Verification, not just extraction.
Supplement QA teams move third-party lab results from CoA PDFs into spreadsheets by hand — and the part that breaks is the part that carries the liability: the limits, the units, the pass/fail.
General document-AI can pull the fields. What it doesn't do is recompute the verdict and reconcile it against what the lab printed. So the safety-critical step — is this number actually within spec? — stays manual and error-prone.
The one thing we do that an extractor doesn't
For every analyte, a CoA carries three coupled values: the limit (the spec), the result, and a pass/fail. Coatables normalizes the limit notation (≤, NMT, NLT, ranges, Absent in 25 g) and units, recomputes pass/fail from the numbers, and then compares our computed verdict to the lab's printed one. When they disagree, you get a verdict_mismatch flag — the moment a misread number would otherwise have slipped through.
A lab can print PASS on a row where the result exceeds its own limit. An extractor copies that PASS faithfully. Coatables flags it. That's the whole difference.
Honest by design
- The verdict column is for review, not release sign-off — a human stays in the loop.
- We cover a bounded set of common supplement lab formats, deliberately, so accuracy stays high.
- Scanned documents are flagged for confidence, never promised perfect.
- We don't store your CoAs — see privacy & data handling.
Who builds it
Coatables is built by Metawear s.r.o. — metawear.cz. It's part of a small family of focused verification tools (sibling: Veto, which vets SQL before an AI agent runs it). Same principle: a deterministic check on the safety-critical step, not another black box.