If your organization produces public-facing PDFs, you need to adhere to PDF accessibility standards. Easier said than done.

Thanks to Jessica Chambers, CPWA, one of our on-staff accessibility experts; and James Graham, our Senior Digital Designer, for weighing in on this topic.

The Matterhorn Protocol, the gold standard for PDF accessibility, lists 136 specific failure conditions across 31 checkpoints. 

136 ways your PDFs can be inaccessible.

The protocol serves as the definitive testing framework for PDF/UA compliance under ISO 14289-1, the international standard for accessible PDFs. While many failure conditions can be detected automatically, nearly half (currently) require human judgment to evaluate properly.

The stakes became even more clear following federal compliance deadline extensions that revealed how many organizations’ automated testing had missed critical accessibility barriers. 

While it’s easy to fall into thinking that this complexity is just bureaucratic excess, please remember that roughly 1 in 4 adults need some sort of accessibility accommodations online. These 136 failure conditions represent a complete inventory of how digital documents exclude 25% of users.

But, as with anything, it’s not a perfect system.

Every way a PDF can exclude users

The Matterhorn Protocol operates as a validation checklist, not a remediation guide. Each of the 31 checkpoints addresses a specific aspect of PDF accessibility, from document structure to navigation elements.

Some examples, just to give you an idea of what we’re working with:

  • Checkpoint 01 focuses on semantically appropriate tags and contains 8 separate failure conditions including missing structure elements, incorrect tag nesting, and improper heading hierarchies. 
  • Checkpoint 07 examines dictionary and language settings across 4 failure conditions. 
  • Checkpoint 30 evaluates digital signatures through 3 specific ways they can violate accessibility standards. 

This granular approach means that each individual PDF element, like a data table, gets evaluated across multiple checkpoints for structure, headers, scope, and navigation.

The research on expert-validated accessibility annotations shows that common PDF accessibility failures cluster in predictable patterns, but each failure type requires different detection methods and remediation strategies. 

Matterhorn’s 136 failure conditions map directly to these real-world barrier patterns. In practice, this tends to mean that documents passing basic automated checks often fail dozens of Matterhorn checkpoints when subjected to human evaluation.

Real-world example: image alt text

Automated tools excel at detecting missing alt text but poorly evaluate whether “image of person” provides adequate context for the situation. We’ve written at length about why AI alt text leaves a lot to be desired, if you’re curious to learn more.

Research also suggests that AI systems consistently perform worse for users with disabilities across multiple domains, further outlines the fundamental limitation of algorithmic accessibility assessment.

Reading order validation illustrates this automation boundary most clearly: Software can reliably detect tag sequence but can’t judge whether the logical flow makes sense to someone navigating by screen reader. 

For example: Consider a PDF annual report where an automated scan confirms proper heading structure and tagged paragraphs, passing technical checkpoints. But human evaluation reveals that the screen reading order jumps from the executive summary to a sidebar advertisement, then to financial footnotes, before reaching the main content.

Table structure is similar. Automated tools detect missing header elements, but humans have to verify that the tables maintain logical relationships between data points, row headers, and column spans. The accessibility research published in ACM Digital Library confirms this pattern across multiple interaction contexts.

Compliance requirements per document

Manual verification of human-judgment conditions requires substantial time per document, depending on complexity and length. 

The U.S. General Services Administration Section 508 testing protocols keeps track of cases where automated pre-screening reduces total testing time. It identifies machine-detectable issues first, but acknowledges that complex documents with tables, forms, and embedded media still require checkpoint-by-checkpoint human review that cannot (yet) be meaningfully accelerated.

Real-World Testing Workflows

A financial report with embedded charts and data tables requires separate evaluation of logical flow, alternative text adequacy, table header relationships, and color contrast meaning across dozens of elements. The federal compliance assessment findings show that agencies initially underestimated testing duration by 60-70% when relying on automated tools alone.

Matterhorn’s thoroughness means most organizations need dedicated accessibility testing roles rather than adding PDF review to existing responsibilities. Content creators who attempt Matterhorn validation alongside their primary duties miss critical checkpoints or rush through contextual judgments that require sustained focus. 

It’s definitely something that needs a professional eye. For more context and specifics, please see our Senior Digital Designer’s excellent post about accessible PDFs.

