Autism Was Never the Epidemic

Before I said any of this out loud, I watched what happened to people who did.

The hostility came first. Not from clinicians, not from employers — from autistic people online. A small group, but relentless: that what we were describing was anecdotal, that we were harming the community, that we should stop. I spent a long time not posting because of it. I learned how to say the thing in a way that might survive them, and I got better at it, and I want to be clear that I should not have had to.

When I finally posted about autistic burnout and late-in-life identification, I had fewer than eight hundred followers. I said what happened to me: that I went into a workplace able to work and came out unable to, and then spent two years being asked to prove I had been impaired all along.

I did not think I was the only one. I have spent more than three years facilitating peer support spaces for late-identified autistic adults, and I have heard some version of this story every single time. Not occasionally. Every time. What I did not have was anywhere to put it. A room full of people telling you the same thing for three years is not a dataset, and I knew that, and I said it out loud anyway.

They came for me anyway. Same dynamic I had just lived through at work — my ability to explain myself treated as evidence I didn’t need anything, my account of my own functioning treated as a thing to be litigated rather than heard. And this was the last place left. The clinical route had not worked. The administrative route had not worked. What remained was the informal square, people talking to each other about their own lives, and the silencing followed me into it. That is not a side note about internet manners. When the formal systems cannot register you and the informal one is policed, there is no venue in which the thing can be said at all.

And then the other thing happened.

Thousands of comments. Me too. This is my life. Nobody has ever described this. Women in their forties and fifties who had been in therapy since their twenties. People carrying a personality disorder in a file somewhere, treated seriously, for years. People who found out at sixty-eight. People commenting from homeless encampments. People using substances to survive and cope. People unable to sustain employment. People barely holding on. My following is near seven thousand now, and almost all of that is people arriving to say some version of the same sentence. The stories broke my heart into pieces.

Hold those two responses next to each other, because they are the same finding. Thousands of adults identified themselves, unprompted, to a stranger with no platform, because no formal mechanism exists that would have asked them. And the fact that they had to do it that way — in a comment section, because there was nowhere else — is exactly what was used to disregard them. Three years of hearing it in peer support was anecdote. Thousands of comments were anecdote. There is no quantity of this that becomes evidence, because the only instruments that would convert it into evidence were never built.

That is the shape of the whole problem, and it is why this essay is built the way it is. I went and got the papers. What follows is what the papers say, and where they run against me, I have said so.

The papers say there is no mechanism.

We spent decades arguing about something else entirely — an “autism epidemic,” as though the public health problem were that too many autistic people had suddenly begun to exist. Rising numbers, alarmed headlines, a hunt for the cause of the surge.

The real crisis points almost the opposite direction. Not autistic people appearing out of nowhere, but autistic people who were already here — moving through childhood, school, work, marriages, addiction treatment, psychiatric wards, and decades of ordinary life — without anyone around them, including the clinicians paid to notice, having an adequate framework for what they were looking at.

That is not a diagnosis problem. It is a public health crisis of missed autistic adulthood, and we have built almost no accounting for it.

Start with how unstable “late” is as a term. Russell and colleagues’ preregistered systematic review, published in Autism Research in 2025, surveyed 420 studies spanning 1989 to 2024 and every continent but Antarctica, and found no field-wide agreement on what “late” even means. Only about a third of those studies defined a threshold at all. Among the ones that did, the cutoff ranged from age 2 to age 55 — a median of 6.5 years, a mean of 11.5, and a distribution with two peaks, one clustering around age 3 and the other at 18, depending on where the study was done and how old its sample was.

Read those two numbers together and the incoherence is the finding. The gap between the median and the mean is a thin tail of studies about adults, dragging an average upward across a field whose center of gravity is still toddlers. The two modes are two different arguments about what “late” means — late for early intervention, and late for a timely diagnosis — running in parallel under one word.

That is not a rounding error. That is 420 papers invoking a phenomenon the field cannot yet bound. (The inclusion criterion was merely that a paper use the term; only 146 defined it.) Before we can ask how many people this affects, we have to admit the category is still being built in real time — which tells you how recently anyone began asking the question seriously.

