Desrochers Validation, Félix Desrochers What I don't do, and sources

Uncertainty budget documentation

I document and verify the Excel workbooks your measurement uncertainty budgets live in, and hand back a record your technical manager can sign. The first one is free.

Clause 7.6 requires an uncertainty estimation for every measurement function you perform, and in practice each one is a spreadsheet. Published summaries of A2LA deficiencies put uncertainty budgets missing required components at number two on the top-ten list. Those are secondary sources, not A2LA's own publication, and I'd rather you knew that than took my word for it. Every other claim on this page is sourced on the reference page, marked primary or secondary.

What you get

The record is built around the NIST Office of Weights and Measures Software Verification and Validation form (GLP 15, Appendix A, published free by NIST). It has 47 checks. Around thirty come back answered from the file itself, and I leave the eleven only you can answer, meaning backups, passwords, macro behaviour and which procedure you chose, blank for your signature. Every one of the 47 appears in the record, so nothing is quietly skipped.

How many is "around thirty" depends on your workbook, and I would rather give you the range than a number that sounds more certain than it is: some checks will not apply to your file at all, and two of them need things I ask for separately, namely last year's version of the same workbook and your accredited scope. If the record cannot answer an item, the reason is written in it.

You send one workbook. I return a signed-ready PDF, and a spreadsheet of the same results for your own QA system. The PDF opens with a numbered cover, a contents page, and a one-page summary telling you what was found and what you need to do. Everything after that is evidence, in this order:

  1. Identification: file name, hash, size, every sheet, and what depends on it.
  2. A requirements specification derived from the file itself. Most labs have none, and without one there is nothing to validate against.
  3. Test cases and results, with expected values derived from those requirements and your input columns, never from the formulas being tested.
  4. Boundary and negative tests: zero, negative and text substituted into an input, with your own formula chain re-evaluated each time to show what it does.
  5. Findings, against the GUM and ILAC P14: what your budget contains, what a complete one should, and the gap. One entry per distinct issue, with every place it was seen.
  6. Coverage against the NIST form, and the evidence behind every check that did not pass.
  7. Change control: protection state, hand-entered values sitting inside formula columns, volatile functions and references outside the workbook, plus a block for your version and authorisation.
  8. A statement of limitations, covering what I did not test and why.
  9. A signature block for your technical manager.

Then three appendices: all 47 checks, every test case rather than only the failures, and a formula inventory. You are meant to read page 3 and check the rest.

What I can recompute is bounded, and the record says where the boundary is. Uncertainty budgets laid out the way calibration laboratories build them, one row per contribution with a value, a divisor and a sensitivity coefficient, and linear calibration curves. A budget built by another convention, such as the Kragten method common in analytical chemistry, gets every hygiene and integrity check but is not recomputed, and the record says so rather than letting it look like a pass.

Section 2 is the coverage map: every worksheet in your file, how many formula cells it holds, which calculation was recognised on it, and how many tests ran against it. Where a sheet holds a calculation I cannot test, it says so and names it. That is not a disclaimer at the back. It is a column you can read.

Read a complete sample dossier →

Price

ItemPrice
First workbook, for my first ten customersFree
First workbook after that$750
Each additional workbook$250
Annual re-verification, per workbook$150
Ten-workbook bundle, first year$3,500
Ten-workbook bundle, annual renewal$1,200

The free first file is an introductory offer for my first ten customers, not a standard price. The first file costs more than the ones after it because it includes the reverse-documentation the others reuse. If you have thirty budgets, the bundle is the number worth looking at, not the per-file one. I'm telling you where this ends because it's true, not to create urgency.

Turnaround is ten business days from the file arriving. If I can't read your workbook I'll tell you the same day, and there's nothing to pay.

Sending files to a stranger

Reasonable objection. I work on the formula layer, not your data. Delete every data row, strip sample IDs and client names, and send a workbook whose formulas are intact. That is 100% of what I need, and a redacted file gets the same dossier as a complete one.

Written data handling on request: kept only for the engagement and whatever retention period you name, no subprocessing, no use as training data, deletion confirmed in writing. Standard NDA if you want one.

What I won't sell you

There are eleven more things I don't do, with what each means for you, on the reference page.

What I haven't got

No record I have produced has been through an accreditation assessment yet. I started in 2026. Being able to complete a NIST form is not the same as your assessor accepting it, and I have no evidence about the second thing, because nobody has taken one in yet. You would be finding that out with me.

That is the honest reason the first file is free, and it is the main risk you are taking. I would rather you weighed it now than discovered it later.

Three things you'd otherwise have to ask

Are you a metrologist? No. I don't build uncertainty budgets and I don't decide what belongs in yours. I document and test the one you already have.

What accreditation do you hold? None, and neither does anyone else. There is no accreditation for validation providers. Your technical manager signs the record. I never sign, attest or approve anything.

How long have you been doing this? I started in 2026, and no record of mine has been through an assessment yet. You're taking the risk on someone new, so you shouldn't also be paying for it. That is the whole reason the first file is free.

En français

Je travaille en français comme en anglais. Je suis établi en Mauricie, et le dossier et les échanges peuvent être entièrement en français. Écrivez-moi en français et je répondrai de même.

Contact

Email me the workbook, or email me first with questions. Either is fine.
felix@desrochersvalidation.com