range must be refused, and those vectors check that they are.
- It does not show that the equations suit your measurement, that your inputs are right, or that results are rounded the way your procedures require. Those are yours to judge.
The same runner is used by our own continuous integration on every change, and a guard blocks any calculated result from moving without an announced version change. Browse all test vectors · Download the vector file