Where the evidence stops being ambiguous is the cost.

Kentrou and colleagues, working from a Dutch longitudinal register of 1,211 primarily late-diagnosed autistic adults, found that nearly a quarter reported at least one prior psychiatric diagnosis they now perceive as a misdiagnosis. Personality disorders led the list, followed by anxiety disorders, mood disorders, and chronic fatigue and clinical burnout diagnoses. Women reported perceived misdiagnosis at roughly twice the rate of men — but the mechanism underneath that gap is worth sitting with. Among participants who had received any prior psychiatric diagnosis at all, men and women were equally likely to consider one of them wrong. The gap opens earlier than that: 54 percent of the women had been handed a psychiatric diagnosis before anyone said autism, against 31 percent of the men. And the label was disproportionately a personality disorder — 18 percent of women against 6 percent of men, a threefold difference that held after adjusting for age, income, education, and ethnicity.

There was no measurable difference in autistic traits between the men and the women in that sample.

The same structure appears again when you sort by age at diagnosis, and it does something more interesting than I first expected. Among people who had received any prior psychiatric label, the share who came to regard one as wrong peaks in the middle of the lifespan and falls off at both ends: 37 percent for those identified before eighteen, 69 percent for those identified between twenty-five and thirty-four, and 39.5 percent for those identified at fifty-five or later. The oldest group looks, statistically, almost exactly like the children.

The reason is not that they were served better. It is that only 38 percent of them had ever received a prior psychiatric diagnosis at all, against 59 percent of the thirty-somethings. You cannot be misdiagnosed by a system that never processed you. The people who came up before adult autism was a thing anyone said out loud were not spared the error — they were spared the encounter, and everything that would have come with it. That is the missing adulthood in its purest form: not a wrong name, but no name, and no file.

Both effects run through the same door. Being a woman predicts perceived misdiagnosis in the full sample and not among people who already have a psychiatric label. Age at diagnosis behaves identically. The disparity is not in who disputes their diagnosis. It is in who gets one.

The authors are careful about what this is. These are perceived misdiagnoses, self-reported, and the study cannot separate an outright clinical error from a condition that was real but arose from living undiagnosed. That distinction matters less than it first appears. Under either reading, the treatment addressed the surface while the thing generating it went unnamed for decades. It is also a Dutch volunteer register — around one percent of that country’s autistic population, no control group, highly educated, and recruited through a national autism association. What it establishes is that the diagnostic failure is not an American peculiarity. What happens afterward — the invoice, the absent services — is a separate argument, and the American one.

The system did not merely fail to see autistic women. It saw them clearly enough to name them something else, and then treated the name.

Run that forward across a lifespan and the pattern compounds. The minority stress model, extended to autistic populations by Botha and Frost, supplies the mechanism. In a sample of 111 autistic adults, everyday discrimination, the expectation of rejection, internalized stigma, and the deliberate concealment of autistic behaviour each predicted lower well-being and higher psychological distress — and did so above and beyond ordinary stressful life events, which the study controlled for. The design is cross-sectional and the sample is small; the authors say outright that questions of causality remain open. What it does is reframe the question. Poor mental health stops being a property of autism and becomes, at least in part, a measure of how the surrounding society behaves toward autistic people. The same paper notes that under one percent of autism research funding in the United States and the United Kingdom has gone to adults on the spectrum or to the social factors that might explain their mental health.

Raymaker and colleagues gave autistic burnout its first empirical definition, drawn from nineteen interviews and nineteen public accounts: a syndrome arising from chronic life stress and a mismatch between expectations and abilities without adequate support, characterized by pervasive exhaustion, loss of function, and reduced tolerance to stimulus, typically lasting three months or more. One participant described it as having all of your internal resources exhausted beyond measure and being left with no clean-up crew, which is where the paper gets its title and most of its circulation. It is a small exploratory study and the authors say so.

But the detail in it that belongs here is one nobody quotes. Many of their participants reported receiving their autism diagnosis because of burnout — because they had lost the compensation skills that had been hiding them. The identification did not arrive through a screening pathway or a clinician’s suspicion. It arrived when the masking collapsed and there was finally something visible to assess. That is a diagnostic system that functions only after the person it missed has broken down.