Technical standards as real accessibility needs

Each of the 136 failure conditions exists because it represents a documented barrier that prevents assistive technology users from accessing content.

Missing landmarks force users to navigate line-by-line through entire documents to find content sections.

Incorrect heading hierarchy creates false document maps where H3 elements appear before H2 elements, leaving screen reader users lost in illogical content structures.

Untagged list structures present as disconnected text fragments rather than organized information, requiring manual interpretation of formatting cues that visual users process automatically. 

The research on learning barriers created by inaccessible documents demonstrates how these technical failures compound into genuine exclusion from information access. The human judgment requirements reflect accessibility realities where context and meaning cannot be automated, because this type of accommodation is contextual. 

Documents passing automated checks while failing human evaluation create the most frustrating user experiences because they appear accessible to testing tools while remaining functionally unusable. Technical compliance without contextual validation produces PDFs that appear accessible, but aren’t. 

PDF accessibility is hard. We won’t pretend it’s not. The 136 conditions laid out in Matterhorn represent the systematic catalog of ways that digital documents exclude disabled users from participating in information-driven society. 

It’s not acceptable to let people fall by the wayside just because it’s hard.

And, if we have anything to say about it, the process of checking for these conditions will get faster and more consistent soon in the future.

References

  1. ADA.gov. (2024, March 8). Fact Sheet: New Rule on the Accessibility of Web Content and Mobile Apps Provided by State and Local Governments. ADA.gov. https://www.ada.gov/resources/2024-03-08-web-rule/ 
  2. Akiba, D., Pagliara, S., Truss, M., Nwokoye, C., & Waters, G. (n.d.). AI testing, evaluation, verification and validation for accessibility: a comprehensive framework. Front. Digit. Health, 7, 1679603. https://doi.org/10.3389/fdgth.2025.1679603 
  3. cmtWPAdmin. (2025, May 2). New Federal Requirement: Digital Accessibility Compliance by 2026 – Accessibilty. Umich.edu. https://accessibility.isr.umich.edu/new-federal-requirement-digital-accessibility-compliance-by-2026/ 
  4. Extension of Compliance Dates for Nondiscrimination on the Basis of Disability; Accessibility of Web Information and Services of State and Local Government Entities. (2026, April 20). Federal Register. https://www.federalregister.gov/documents/2026/04/20/2026-07663/extension-of-compliance-dates-for-nondiscrimination-on-the-basis-of-disability-accessibility-of-web 
  5. International, F. (2025, November 20). Accessible Documents: Ensuring Access for Everyone. Fiu.edu. https://core.fiu.edu/blog/2025/accessible-documents-ensuring-access-for-everyone.html 
  6. Kumar, A., Padath, T., & Wang, L. L. (2025). Benchmarking PDF Accessibility Evaluation: A Dataset and Framework for Assessing Automated and LLM-Based Approaches for Accessibility Testing. Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility, 1–24. https://doi.org/10.1145/3663547.3746380 
  7. moiala, linda. (2026). Matterhorn Protocol 1-02. Scribd. https://www.scribd.com/document/375542639/Matterhorn-Protocol-1-02 
  8. PDF/UA-1, PDF Enhancement for Accessibility, Use of ISO 32000-1. (2024). Loc.gov. https://www.loc.gov/preservation/digital/formats/fdd/fdd000350.shtml 
  9. Section508.gov. (n.d.). Www.section508.Gov. https://www.section508.gov/test/testing-overview/ 
  10. Tapping Technology – Link Global: 508 and the Facts. (2026). Maryland.gov. https://msa.maryland.gov/megafile/msa/speccol/sc5300/sc5339/000113/000000/000451/unrestricted/20040944e/art_4.html 
  11. U.S. Access Board – General Services Administration Publishes Annual Governmentwide Section 508 Assessment. (2026). Access-Board.gov. https://www.access-board.gov/news/2026/03/05/general-services-administration-publishes-annual-governmentwide-section-508-assessment/ 
  12. Wallace, E. (2026, April 13). Stony Brook CS Researchers to Present Seven Papers at CHI 2026, Showcasing Advances in AI, Accessibility and Interaction Design – SBU News. SBU News – News & Features at Stony Brook University. https://news.stonybrook.edu/computer-science/stony-brook-cs-researchers-to-present-seven-papers-at-chi-2026-showcasing-advances-in-ai-accessibility-and-interaction-design/