Tagged and archival PDFs from ReflowPDF: the files, and how to check them
Exports that validate as PDF/UA-1, PDF/UA-2 and PDF/A-3a in veraPDF, each with its report. What the tags are, what the archive adds, and what no validator can judge for you.
, updated , a 5-minute read
Each file below came out of ReflowPDF, and each link beside it is the veraPDF report it was published with. Check them yourself in veraPDF, the reference validator from the PDF Association.
| File | Claims | veraPDF |
|---|---|---|
| Invoice | PDF/UA-1 | report |
| Financial report, several pages of tables | PDF/UA-1 | report: 9,417 checks passed, 0 failed |
| CV | PDF/UA-1 | report |
| Council agenda | PDF/UA-1 | report |
| Council agenda | PDF/UA-2 | report |
| Financial report, 4 pages | PDF/UA-2 | report: 22,745 checks passed, 0 failed |
| Invoice | PDF/A-3a | report |
| Financial report | PDF/A-3a | report: 26,228 checks passed |
What a tagged PDF is
To a screen reader, an untagged PDF is glyphs scattered at coordinates, with nothing to say which of them form a heading or a table, or in what order to read them. A tagged PDF carries a second, invisible layer, a structure tree (Document → H1 → P → Table → TR → TH/TD …), that says what each piece of content is. PDF/UA (ISO 14289) is the standard that says how to build that tree correctly.
Building the tree means knowing the structure. An engine that turns a page into drawing operators and keeps nothing else has nothing to tag. Some engines do tag (Prince, Antenna House and PDFreactor among the commercial ones; WeasyPrint has a PDF/UA-1 variant), and many popular ones, dompdf and wkhtmltopdf among them, do not.
ReflowPDF tags because of what it edits. A heading in the editor is a heading and a table is a table, because that is the document you work on, not a guess from font sizes. When the PDF is made, the engine walks the same tree and writes the matching tags. By default every export carries:
- a structure tree (
/StructTreeRoot) with/MarkInfo /Marked true; - marked content tying every glyph and image to the tree; backgrounds, borders and rules marked as typed artifacts so readers skip them;
- headings tagged from H1 without skipped levels (an H4 after an H2 is tagged H3), while the editor points out the skip in the document itself;
- tables as
Table → THead/TBody → TR → TH/TD, with/Scopeon every header cell, also when a table continues on the next page; - lists as
L → LI → Lbl, LBody; - figures with
/Alt, and links with/Contents; - the document’s language and title, and the
pdfuaididentifier.
“Tagged PDF” can be switched off in Page settings for a smaller file; the API option is tagged: false.
How it is checked
In June we ran every built-in template of the time through veraPDF, and 16 of the 50 failed. The validator found real faults: artifacts opened with the wrong operator (/Artifact BDC where the specification wants BMC), table rows that lost their table when it continued on a new page, headings that skipped a level, images without a description. Each is handled in the engine now, for any document.
Since then the checks have grown:
- veraPDF, before an engine release reaches the editor, over tagged documents that include the templates and the files above. A file that stops passing is a fault in the product.
- PAC, as a second opinion. It implements the same standard independently and reports what the machine rules leave to judgement. It found faults veraPDF passed: bare
/Artifactregions that tell a screen reader nothing (now typed/Layoutand/Pagination), SVG figures without a/BBox, footnotes no link pointed to. All fixed in the engine. - The rules a file exercises. The veraPDF PDF/UA-1 profile has 106 rules. A rule about something the file does not contain (encryption, optional content, multimedia) passes because there is nothing to check, so a clean report says less than it seems; a reference document that contains every construct the engine writes is the fairer test: the PDF/UA-1 reference document contains each construct once.
Archival: PDF/A-3a
PDF/A (ISO 19005) is the archival standard: a file that opens and looks the same decades from now. Every font is embedded, colour is tied to an embedded ICC profile, and encryption and outside references are not allowed. The suffix says what else is promised:
- 3, the part: files of any kind may be embedded in the archive. PDF/A-2 allows embedded files only when they are PDF/A themselves; PDF/A-1 allows none.
- a, the level: the full tag tree, the same structure a screen reader needs. Level b promises only how the pages look.
ReflowPDF writes the archival layer on top of the tags: it embeds every font it used, adds an sRGB output intent, writes the pdfaid identifier, and does not encrypt. With Keep the PDF editable on, the document’s editable source goes into the archive as an associated file (/AFRelationship /Source), encrypted so that only ReflowPDF opens it, and the archive opens in ReflowPDF as a document again; the PDF itself is not encrypted.
