A claim like “passes PDF/UA” is cheap without numbers. Here is what was measured on TagSmith-PDF 0.7.0, how, and where it stops.
The measurement
On 12 October 2026 the release build repaired a set of real published PDFs, and veraPDF 1.30.2 validated every result against the PDF/UA-1 profile. A file counts as passing only when veraPDF reports no failed rules.
| Document set | Fully PDF/UA-1 conformant |
|---|---|
| EU institutional PDFs (EUR-Lex) | 31 of 31 |
| Dutch national reports (RIVM) | 40 of 40 |
| US tax forms (adversarial control set) | 26 of 26 |
| All real public-sector PDFs | 97 of 97 |
Across those 97 files, veraPDF counted 62,899 failed checks before the repair and 0 after. The tax forms are in the set on purpose: they are form-heavy and awkward, so they act as a control that is hard to pass by accident.
The demo set
The 97 documents are real, published files and cannot be redistributed. To let anyone see the output, there is also an invented set, a fictional town called Harbourford: Word, Excel and HTML files, a scan and some PDFs. TagSmith-PDF made all 7 conformant with 65 fixes, and wrote 29 picture descriptions during the run. The originals, the repaired files and the reports are on the product page.
The software tests
The release build ran 344 automated tests: 0 failures, 0 errors, 2 skipped.
What this does not prove
- 97 documents is a sample. It shows the engine handles many kinds of file. It does not promise that every PDF will pass.
- A veraPDF pass is not a certification. The machine-checkable rules hold, and a person still has to judge picture descriptions, reading order and contrast.
- The built-in AI is small and it can be wrong. That is why its descriptions stay drafts until someone accepts them.
- These numbers describe 0.7.0 only. A later version is a different build.
The method is scripted so the same measurement can run on every release. If you want to check the claim on your own files, the app has a 14-day trial with every feature switched on. Details are on the product page.