Section 504 and the documents you generate

Section 504 of the Rehabilitation Act of 1973 applies to recipients of federal financial assistance from the Department of Health and Human Services. HHS's final rule extends it explicitly to web content and mobile applications, against WCAG 2.1 Level AA.

Most coverage of that rule discusses websites. The part that tends to get missed is that a health system's accessibility problem is usually not its website — it is the statements, explanations of benefits, appointment letters and test results it generates by the hundred thousand.

The dates moved, and a lot of published guidance is stale

The original compliance date was 11 May 2026. On 7 May 2026, HHS issued an interim final rule extending it:

An extension is not a reprieve for anyone generating documents at volume, because the work is not a one-time remediation project. It is a change to a pipeline, and then every document after it is correct by construction.

Why documents are the expensive part

A website is a fixed number of pages. You can audit them, fix them, and be done. Generated documents are unbounded: every statement produced next month is a new artefact, and if the template that made it produces untagged output, the template manufactures the problem faster than any team can clear it.

The market rate for remediating an existing PDF runs roughly $5 to $25 per page, depending on complexity, volume and turnaround. Apply that to a monthly statement run and the arithmetic answers itself: at 50,000 statements of two pages each, a single month of output is outside any plausible budget. Nobody remediates their way out of a pipeline.

Fixing generation is the only move that scales, and it has the useful property of being finite. It is one piece of work, once.

What "accessible" means for a statement, concretely

A patient statement is a table of dates, codes, charges and balances. Under PDF/UA-1 that means, at minimum:

• The table tagged as a table, with header cells tagged as headers and scoped, so a screen reader can announce that $482.10 is the patient responsibility for the 14 March line. Without that association, the document is read as a stream of unattached numbers — which is not a degraded experience, it is an unusable one.
• Headings in order, so the reader can move between sections rather than listening from the top.
• The clinic logo either given alternative text or marked as an artifact. Not tagged as a figure with nothing in it.
• Document language set, so the reader is not pronounced in the wrong language, and a real title rather than a filename.
• Fonts embedded, so text extraction does not break on another machine.

These are the same requirements that catch generated documents everywhere. Healthcare is simply where the volume and the sensitivity meet.

What a tool can honestly tell you

The PDF Association's Matterhorn Protocol breaks PDF/UA-1 into 136 failure conditions. 87 can be settled by software. 47 require human judgement. 2 have no defined test. A machine can prove a figure carries alternative text; it cannot tell you the alternative text is accurate.

So no software can certify that your statements comply with Section 504, and you should treat any vendor claiming otherwise with suspicion. What software can do is close the 87 conclusively and hand you evidence, so the human review that remains is small, targeted, and about the things that actually need a person.

Where Wellform fits

Wellform is an HTTP API: post the HTML you already generate, get back a PDF that conforms to PDF/UA-1 with a machine-readable conformance report attached to every document. Add "verify": true and the file is independently checked by veraPDF, the open source reference implementation the standards bodies maintain, with its verdict recorded next to ours and any disagreement shown rather than hidden.

That report is the artefact to keep. When someone asks what you did about Section 504, "every document we produced carries an independent conformance record" is a considerably better answer than a remediation invoice.

If your documents come out of a headless browser today, the reason they are failing is probably this. If you have been told your PDFs are already tagged, it is worth knowing why that is not the same claim.