Learning design for neurodivergent learners

The learning-design industry has a stable, well-funded answer to neurodivergentneurodivergent. Having a mind that works differently from the typical — e.g. autistic, ADHD, dyslexic. A descriptive word, not a deficit label. learners: build for the majority, then bolt on accommodations for the edge cases. Closed captions on request. Extra time on exams, if you file the paperwork. A quiet room, three weeks after you asked. The model treats neurodivergent cognition as a deviation from a neutral default, and treats the accommodation as a favor the default graciously extends. Almost every piece of learning software you’ve used — every LMS, every corporate compliance course, every MOOC — was designed this way.

The trouble is that the default isn’t neutral. It’s a design choice, made on behalf of a neurotypical majority that mostly suffers it badly too. Sync the schedule, lecture orally, fragment the day, reward streaks, assess by social performance — these aren’t laws of cognition, they’re conventions of institutions. And here’s the part the industry keeps rediscovering: when you design from neurodivergent cognition instead of accommodating it afterward, the result tends to be better for everyone. This is the curb-cut effect applied to learning. Curb cuts were mandated for wheelchair users and are now used by strollers, suitcases, and delivery carts; a recorded, captioned, structured lecture was demanded by disabled students and is now used by every student who ever got sick, got tired, or got confused at minute forty. Designing for the edge has a funny habit of improving the center. (If you want the frame itself taken apart first — including the honest cases where the curb cut isn’t free — read the accommodation tax is a myth, the companion argument this piece builds on.)

The claim has a name and a provenance, which matters because it is doing a lot of work. Angela Glover Blackwell coined the curb-cut effect in 2017 to describe exactly this pattern: a provision won by one excluded group, which then becomes infrastructure everyone uses. And the learning version has direct evidence behind it, not just intuition — Gernsbacher (2015) reviewed the captioning research and found that captions improve comprehension and retention for viewers without hearing loss too. Not a concession granted to a minority: an improvement to the medium. That is the standard the rest of this argument should be held to. Where it meets the standard, it is worth saying plainly; where it doesn’t, saying so is the point.

So this piece is the reading list in reverse. Instead of asking “what do autistic and ADHD learners need patched in,” ask “what does learning design look like if their cognition is the starting point.” The answer is concrete, specific, and mostly cheap — and it maps almost one-to-one onto what good credentialing infrastructure should have been doing anyway.

The default is not neutral

Universal design, as practiced, is usually neurotypical design with ramps added. The architecture is unchanged: real-time oral instruction, synchronous attendance as a proxy for engagement, social fluency as an implicit assessment criterion, fragmented days dressed up as “engagement.” Then the ramps: a captioning toggle here, a deadline extension there, each one opt-in, each one visibly special, each one generating administrative labor on both sides of the request.

The reframe is simple and it matters: there is no view from nowhere in learning design. Every platform has a theory of the learner baked in — how that learner parses language, sustains attention, recovers from a bad day, demonstrates mastery. The question is only whether the theory is stated or smuggled in. When it’s smuggled in, it’s almost always a theory of the neurotypical learner, and the accommodations budget is what’s left for everyone else.

Designing from neurodivergent cognition produces a different architecture, and the first thing you notice building it is how often the “accommodation” is just good design that the neurotypical default was too lazy to build. The needs below aren’t exotic. They’re specific.

What autistic learners tend to need

Written over oral. This is the single highest-leverage change. Oral instruction is fragile: it’s real-time, it isn’t searchable, it’s gone the moment it’s spoken, and it quietly smuggles in a social-parsing requirement — the learner isn’t just processing content, they’re parsing tone, intent, and the instructor’s implicit expectations all at once. Written instruction is re-readable, searchable, and identical for every reader. A learner who needs to read the spec three times isn’t being high-maintenance; they’re using the medium the way the medium works. The same written spec, incidentally, is what the neurotypical colleague pastes into the ticket six months later. Everyone wins; one group was just unwilling to pretend otherwise.

Structure and predictability. Clear processes, defined roles, low ambiguity. Not because autistic learners are rigid, but because ambiguity is load — every ill-defined rubric and “we’ll figure it out live” meeting is cognitive budget spent on decoding the frame instead of learning the content. Explicit structure is a transfer of cost from the learner to the designer, which is where it always should have lived. And here again the majority follows behind: nobody ever complained that a rubric was too clear.