PDF/A-3a is a choice, not the default: the archive standard asks for a little more in the file, an output intent with an ICC colour profile (about 4 KB), and it pairs only with the first edition of PDF/UA. Turn on Export as PDF/A-3a in Page settings on the Pro and Max plans, or send pdf_a: "3a" to the developer API.
Where a validator stops
No tool can make a document fully conformant by itself, because PDF/UA asks for human judgement where software cannot decide:
- Is the alt text meaningful (“Q1 revenue, up 12%”) or not (“image1”)?
- Is the reading order the order of the argument?
- Are headings used for structure, or only because the text was big?
The editor helps with the human half: it asks for alt text and offers a “decorative” switch, takes the language in Page settings and the title from the document’s name at the top left, and points out headings that skip a level. The words are still yours, and for documents with legal weight a human review is the right last step.
A checklist for any PDF, from any tool:
| Question | How to check |
|---|---|
| Is it tagged at all? | pdfinfo file.pdf → Tagged: yes |
| Does it pass the machine rules? | verapdf -f ua1 --format text file.pdf → PASS |
| Does a second implementation agree? | Open it in PAC and read the warnings, not only the errors |
| Are decorations artifacts? | They should be typed artifacts (/Layout, /Pagination) |
| Does every figure have meaningful alt text? | Read each one out loud; no tool can judge this |
| Does the file match its claim? | head -c 8 file.pdf, and the catalogue’s /Version if the header is older: PDF/UA-1 and PDF/A-3a need PDF 1.7 or earlier (our engine writes %PDF-1.7), PDF/UA-2 needs PDF 2.0 |
The second edition of PDF/UA is written against PDF 2.0 and has its own post.
The laws that ask for it
The European Accessibility Act applies since June 2025; EN 301 549 governs public procurement in the EU; Section 508 requires an authoring tool that exports PDF to be able to export PDF/UA-1; and the US Department of Justice’s rule under ADA Title II sets dates for state and local governments, which our page for them covers. If you publish under any of them, the PDF has to be tagged.
Questions
- Where can I get an example PDF/UA file to test?
- The files at the top of this article are exports from ReflowPDF, each with the veraPDF report it was published with: an invoice, a financial report and a CV as PDF/UA-1, a council agenda as PDF/UA-1 and PDF/UA-2, a four-page financial report as PDF/UA-2, and an invoice and the report as PDF/A-3a.
- What is a tagged PDF, and what is PDF/UA?
- A tagged PDF carries a structure tree that says what each piece of content is (heading, paragraph, list, table, figure, link) and in which order to read it, plus the document’s language. PDF/UA (ISO 14289) is the standard for doing that correctly. Without tags, a screen reader has only positioned glyphs.
- What does PDF/A-3a add?
- PDF/A (ISO 19005) makes the file self-contained for the long term: every font embedded, colour tied to an embedded ICC profile, no encryption and no outside references. Level a requires the full tag tree, and part 3 allows files of any kind to be embedded.
- Is a ReflowPDF export certified as PDF/UA?
- No tool can certify a document on its own: PDF/UA asks for human judgement on alt text, reading order and the use of headings. Our exports validate against veraPDF, the reference validator; the parts a machine cannot judge stay with the author, as with any authoring tool.
- How do I export PDF/UA-2 or PDF/A-3a in ReflowPDF?
- In Page settings: “Claim PDF/UA-2” or “Export as PDF/A-3a”, on the Pro and Max plans. Tagged PDF/UA-1 is the default on every plan. Through the developer API the options are pdf_ua: 2 and pdf_a: "3a".
Check a PDF of your own against PDF/UA-1 with veraPDF, free, in the browser.
More from the blog
PDF/UA-2 (ISO 14289-2): what changes, and how we verify it
PDF/UA-2 is written on PDF 2.0. Five changes from PDF/UA-1, how to check each, and what veraPDF does not check., a 9-minute readThe Matterhorn Protocol: 136 ways a PDF can fail PDF/UA
PDF/UA-1 checkers test the conditions of the Matterhorn Protocol: 31 checkpoints, 136 failure conditions, and the split between machine checks and human judgement., a 3-minute readHow we measure CSS conformance on the web-platform-tests
We changed the unit behind our conformance headline: a reftest counts as one test, a scripted file counts as its subtests. Both scales, same dataset., a 5-minute read