Masking is what precedes that, and the qualitative literature is consistent that it is costly, exhausting, and implicated in the cascade — though it is described by participants as a strategy with a price, not as an inescapable baseline.

None of that is what happens when someone is identified a little later than average. That is what happens when a person spends twenty, thirty, fifty years inside systems — medical, psychiatric, educational, occupational — that had no framework for the cognitive ecology they were actually living in, and treated the resulting distress as the primary problem rather than the downstream signal of a much older mismatch.

And who gets missed is not evenly distributed, which is what turns “we didn’t know” from private misfortune into a public health finding.

Autism was conceptualized and operationalized largely around boys — Kanner in 1943, Asperger in 1944 — while earlier work documenting girls, including Sukhareva’s clinical observations from the 1920s, went overlooked in the West for decades. That founding male spotlight did not simply miss girls and women once. It seeded a cycle: the template narrowed who got studied, which narrowed the criteria, which narrowed who got flagged for assessment, which narrowed the sample in the next study built on those criteria.

Masking compounds it, though not in the way it is usually invoked. Garcia and colleagues, surveying 253 autistic adults in Brazil — a community-recruited sample, level 1 support needs, about two-thirds holding a formal diagnosis — found that women did camouflage more than men, and that camouflaging did not predict whether they were identified early. (It weakly tracked age at diagnosis in women as a continuous measure, and not at all in men; it predicted early identification in neither.) What their women reported instead, at 85 percent against 64 percent of men, was a prior diagnosis of some other psychological condition before anyone said autism: borderline personality disorder, major depression, bipolar disorder. Autistic traits themselves barely differed across genders — the same null that turned up in the Dutch register, in a different hemisphere with a different instrument.

Camouflaging in that sample predicted higher anxiety, stress, and depression, and a diminished sense of belonging — in men as well as women, more strongly in women. But the authors are explicit about where the diagnostic failure sits, and it is not primarily in the masking. Their stated foremost obstacle is clinicians’ slowness to recognize autistic characteristics, shaped by stereotypes built on behaviours more common in boys, with instrument insensitivity as the second-order problem. They cite work in which roughly a fifth of autistic females met the autism threshold on the ADOS against half of autistic males.

That distinction matters, because the popular version quietly locates the problem in what autistic women did. The finding locates it in what clinicians could not see — and in tools built to confirm a picture these women were never in. An instrument is a fixable artifact. A clinician who cannot perceive autism in the woman sitting across from him is the missing adulthood happening in a single room, in real time, and it is the same failure Kentrou found: the gap opens at whether you get a label, which is a decision a person makes.

Undiagnosed co-occurring ADHD compounds it again, autistic and ADHD traits obscuring each other well into adulthood.

But sex is only one axis, and I want to be careful not to write as though it were the whole story, because the gap I keep describing has more than one shape and mine is not the most common one.

Two things determine whether a support need ever becomes visible to a system: whether you can articulate it, and whether you have the resources to arrange your life so that you never have to. Those two things come apart, and they come apart in both directions.

If you have money, flexibility, transportation, healthcare, a private place to recover, and control over your own work, you can build accommodations around yourself and never ask anyone for anything. You can leave the job. You can pay someone to do the thing your body cannot. You can construct an ecology in which the disability stops being visible — including to the people counting. You are missed, and you are also, largely, fine.

Run it the other way and you get something worse. If you can name exactly what you need, explain the mechanism, and read the policy better than the person administering it, but you do not have the money or the autonomy to provide any of it yourself, then your fluency becomes the evidence against you. She’s articulate, so she must be managing. She filled out the form correctly, so she cannot need help with forms. I know this one from the inside: I am not a person with resources, and I have spent a great deal of my life being treated as one because of how I speak. That is not a small misreading. It forecloses the support that would have made the capacity real, on the strength of a fluency that costs something to produce.

Race runs through both, because whether the same fluency reads as competence or as arrogance, whether the same distress reads as overwhelm or as instability, whether the same request reads as self-advocacy or as demanding, is not determined by the fluency. One person is accommodated. The other is disciplined.

