Data about disabled students, in a tool they could not use

States report data to the federal government about children with disabilities under the Individuals with Disabilities Education Act — Part B for school-age students, Part C for infants and toddlers. Who is being served, under what category, with what outcomes, and where the gaps are. It is how the country finds out whether children with disabilities are actually getting the education they are entitled to.

The tools for reading that data were, with some regularity, unusable by people with disabilities.

That is not an irony worth being clever about. It is a straightforward failure with a specific cost: the people with the strongest claim to understanding this data — disabled adults, parents of disabled children, advocates, researchers who are themselves disabled — were the ones most likely to be locked out of it.

What we built

I led accessibility and user experience at the Rhonda Weiss Center for Accessible IDEA Data, which AEM Corporation leads under a multiyear grant from the U.S. Department of Education's Office of Special Education Programs. The Center's job is not to publish the data itself — it is to improve the capacity of states to publish it accessibly, which is a harder and more durable problem than fixing one tool.

We built an accessible reporting system, developed with around 80 stakeholders, including people with disabilities.

Two decisions shaped the rest of it.

We put accessibility decisions into design tokens, extending the U.S. Web Design System rather than building beside it. A colour that passes contrast is not a preference someone applies correctly; it is a value a downstream team inherits whether they are thinking about it or not. The same goes for focus treatment, text sizing and the palettes used for categorical data — the last one matters enormously in a reporting tool, because a chart with colour as its only channel is a chart that excludes people before they have read a number.

We tested with disabled participants continuously — over thirty studies, including blind and neurodivergent participants — rather than validating at the end. That is a deliberately expensive choice and the only one that finds the problems that matter. Automated checking finds a missing label. It does not tell you that your table of state-level outcomes is technically navigable and practically impossible to hold in your head at screen-reader pace.

The delivered tool reached the highest applicable conformance level. I am careful with that sentence: it is a statement about the tool we delivered, on the criteria that applied to it, and not a claim about everything the department publishes.

Data visualisation is where accessibility usually quits

Most teams do reasonably well on forms and navigation, then hit charts and give up, because the honest answer — that a visualisation needs a real non-visual equivalent, not alt text saying "bar chart" — is expensive.

Equitable data representation was most of the design work here, and I have spent a fair amount of time since arguing it publicly, at CSUN and Closing the Gap among others. The short version: the question is not how to describe the picture. It is what the reader is trying to find out, and what the shortest path to that answer looks like when the picture is not available. Those are frequently different structures, and the second one is rarely just a table.

Why this one stays with me

Most accessibility work is about not excluding someone from a transaction. This one was about who gets to participate in an argument — about whether the education system is serving children with disabilities, using the evidence that is supposed to settle it.

A parent who cannot read the report cannot challenge the report.

Where this goes

The pattern I would take everywhere from this is the token one: whenever an accessibility decision can be made once, centrally, and inherited, it should be — because the alternative is asking hundreds of people to remember it, and they will not, and no one will find out until someone is already shut out.

It is the same argument I now make about design systems in code and about machine-readable standards. This was where I first watched it work at the scale of a federal reporting programme, with the people it was about in the room.