It does not fail randomly. It fails in the same direction everywhere.

Girls fare worse again: 42% of girls with disabilities complete primary school, against 51% of boys. The disadvantage compounds rather than averaging out.

These are not people who received the wrong support. They are people for whom the question of support was never opened.

Recorded prevalence follows national income — roughly 0.5% in low-income countries against 1.5% in high-income ones. Nobody believes autism is three times rarer in poorer countries. What differs is the machinery for noticing.

A diagnosis rate is therefore not a measure of how many autistic people there are. It is a measure of how many a country has the capacity to find.

Autistic girls are diagnosed later than boys — around 5.6 years against 4.8 on average — and the mechanism is now well described. Camouflaging, the effortful suppression of autistic traits to pass as typical, is measurably more common and more sustained among girls and women. Those who camouflage most are diagnosed latest, and the relationship is stronger for females than males.

So a child is read as shy, or quirky, or a bit intense, and the referral is never made. Some estimates put the proportion of autistic females still undiagnosed at eighteen as high as 80%.

The system rewards visible difficulty. Coping is invisible to it — and coping is exhausting, which is the part nobody scores.

Black children with comparable presentation have historically been diagnosed one and a half to two and a half years later than white children. Nearly half were assessed by multiple providers before anyone reached a conclusion; around one in seven saw six or more professionals first. Latino children average roughly eight medical visits before diagnosis.

Black, Hispanic and Indigenous children are also more likely to be identified only once an intellectual disability is present as well — meaning the milder presentation, the one where early support changes the most, is precisely the one that goes unseen.

The parents noticed. They raised it. What differed was how many times they had to raise it before somebody wrote it down.

Everything above describes systems that are trying. In the United States, 39 states and the District of Columbia report shortages of special education staff; most states have reported one in most years for two decades. The United Kingdom, Australia and much of Europe report the same structural gap.

There is no version of this that is solved by better referral pathways, and no realistic number of specialists that closes a gap this size. The constraint is arithmetic.

Each of these failures traces back to the same cause, and each one is a design requirement rather than a lament.

Direct, because half the people affected have no institution to be referred within. Free, because the families furthest from support are furthest from means. Native, not translated, because standing with institutions is culturally specific and deciding not to engage can be entirely rational. Learner-led, because a young person who masks is invisible to observation but perfectly able to describe what it costs them — if anyone asks in a form that does not itself require performance.

None of that is a claim to have solved this. It is the reason the constraints are the shape they are.