Which brings me to something uncomfortable about the evidence I have just spent several thousand words assembling.

Raymaker’s burnout interviews were entirely non-Hispanic white. Papadopoulos’s sample was 90 percent white and recruited through an online text-only survey, which the authors note excluded nonspeaking participants and those using augmentative communication. Kentrou’s register was 95.7 percent Dutch, 46.6 percent highly educated, 92 percent without intellectual disability. Garcia restricted recruitment to level 1 support needs by design and excluded anyone without completed elementary education. Russell’s review found no studies at all on late diagnosis from South America, and found that 92 percent of the papers quantifying it involved no autistic person in their design or interpretation.

Every one of those authors flags this. None of them are hiding it. But read the limitations together and they say something the individual papers cannot: the literature documenting the missed population is drawn overwhelmingly from the part of it that became legible. It selects, specifically, for the capacity to be articulate in writing, in English or Dutch or Portuguese, in a survey found through an autism organisation. It captures people like me reasonably well. It captures almost nothing about people whose distress never became articulate enough to be recorded — the ones who ended up houseless, in addiction treatment, in psychiatric holds, in prisons, in one room with a world that has become very small.

Exclusion produces data too. The people who are not in the samples are telling us something about the samples. And any count built to fix this has to be built for them, not for the people who were already findable.

The genetics complicate this, and not in the direction my argument would prefer.

In 2025, Zhang and colleagues published the first large investigation of how genetics relates to age at autism diagnosis, drawing on Danish registry data and the American SPARK cohort. Testing six competing models, they found that a correlated two-factor structure fit best: one polygenic factor associated with earlier diagnosis, one with later, only modestly correlated with each other. The authors are careful to call these relative terms describing a gradient rather than two discrete kinds of autism, and a substantial share of the variance in one cohort was explained by neither factor.

Read the second factor carefully, because it does not say what someone making my case would want it to say. It is substantially genetically correlated with ADHD, depression, PTSD, and self-harm. The authors offer two readings. One is diagnostic overshadowing, where co-occurring mental health conditions delay recognition of the autism underneath. That is my argument. The other is diagnostic misclassification, where people with other conditions receive an autism diagnosis they do not warrant. That is the opposite of my argument. The paper cannot fully separate them.

It does tilt, though, and slightly my way. When they decomposed the genetic signal, later-diagnosed autism could not be reduced to the additive polygenic effects of earlier-diagnosed autism plus six other mental health conditions — schizophrenia, ADHD, anorexia, depression, bipolar disorder, and PTSD. Whatever the later factor is, it is not simply other conditions wearing an autism label. That is evidence against the pure misclassification reading, not proof against it.

Nor does the paper rescue the claim about sex. Both polygenic factors correlated more strongly with autism in males than in females. Its actual conclusion about sex is a caution: apparent sex differences may partly reflect differences associated with age at diagnosis, which means research on sex differences has to account for that confound rather than read it as signal.

So I am not going to lean on genetics for the sex argument. That case rests on the historical record — the male-derived template, the decades of overlooked observations of girls, the clinical-recognition findings, the misdiagnosis rates — and it stands on its own without a genetic warrant it does not have. And whatever distinguishes late-diagnosed people at the genomic level, it is not producing a population that is doing fine. A quarter misdiagnosis rate, chronic minority stress, and burnout described as exhaustion beyond measure are not the profile of a group who can be left where they are.

There is one thing in that paper that belongs here, though, and it is the detail I cannot stop thinking about. The later-diagnosed factor is anchored in cohorts whose median ages at diagnosis are 14.5 and 15.8 years. And the quantitative analysis of what predicts diagnostic timing — the study’s central question — excluded anyone over twenty-two. The authors explain why: older adults may have missed an earlier diagnosis because of shifts over their lifetimes in how autism was understood. The exclusion is methodologically correct. It is also completely revealing. “Later-diagnosed autism,” in the best available genetics, mostly means adolescence. The woman who is forty-seven cannot appear in the analysis at all, because her existence would confound the measurement.

The gap is not only in the public health data. It reaches into the genome studies too.

Then there is the infrastructure, which is where this stops being a diagnostic question and becomes a systems one. What happens when someone does find out, at 35, or 50, or 68?

