Accessibility and universal design for learning specialization

What we are working on together in accessibility, and where my practice participates

One problem, followed through five places where learning design helps decide whether a right to access becomes access people can use: from obligation, to the people it is meant for, and back again.

This is the linear version of the Watershed page. It holds the same content with every section expanded, in reading order, with no interactive controls.

The idea

Why a watershed

A watershed is all the land that drains to one river. What happens upstream, where the rain falls and what the ground holds, decides what reaches the river downstream.

Access works the same way. Decisions made early, about who a course, document or service is for and what it assumes, decide who has to ask later. Many people already work on each part of this: disabled people and their organizations, accessibility committees, advisors, researchers and standards bodies. My practice is one contribution among them, and this page shows where it joins.

The one problem

Access promised in law and policy does not reliably become access in practice

Four measures show the size of the gap in Canada and online: how many people it concerns, how work differs for them, how often the access people ask for arrives, and how much of the web fails basic checks. Each figure is shown with what it measures and what it does not.

Prevalence

27% of Canadians aged 15 and older, 8.0 million people, had at least one disability that limited their daily activities in 2022.

Measures
The Canadian Survey on Disability, Statistics Canada’s official source on disability. The rate rose from 22% in 2017.
Does not show
People living in institutions and other collective dwellings, on First Nations reserves or on Canadian Armed Forces bases, who are outside the survey. It is self-reported.

Work

62% of Canadians aged 25 to 64 with disabilities were employed, against 78% of those without disabilities.

Measures
Employment in May 2021 from the Census, linked to disability status from the 2022 survey. The rate falls to 30% among people with very severe disabilities.
Does not show
Job quality, pay or fit. The reference week fell during the COVID-19 pandemic.

Accommodation

54.4% of Canadians with disabilities aged 25 to 64 who required workplace accommodations did not receive all of them in 2022.

Measures
Unmet workplace accommodation needs, from the Canadian Survey on Disability. The share fell from 62.2% in 2017.
Does not show
Which needs went unmet, or why. It covers the workplace, not education or services.

Digital

95.9% of the top one million home pages had detectable failures of the Web Content Accessibility Guidelines in 2026.

Measures
An annual automated scan of the world’s most visited home pages. The share rose from 94.8% in 2025.
Does not show
Failures automated tools cannot detect, so the true figure is likely higher. Home pages only, not whole sites.

Read together, the figures say something more precise than that a gap exists. More than one in four Canadians aged 15 and older has a disability. Employment rates differ by sixteen points. More than half of working-age people with disabilities who need workplace accommodations do not receive all of them. Nearly every major home page fails checks that are well understood and largely fixable. Worldwide, an estimated 1.3 billion people experience significant disability, and some die up to 20 years earlier than people without disabilities.

The gap is not mainly a shortage of rights or standards. Canada and British Columbia have both. It sits between obligation and practice: in whether access is designed in or left to be requested, who carries the work of asking, whether arrangements hold when needs change, and whether anyone checks. That is the ground learning design works on.

Selection

Why these five

Many problems are large. These five were chosen because each passes three tests at once, and because they depend on each other.

Each is large: each touches a large share of the 8 million Canadians with disabilities, and many more worldwide. Each is neglected in a specific way: not for lack of law or goodwill, but because the part that decides the outcome, the design of what people actually meet, is rarely resourced as carefully as the policy around it. And each is tractable: the means to move it are known, and many cost little. The strength of the evidence differs from case to case, and each case says what it rests on.

They are also one system. A plan that never reaches daily practice changes nothing. Practice that depends on requests misses the people least likely to ask. Processes that are slow or repeated wear down the people who do ask. Arrangements that are never reviewed fail when needs change. A gain in one place is lost in the next unless the next is designed too, which is why the cases are read as one watershed and not five problems.

The fifth is different in kind. Artificial intelligence changes the conditions for the other four at once: how materials are made and adapted, how captions and image descriptions are produced, how plans are put into practice, and how anyone checks whether access holds. It is the climate the other four work in.

