From XML to Accessible PDF: Full Automation with XSL-FO and PDF/UA

Creating accessible PDFs sounds simple, but WCAG, PDF/A-1a, and PDF/UA are ruthless about structure, semantics, and metadata.
A single mistake in tag order, table structure, or missing metadata — and the PDF is no longer accessible or archival-grade.

In complex publication chains, with documents from many sources, manual fixes simply don’t scale.

At Elk Solutions we automate the entire process. Our pipeline starts with XML as the uniform source — no matter where that XML comes from: databases, STOP/TPOD, legal systems, CMS, custom files, or specialized schemas.

The Problem: The Illusion of “Complete” PDF Export

Many organizations rely on manual or semi-automatic PDF export. Even with decent tools, structural issues persist:

• Incomplete or incorrect tag structure
• Headings and paragraphs not recognized
• Lists that look right visually but are semantically wrong
• Tables missing th info or with broken cell spans
• No readable navigation for screen readers
• Incorrect or missing XMP metadata

Accessibility becomes a “nice to have”, while WCAG and archiving standards make it mandatory.

Reality:

• Manual validation fails.
• Post-fix is expensive and error-prone.
• Consistency is impossible when sources vary.

The Solution: XML as Source + XSL-FO as Layout

Where traditional pipelines “cut and paste” PDFs, we take a different route.
We use XML as the foundation — structured, predictable, semantically rich.

Then we render XML to XSL-FO, which gives us pixel-perfect and semantically correct output.

XSL-FO gives full control over:

1. Document structure (headings, paragraphs, sections)
2. Lists (bullets, nesting, numbering)
3. Tables (headers, body, cell relations)
4. Alt text and descriptions
5. Metadata for PDF/A-1a and PDF/UA
6. Reading order and logical structure

Because XML is the source, it doesn’t matter how it was created.
Everything becomes uniform, repeatable, and 100% controllable.

The Engine: Smart Structure Recognition & Validation

Our transformation engine does more than “convert”.
It understands structure and fixes what’s missing.

Examples:

• Normalizing heading levels
• Checking and rebuilding list structures
• Enriching or repairing table structures
• Injecting missing semantics
• Assigning roles, alt text, and relationships
• Deriving logical reading order

These steps are built around international standards:

• PDF/UA (ISO 14289) — https://pdfa.org/resource/iso-14289-pdfua/
• PDF/A-1a: full tagging and XMP integrity
• WCAG 2.2: accessibility for non-HTML docs
• Matterhorn Protocol: 136 checkpoints for PDF/UA validation

By baking this knowledge into the engine, accessibility is not a bolt-on step but integral to the publication flow.

The Result: Automatic, Valid PDFs

1. Perfectly Tagged PDF/A-1a + PDF/UA

Each PDF is delivered with:

• full semantic tags
• correct heading and list structures
• table recognition with headers and scopes
• correct reading order
• language and document metadata
• embedded, accessible fonts

2. Scalable, Fast, Repeatable

Applies to:

• legal documents
• policy papers
• inspection reports
• dynamic data sheets
• STOP/TPOD publications

— always identical output. No dependency on individual users or Office setups.

3. Automatic Validation Chain

We validate by default with:

1. veraPDF — official open validator for PDF/A
   https://verapdf.org
2. PAC (PDF Accessibility Checker) — strictest PDF/UA validator
   https://pac.pdf-accessibility.org/en
3. Matterhorn Protocol — full PDF/UA checklist

This is not optional: every PDF passes this chain before it leaves the pipeline.

Conclusion: Why This Works

Starting from XML and using XSL-FO for consistent layout yields a publication chain that:

• Guarantees accessibility, not “approaches” it
• Meets standards without post-processing
• Is fully repeatable, independent of tools or people
• Scales from one document to hundreds of thousands
• Stays future-proof, because XML and FO are open standards

Accessible PDFs don’t have to be a last-minute fire drill.
They can be guaranteed output of a smart, structural pipeline.

Don’t start at the PDF.
Start at the structure — and everything else follows.