Hyperfocus honored, not scheduled around. Deep, sustained attention is the autistic cognitive superpower and the modern calendar is its kryptonite. Long uninterrupted blocks produce extraordinary throughput; the fragmented “collaborative” day — standup at nine, sync at eleven, review at one, workshop at three — shatters it. This isn’t anecdote. In the research summaries compiled by autistic-led communities, AuDHD participants reported that attending standup hurt their wellbeing — the very ritual designed to “include” the team functioning as a daily tax on the people who most need unbroken time. (Autistic-led research priorities point in this direction repeatedly — see what autistic people actually want researched; the co-occurrence cluster — autism, ADHD, EDS, POTS, MCAS, and the rest — means these needs compound rather than take turns, as laid out in the AuDHD/EDS/POTS cluster.)

Interest as engine. Autistic and ADHD motivation systems run on interest, not on external reward mechanics. A learner engaged with the topic will out-produce any incentive scheme you can design; a learner disengaged by it cannot be bribed into mastery with points, badges, or streaks. Design that treats interest as the engine and lets the learner route toward depth will see comprehension numbers no streak mechanic ever produced.

What ADHD learners tend to need

The ADHD list overlaps the autistic list and diverges in instructive ways — and, worth naming: a large fraction of your autistic learners are AuDHD, so the lists are co-resident in the same person, not two personas you can A/B test.

Time-boxing and externalized structure. The ADHD executive system struggles to generate schedule internally and benefits enormously when the structure is externalized: Pomodoro blocks, visible time-boxing, working sessions with explicit starts and ends. The point isn’t discipline cosplay; it’s offloading the scheduling function onto the environment so the cognition can do the work. An LMS that lets a learner say “twenty-five minutes, then a break” is doing executive function as a feature.

Notification sovereignty. Attention that is fragile does not survive a platform that treats every interruption as engagement. Tooling designed for ADHD cognition protects focus by default: notifications off unless the learner turns them on, no badges for opening the app, no streak flames for the daily ritual. This is a direct inversion of the industry’s engagement metrics, and it’s the right inversion — the metric you’re optimizing shouldn’t be the platform’s time-on-site but the learner’s time-on-task, which is often maximal when time-on-site is minimal.

Asynchronous review. “Ask me to record the call and re-watch it” is not a request for a favor; it’s a description of how working memory functions under ADHD. Real-time comprehension and retrieval-from-recording are different cognitive operations, and replayability converts the second into a substitute for the first. A learner who can re-watch the lecture at 1.5x, pause it, and re-run the hard ten minutes will beat a learner who was “present” for the whole thing. Replayable beats real-time — for learning, almost without exception. The synchronous-first default persists not because it teaches better but because it’s cheaper to schedule and easier to measure attendance in.

Where current LMS design breaks both

Now the indictment, briefly, because the failure modes are the design requirements read backwards.

• Forced synchronous everything. Live lectures, mandatory attendance, timed discussion posts. Every one of these optimizes for the institution’s measurement convenience against the learner’s cognition. The content rarely requires the synchrony; the scheduling does.
• Social-pressure features. Leaderboards, public completion walls, nudges that shame you in front of peers. These convert learning into a social performance and then assess the performance. If your learner is maskingmasking. Hiding your natural way of thinking or behaving to fit in. Common among neurodivergent people; it's exhausting and linked to burnout. already — camouflagingcamouflaging. The work of hiding autistic traits and performing neurotypical ones — closely related to masking. A real skill with real costs; the most commonly reported cost is exhaustion. the cognition to survive the environment (see autistic burnout and masking in knowledge workers) — social-pressure features add the environment’s load on top of the content’s.
• Gamified streaks that punish bad days. The streak mechanic assumes tomorrow looks like today. It doesn’t — not for ADHD regulation, not for autistic burnout cycles, not for anyone with a body or a life. A streak is an incentive that converts one bad day into a lost record and a defection. It is a retention metric wearing a learning mechanic’s clothes.
• Assessment that conflates social performance with skill. Participation grades. Oral exams as default. “Soft skills” rubrics that score eye contact and deference under a learning-outcome label. The credential then certifies the performance, not the skill — which is precisely the validity failure that makes credentials untrustworthy downstream. If the assessment is measuring social fluency while claiming to measure mastery, it’s telling the verifier a lie, and it’s telling the neurodivergent learner that their skill doesn’t count until they perform it in the approved dialect. This is the assessment problem in miniature — and it is the same argument as the credentialing one, seen from the learner’s side instead of the verifier’s.