The map

Five tributaries, one river

Read in order, the stages follow the water. Each stage flows into the next. What stage 4 finds returns to stage 1. Stage 5 affects every stage at once.

  1. Obligations into practice (The source). A duty in law becomes something staff can do on an ordinary day.
  2. Access designed in, not requested (Who can use it). Courses, documents and services work for disabled people from the first draft, without depending on a request.
  3. The work of asking (Who carries it). The work of securing access is shared fairly, and no one repeats a request that a better process would make unnecessary.
  4. Needs that change (Feedback). Access is reviewed as needs change. What is learned returns to stage 1.

The river: people taking part on equal terms. Where the four stages lead. Each stage decides how much access is lost on the way.

Stage 5, the climate: Access through the AI transition. Changes how materials are made, adapted and checked at every stage at once.

How each case is read

Three questions, asked of every case

Each question is asked twice: what is, and what could be. Scale asks how many people a problem affects, and how much. Neglectedness asks how many resources already go to it, and where one more contributor would add most. Solvability asks how much of what remains more effort would solve, and what it would take.

Scale
What is: how many people does it affect, and how much? What could be: if it were solved, how much better would things be, and for how many people?
Neglectedness
What is: how many resources already go to it? What could be: where would one more contributor add most?
Solvability
What is: how much of what remains would more effort solve? What could be: what would it take?

Each case also names what design cannot do alone: the conditions held by others, such as leadership, funding, staff time, procurement and legal advice on what the law requires.

The cases

Five cases, each read the same way

1. Obligations into practice

The source

How does a duty in law become something staff can do on an ordinary day?

What solving it would make possible: accessibility plans that change what people are trained to do, how procedures run and how courses are built, reviewed against what people report.

The evidence: In B.C., more than 750 public sector organizations must have an accessibility committee, an accessibility plan and a way to receive public feedback, and must review the plan at least every three years. Ontario’s 2019 review of its accessibility law found progress far slower than the law intended.

Where my practice participates: translating accessibility plans into training, procedures and course design that teams can sustain, worked out with the people the organization serves.

What it assumes: Plans have leadership support and staff time, and the people who use the organization’s services are part of writing and reviewing them.

2. Access designed in, not requested

Who can use it

How can courses, documents and services work for disabled people without depending on a request?

What solving it would make possible: materials and services people can use on the same day as everyone else, with individual accommodation kept for needs that are genuinely individual.

The evidence: Of the top one million home pages tested in 2026, 95.9% had detectable accessibility failures, and about 96% of detected errors fell into six common types. A 2023 meta-analysis of 20 studies found a moderate positive effect of universal design for learning on achievement.

Where my practice participates: inclusive learning design and universal design for learning applied to courses, training and learning systems, and accessibility reviews of what already exists.

What it assumes: The platforms, devices and assistive technology people use can work with what is built, and teams are given time to build it well.

3. The work of asking

Who carries it

Who carries the cost of getting access, and how much of that work is necessary?

What solving it would make possible: accommodation processes that are clear and rarely repeated, with requirements that state the capability needed rather than one way of performing it.

The evidence: In 2022, 54.4% of Canadians with disabilities aged 25 to 64 who required workplace accommodations did not receive all of them. Among U.S. employers who reported costs to an advisory service, 61% said the accommodations they made cost nothing, and 33% reported a one-time cost, with a median of US$300.

Where my practice participates: research on how Canadian professional programmes write their essential requirements, the standards applicants measure themselves against before any accommodation process begins.

What it assumes: Requirements can be revised by the people who hold them, and the process can be simplified without losing what a requirement genuinely protects.

4. Needs that change

Over time

Do access arrangements hold when needs change, and does anyone check whether they worked?

What solving it would make possible: access that is reviewed as needs change and adjusted without starting again, with what is learned returned to the plan.

The evidence: In 2022, 64.2% of Canadians with disabilities had dynamic disabilities: limitations that are progressive, recurrent or fluctuating. Mental health–related disabilities increased more than any other type between 2017 and 2022.