They discover that nearly everything built in response to autism — early intervention programs, school accommodations, pediatric research funding, parent support networks, most of the clinical literature — was built around autistic children. Not because autism stops at eighteen. Because the population everyone pictured, and funded services around, was young.

A late-discovered adult does not get to walk into that. They get a diagnosis, sometimes, if they can find a clinician who assesses adults at all, and then very little on the other side of it. No equivalent of early intervention for the fortieth year of masking. No workplace accommodation pathway designed with them in mind. No mental health system trained to treat someone whose presenting problem is three decades of a framework nobody had.

You can see what that absence produces if you look at where these adults surface instead. Papadopoulos and colleagues surveyed 475 autistic adults recruited specifically because they self-medicate with substances — a sample built to study the practice, not to measure how common it is. Among them, 87 percent identified with or had been diagnosed with anxiety, 75 percent with depression, and 80 percent with ADHD, a rate so far above the roughly 28 percent expected in autistic populations that the authors caution their findings really describe AuDHD adults who self-medicate rather than autistic adults generally. The study measures no diagnostic timing and cannot speak to causation. What it shows is a population dense with unmet need, moving through substance use, that arrives in services already carrying three or four labels. Whether anyone in those services ever asks the underlying question is not something the study measures either. That is rather the point.

And this is where I have to report something that complicates what I am asking for. In the Botha and Frost data, being out about autism was associated with worse outcomes — lower psychological well-being, higher distress — the reverse of what the same measures show for sexual minorities. Holding a formal diagnosis was likewise associated with higher psychological distress. The authors read this as the cost of disclosure in an environment that punishes it: telling people opens you to the discrimination that being unreadable had partly deflected.

I want to be careful with that finding. It is one cross-sectional study, and disclosure is not diagnosis. But taken seriously it says something uncomfortable and useful: identification, delivered into a hostile environment with nothing attached, is not automatically a gift. It can be another exposure. Which is an argument for building the receiving end, not for leaving people unnamed.

Which raises the question of how many people we are talking about, and here the answer is stranger than a simple absence.

There is a number. The CDC’s Dietz and colleagues estimated in 2020 that 2.21 percent of American adults are autistic — about 5.4 million people. It gets cited as though it settles the question. Look at how it was produced. The authors say plainly in the abstract that national and state estimates of autistic adults do not exist because no surveillance system is funded to collect them. So they modeled it: state population and mortality data, combined with a national parent-report survey asking whether a doctor had ever told them their child, aged three to seventeen, had autism — projected forward across the lifespan with a mortality adjustment.

That is not a measurement of autistic adults. That is the rate of parent-reported childhood diagnosis, extrapolated on the assumption that what parents report about children now applies equally to fifty-year-olds and eighty-four-year-olds who came up under entirely different diagnostic criteria. (Eighty-four is where it stops, because that is where the general-population mortality data ran out.) It is a serious piece of work and it was the responsible thing to do given what was available. It is also an admission. The best national figure we have for this population is arithmetic performed on a different population.

How much rests on the arithmetic is easy to see: change the input age band from three-to-seventeen to six-to-seventeen and the national estimate moves from 2.21 to 2.38 percent. An eight percent swing, from one defensible choice about which children to count.

But the detail that should stop you is the sex breakdown. The model puts autism prevalence among adult men at 3.62 percent and among adult women at 0.86 percent — a ratio above four to one, inherited wholesale from parent-reported childhood diagnosis. The authors flag it, say the reason is unknown, and suggest it may partly reflect differential diagnosis by gender.

Sit with what that means. The only national figure for autistic adults encodes, as a fact about adults, the very gap this essay has spent several thousand words documenting. A woman missed at seven is missed again at forty-seven — in the count. And there is one more turn: the abstract offers the estimate as covering adults “diagnosed and undiagnosed,” while the limitations concede that the childhood input excludes children who were never diagnosed. The undiagnosed are in the sales pitch and out of the arithmetic.

The architecture underneath it is the same story. The autism prevalence number in every news story comes from a monitoring network that tracks children. That is the design. We built a surveillance system that looks at one end of the lifespan and then acted surprised each time an adult turned up unaccounted for.