The design bet

One of the tools this essay argues for already exists here: the Visual Schedule builder — predictable sequences, tap-to-complete, printable. Free, no account, built for the learners this essay describes.

Here’s where this stops being a complaint and becomes an architecture. The bet for what we’re building at Mneurix is three moves, and they generalize to any learning platform.

Build the system of record first; let the intelligence layer honor it. The durable artefact of learning shouldn’t be the attendance log or the streak counter — it should be the work: the artefact produced, the rubric applied, the reasoning behind the grade, sealed in a credential a third party can verify without trusting anyone’s word. That is the credentialing argument made from the learner’s side, and evidence is what converts a claim into an inspectable claim. Record the work first; derive the analytics from the record, not the other way around.

Default to written, asynchronous, replayable, structured. Not as accommodations — as the default. Written spec over oral brief. Recorded lecture over live mandatory session. Async submission over timed in-class performance. Explicit rubric over “we’ll know it when we see it.” The neurodivergent learner gets the medium they need without having to disclose, request, and wait; the neurotypical learner gets a better product they didn’t know to ask for; the institution gets a credential that actually evidences skill, because the assessment measured the work instead of the social performance around it. The synchronous, oral, one-shot default can remain as an option for learners who genuinely prefer it — that is universal design, not a replaced default.

Make accommodations cheap to grant, not special to request. The accommodation-by-request model has a stealth cost baked in: it requires disclosure, and disclosure carries risk. Every learner who needs written instruction but doesn’t want to file a diagnosis goes without — and the employer or provider quietly inherits a disclosure problem they caused. When the accommodation is the default, the request disappears, the disclosure risk disappears, and the administrative labor on both sides goes to zero. Design makes the accommodation infrastructure out of nothing but a changed default. That is the cheapest accessibility program ever invented, and it’s the one most platforms refuse to run.

The strongest objection

Two objections deserve taking seriously, and the second is a real cost rather than a misunderstanding.

The first is that edges disagree with each other. “Design from the edge” is a slogan, and edges have conflicting needs: one learner is helped by structure and predictability, another by latitude; one needs reduced stimulation, another needs the engagement that stimulation buys. A platform honouring every edge at once can produce a configuration space nobody can navigate, and a designer told the edge is always right has no principle for choosing when two edges want opposite things. The honest answer is that the edge is a specification for defaults, not a mandate for every case — and the arbitration between defaults has to be explicit rather than assumed.

The second objection is sharper, because it accepts the design argument and points at what it costs. Synchronous, social learning is not merely institutional convenience; it has strong evidence behind it. Freeman et al. (2014), pooling 225 studies, found that active learning substantially outperforms passive lecturing in STEM — including on failure rates — and much of what “active” means is peers working together, in one room, at one time. So an async-first default is not free. It can trade away something that demonstrably works, and “async is better” is a claim about a specific learner’s access problem rather than a general empirical finding about learning.

That reshapes the argument rather than sinking it. The right response is not to replace the synchronous default with an asynchronous one, but to make the asynchronous path a first-class path rather than a fallback, and to keep synchrony as a genuine option with equal standing. The failure mode to avoid is the mirror of the one this essay opened with: an async-first system in which the synchronous, social learner is now the one filing requests. The principle survives contact with the evidence only if it is applied symmetrically — the goal is that no learner’s access depends on asking.

And the first objection is the one this site has to answer hardest, since it builds credentialing products and a credential’s value depends on the assessment meaning the same thing to every reader: an assessment that flexes per learner is a nicer experience and a weaker signal.

Why the system, not the exceptions

The bridge, briefly. Any credential or assessment can only certify what the learning system upstream of it actually measured. A platform that measures presence, performance, and social fluency produces records that carry exactly those signals — and no amount of careful certification fixes the contents. Changing the defaults of the system of record is a one-time cost for a permanent shift in what everything downstream of it means, and it costs the learner nothing they weren’t already paying in masking.

Design from the edge, record the work, default the accommodations, verify the skill. The neurodivergent learner isn’t the burden the learning system accommodates. They’re the specification the learning system should have been built to.