Questions I am following: how the work of securing access and changing needs interact over time, studied through the social model of disability, which locates barriers in systems rather than in people.

What it assumes: Feedback reaches the people who can act on it, and arrangements can be adjusted without a new request each time.

5. Access through the AI transition

The climate

How can AI widen access without people having to take its accessibility on trust?

What solving it would make possible: AI tools that widen access, judged against standards and by disabled people before they are relied on.

The evidence: In December 2025, Accessibility Standards Canada published a national standard for accessible and equitable AI, developed by a committee made up primarily of people with disabilities and members of equity-deserving groups. In 2025, the U.S. Federal Trade Commission finalized an order requiring an AI accessibility company to pay US$1 million over claims that its tool could make any website meet accessibility guidelines.

Where my practice participates: applying the same accessibility review to AI-generated captions, image descriptions and learning materials before they reach learners.

What it assumes: AI-generated alternatives are checked before use, and access to the tools is fair across the people an organization serves.

The theory of change

How the five connect

If access is designed so that:

  1. obligations are translated into everyday training, procedures and design,
  2. with access built in from the first draft and disabled people leading,
  3. with the requests that remain kept clear, fair and rarely repeated,
  4. reviewed over time as needs change, with what is found returned to the plan,
  5. and AI tools are checked against standards and by disabled people before they are relied on,

Then more people can take part on equal terms, more of the time, with fewer of them having to ask,

and organizations learn faster which designs remove barriers, because what feedback finds returns to the plan.

This holds only where conditions outside design are in place or held by partners: leadership, funding, staff time, procurement, and legal advice on what the law requires. It describes a contribution, not a sole cause.

The conditions are joined, which is why the theory is stated as one chain and not five. A plan reaches no one if it never changes daily practice. Practice designed for access still fails people if the remaining requests are slow and repeated. A good process still fails if arrangements are never reviewed as needs change. Review that never returns to the plan teaches the organization nothing, so the next plan starts from the same place. Each link is an assumption that can fail, and each case names its weakest one.

Limits

What the evidence does not yet show

Four limits run through all five cases, and are stated here.

Most Canadian figures are self-reported and come from people living in private households. People living in institutions and other collective dwellings, on First Nations reserves or on Canadian Armed Forces bases are outside the national disability survey.

Some of the evidence comes from elsewhere. The cost figures for workplace accommodations come from U.S. employers who contacted an advisory service, who may differ from employers generally. Whether the same costs hold in Canada is an open question.

Automated checks find only part of the problem. The web figures count failures that software can detect. Barriers that need a person to judge, such as whether an image description is accurate, are not counted.

The provincial requirements are recent. B.C. adopted the regulation requiring public sector organizations to have accessibility committees and plans in 2022, and organizations must review their plans at least every three years. Whether the plans change what people experience will become clearer as the first review cycles finish.

My practice

Where my practice participates

My practice joins at three of the five places: obligations into practice, access designed in, and the work of asking. At the fourth, needs that change, it joins as a question I am studying. At the fifth, it applies the same accessibility review to materials made with AI.

The method is Fieldwork, practised with the learners an organization actually serves, not on their behalf. Interpreting what the law requires sits with legal counsel. This is design work.

How I know it is working

The work is designed against seven levels.

Reach
Did it reach the people the default usually misses? Example: accessible formats available at release, not on request.
Requests
Did fewer people have to ask for what should already have been there? Example: fewer accommodation requests for materials that can be built accessibly.
Practice
Did what people do change, checked at 60 to 90 days? Example: a follow-up on how staff now build materials, not on satisfaction.
Adoption
Are others choosing to use or adapt it? Example: an accessibility office adapting a published resource or method.
Maintenance
Is it kept current? Example: a named owner and a review cycle that matches the accessibility plan.
Embedding
Is it in a template, procedure or standard? Example: part of onboarding, a course template or a written procedure.
Field
Is it being cited, critiqued or built on? Example: citations, published critiques, or contributions to a standard.

Sources

References