That is not a young field slowly catching up. It is a choice about where money and attention go, and the proportion is documented: under one percent of autism research funding in the United States and the United Kingdom has gone to adults on the spectrum or to social explanations for their outcomes. It is easier, and far more fundable, to study a five-year-old than a fifty-year-old who has spent three decades being treated for the wrong thing. Children generate research grants, early-intervention contracts, school district line items. Missed adults generate cost, quietly, spread across psychiatric wards, disability appeals, divorce courts, and addiction treatment programs that never ask the underlying question. None of this requires anyone’s bad faith. Ordinary incentives explain it completely, which is precisely why it has held for forty years.

And then the bill.

A comprehensive adult autism assessment runs from a few hundred to well over two thousand dollars in the United States, frequently uncovered by insurance, frequently requiring a specialist with a months-long waitlist even for people who can pay. The person missed by pediatricians, missed by school counselors, misdiagnosed by psychiatrists, and burned out by workplaces that had no idea what they were looking at is the same person now expected to personally fund the diagnostic process that should have happened decades earlier as a matter of basic public health infrastructure.

We know how to do this differently, because we have done it. In August 2012 the CDC recommended one-time hepatitis C testing for every adult born between 1945 and 1965 — explicitly, in the language of the recommendation, without prior ascertainment of risk.

The reasoning is worth reading closely. Risk-based screening had been the standard since 1998, and it had failed. Of the millions living with the infection, somewhere between 45 and 85 percent did not know. Nearly half of everyone ever infected reported no recognized exposure at all, and the report is blunt that risk-based testing was not identifying most infected people even among the insured. The old system asked people to demonstrate that they warranted testing, and the ones it missed were precisely the ones who could not produce a reason. So the CDC stopped asking for the reason. The USPSTF followed in 2013 with a grade B recommendation, which carries coverage requirements.

Two features of that decision matter here. It was never testing alone — the same recommendation paired identification with intervention and referral to care, and devotes an entire section to linkage, to the plain fact that people who are found and then dropped do not benefit from having been found. And the ask came with a price attached. Testing the cohort was projected to cost $664 million over three years to identify roughly 400,000 infections. Somebody wrote that number down, argued it was worth it, and won.

The analogy is not perfect and I do not want to pretend it is. The CDC’s case rested on a standard set of public health screening criteria: a condition that affects a large population, can be detected before symptoms appear, has a test that is reliable and minimally invasive, and has a treatment that changes the outcome. Hepatitis C clears all four. Adult autism assessment clears the first two and fails the rest — it is hours of clinical interview with no biomarker, and, by the argument above, very little to refer people to afterward.

That last part is the real objection, and it is worth answering rather than dodging: why fund identification of a population when there is no service pathway waiting for them? The honest answer is that this is one ask, not two. Assessment without a receiving end is half a system, and building the receiving end without a count is how you get a program nobody can fund, because no legislature appropriates money for a population that has never been enumerated. Counting is not a prelude to services. It is the mechanism by which services become fiscally arguable at all. That is exactly what the birth-cohort model did: it produced a number, attached a pathway to it, and priced the pair. The coverage followed.

So: a real surveillance number for undiagnosed autistic adults, funded the way we fund any other population-level health tracking, and built from adults rather than projected from children. Publicly funded adult assessment pathways, paired with the workplace, clinical, and mental health infrastructure that makes a diagnosis mean something. And research money that follows the adults who are already here.

None of that requires a new diagnosis, a new category, or another round of the argument about what autism is. It requires noticing that we spent forty years asking why there were suddenly so many autistic children, and never once asked where the autistic adults had gone.

They had not gone anywhere. There was no epidemic. There was a population that was never small, never new, and never actually late to exist — only late to be seen, and then handed the invoice for the delay.

I am a doctoral student, not a peer reviewer, and this is an essay, not a study. I have not conducted research here. I have read the research and made an argument out of it.

