Certificate of Analysis review

The lab said pass. Your spec said otherwise.

A supplier's laboratory checks each result against its own limit — not against yours. When your internal limit is tighter, the certificate still reads PASS, and there is no FAIL anywhere on the page to catch.

Heavy metals · ICP-MS Lot 0001350

Lead 895.3 ppb

In spec Over your limit

0 ppb 500 1,000

This lot — 895.3 ppb Your limit — 500 ppb In spec Over your limit, lab still passes it

Against the lab's limit NMT 1,000 ppb Pass

Against a receiving spec of ≤ 500 ppb Out of spec

Taken from a real, publicly published certificate. The laboratory's PASS is correct — against the laboratory's ceiling. Read against a tighter receiving spec, the same lot is a 79% exceedance.

How it works

No software to install, no integration, no system to learn.

You send

Certificates, as they arrive

Forward supplier COAs to a dedicated address. Any layout, any laboratory, born-digital or scanned.

Checked against

Your spec, not the lab's

Every result is read, normalised to your units, and compared with your own limits — plus anything the document itself leaves ambiguous.

You get back

One clean row per lot

A spreadsheet with out-of-spec results marked, and a record of what was checked, when, and against which version of the spec.

Where certificates get slippery

All five are from real published certificates, not invented examples. None of them would surface from a glance at a PASS/FAIL column.

No verdict at all

A heavy-metals sheet whose specification column read "Record Only" for every element. Nothing on the page to pass or fail against.

No limits printed

A whey concentrate certificate stating results only — no specification anywhere. The comparison has to come from your spec or it doesn't happen.

Same word, different test

Salmonella reported as absent in 25 g, not detected per 750 g, and negative per 1,500 g. A 60-fold spread that collapses into one identical "PASS".

Unreadable by machine

A turmeric certificate whose text layer lists every analyte first and every result second, with no link between them. Guessing isn't an option — those go to a human.

Not a number

One certificate marked "Preliminary Results Only", another "valid without signature". Neither is out of spec; both are questions an auditor can ask.

Wrong document entirely

A calibration standard's certificate filed among ingredient COAs. It has no receiving spec to check — it needs recognising and routing, not parsing.

What this is not

This is decision support, not a sign-off. It structures and checks the documents you provide and surfaces what falls outside your limits. Final quality review, release and hold decisions stay with you, and the authoritative certificate files stay in your records.

It is not a laboratory, not a substitute for your own testing, and not your system of record for FSMA or audit purposes.

See a sample run

Six real supplier certificates from six different laboratories — including every awkward case above — read and checked against an example specification. Each document is linked, so any line can be verified against its source.

Open the sample run Ask a question

Who you'd be dealing with

Name: Maksim — one person, working directly. No account manager in between.
Background: Backend and data engineering. Not a quality professional — the limits are always yours, and the judgement calls stay with your QA team.
Location: Kazakhstan, working remotely with clients in the US and Europe.
Your documents: A one-page data agreement is available before anything is sent: used only for the agreed work, never used to train models, deleted on request.
Getting started: A short pilot on a handful of your own certificates, at no cost — so you can judge the output on your documents rather than on a demo.