I am also not writing about this population from outside it. I am autistic, ADHD, dyslexic, dyspraxic, and a gestalt language processor. I have been through five burnouts. I have been misdiagnosed and mismedicated. I have been an inpatient on psychiatric units. I have had substance use challenges. I masked for decades before I had any word for what I was doing. I have been close to not surviving this, twice. Nearly every finding I have cited above — the wrong label treated seriously for years, the collapse that finally made the thing visible, the self-medication, the exhaustion the literature describes as beyond measure — I have lived some version of.

That is not offered as evidence. One life proves nothing and I am not asking it to. It is offered as position: this is where I am standing when I read these papers, and you are entitled to know that.

What I can tell you about the argument is how it was built. Every empirical claim above is traceable to a source in the list below, and every one was checked against the primary paper rather than against another writer’s summary of it. Where a study is smaller, narrower, or more self-selected than its headline number suggests, I have said so in the text. Where the evidence runs against my argument — the misclassification reading of the genetics, the finding that disclosure is associated with worse outcomes, the possibility that perceived misdiagnoses reflect real conditions produced by living undiagnosed — I have put that in the body rather than a footnote, because an argument that only survives when you hide the counterevidence is not worth making.

I used AI to do it, and I am telling you that on purpose. I used it to check my claims against the source papers line by line, which is how I found the errors in my own earlier draft: a number I had attributed to the wrong survey, a genetics finding I had overstated, a trend I had described as rising when it actually peaks and falls. I used it to draft and revise prose. The argument, the sources, and every judgment about what the evidence will and will not carry are mine.

I use accommodations to do my work. That is what this is. And I want to be exact about why I am naming it, because it is the same point the essay has been making the whole way down: when the accommodation works, the work looks like it never needed one. You are reading a document produced by a person with significant support needs, using support. If it reads as fluent, that is the accommodation functioning — not evidence that it was unnecessary.

If something here is wrong, I want to know. Tell me which claim and which source, and I will correct it.

Botha, M., & Frost, D. M. (2020). Extending the minority stress model to understand mental health problems experienced by the autistic population. Society and Mental Health, 10(1), 20–34.

Centers for Disease Control and Prevention. (2012). Recommendations for the identification of chronic hepatitis C virus infection among persons born during 1945–1965. MMWR Recommendations and Reports, 61(RR-4).

Dietz, P. M., Rose, C. E., McArthur, D., & Maenner, M. (2020). National and state estimates of adults with autism spectrum disorder. Journal of Autism and Developmental Disorders, 50(12), 4258–4266.

Garcia, S. G., Simões-Pires, C. da S., Brum, J. A., & Cabral, J. C. (2026). Taking off the mask: Investigating autism diagnosis and camouflaging in adult women. Autism in Adulthood, 8(4), 677–686. doi:10.1089/aut.2024.0150

Kentrou, V., Livingston, L. A., Grove, R., Hoekstra, R. A., & Begeer, S. (2024). Perceived misdiagnosis of psychiatric conditions in autistic adults. eClinicalMedicine, 71, 102586.

Papadopoulos, C., Adkin, T., Munday, K., & Gray-Hammond, D. (2025). Predictors of depression and anxiety among self-medicating autistic adults. Neurodiversity, 3, 1–14. doi:10.1177/27546330251344453

Raymaker, D. M., Teo, A. R., Steckler, N. A., Lentz, B., Scharer, M., Delos Santos, A., Kapp, S. K., Hunter, M., Joyce, A., & Nicolaidis, C. (2020). “Having all of your internal resources exhausted beyond measure and being left with no clean-up crew”: Defining autistic burnout. Autism in Adulthood, 2(2), 132–143.

Russell, A. S., McFayden, T. C., McAllister, M., Liles, K., Bittner, S., Strang, J. F., & Harrop, C. (2025). Who, when, where, and why: A systematic review of “late diagnosis” in autism. Autism Research, 18(1), 20–34. doi:10.1002/aur.3278

U.S. Preventive Services Task Force. (2013). Hepatitis C: Screening. Grade B recommendation, one-time screening for adults born 1945–1965.

Zhang, X., Grove, J., Gu, Y., et al., & Warrier, V. (2025). Polygenic and developmental profiles of autism differ by age at diagnosis. Nature, 646(8087), 1146–1155. doi:10.1038/s41586-025-09542-6