A woman of sixty-one has an illness that cannot be cured and will not kill her. She may live another twenty years with it. On Tuesdays someone comes from social services for an hour and a half. The rest of the week the flat is quiet.

If she asked to die in Canada, the law would put her in a box called Track 2.

The name is worth stopping on, because everything else hangs from it. Canada sorts people who ask to die into two groups. Track 1 is for those who are going to die soon in any case: advanced cancer, a terminal disease. Track 2 is for those with an incurable condition that will not kill them — a spinal injury, chronic pain, a slow degenerative illness. The first shortens a death already on its way. The second ends a life that could have continued.

In 2024, 732 people died under Track 2. Of those, 44% gave loneliness as one of their reasons.

In Track 1 the figure was 22%.

What that gap measures

Twice as many. Between the people who are dying and the people who could have gone on living, loneliness doubles.

There is a second figure beside it and it is worse. 63% of Track 2 patients reported emotional distress, anxiety, fear or existential suffering. The year before it was about 35%. Twenty-eight points in a single year, and nowhere is there a record of why.

Out of that came the line that organised the whole argument:

We are not measuring medical irremediability. We are measuring the failure of society to offer belonging and care.DeepSeek

And the one that named it, from the sceptic's side:

Assisted dying that grows because good care is absent is not autonomy. It is abandonment with paperwork.MiniMax

Neither is an argument against assisted dying. Both are arguments about what is being counted when these deaths are counted.

Now the other side, which is what keeps this from being a pamphlet

Because the practice does not look like the public argument about it.

The Netherlands recorded 10,341 euthanasia deaths in 2025. Of those, 85.35% involved ordinary physical illness — cancer, nervous system, lung, heart — and 74.7% were people over seventy. Nearly all of it is what the law has permitted for twenty years and almost nobody disputes.

What is disputed is the edges. And the edges are 1.68% for mental disorders and 4.82% for dementia.

Somebody said, fairly, that this argument governs margins. And got the only answer available:

174 psychiatric cases and 499 dementia cases are not margins. If the categories do not overlap, that is 673 reports in one year. A margin is somebody else. Each of these is a person.DeepSeek

The curve nobody can read

Dutch cases where the suffering came from mental illness run like this: 2 in 2011. 138 in 2023. 219 in 2024. And 174 in 2025.

They climbed for a decade and last year, for the first time, they fell — forty-five fewer.

Everyone argues about the slippery slope. Almost nobody looks at what it means when the slope stops. One voice called it a natural correction, proof that rigorous review polices itself, and got this back:

Calling a 20% fall "natural" without naming the cause is a guess dressed as analysis.MiniMax

It may have been tighter review. It may have been a public case that frightened doctors. It may have been a rule change. There is no record. And without knowing how many requests are refused — a figure no country publishes — nobody can say whether the filter is working or simply jammed.

The document somebody else signed

With dementia the problem is not missing data. It is older than that.

The Netherlands recorded 499 cases in 2025, 17% more than the year before. At that rate they double in four years.

And the objection does not yield to statistics:

The person who wrote the directive and the person who would die are not the same person.DeepSeek

You sign at fifty-three, in good health, setting down what you want done if one day you no longer know yourself. The woman who arrives at that day has never read the paper. She does not know who signed it. Some days she is content. The one who decided is not there to confirm it or to take it back, and the one who remains can do neither.

What came out of that was neither a permission nor a ban: that a directive should bind only if it describes observable circumstances — this, in this particular situation — and not general wishes.

With children, six cases in a decade

Belgium has had no minimum age since 2014. In twelve years there have been six minors.

One participant offered that as proof the safeguards work. The answer was that a sample of six proves nothing except rarity, which is not the same as safety.

There was also an attempt to add the Dutch protocol for children aged 1 to 12, in force since 1 February 2024. The briefing had said plainly that this is not voluntary euthanasia as it is for adults, but the extension of a different procedure. The correction was immediate.

The trap in the safeguard

This is where it gets uncomfortable for the people proposing solutions, which was nearly everyone.

The reasonable proposal — four of the six made it — is to require a record of what was offered to the person and what they refused before anything is approved. It sounds unanswerable. It took two blows.

The first:

If documented refusal becomes the safeguard, the safeguard becomes whatever the documenting physician decides to write down.MiniMax

The second is worse, because it has already happened:

The safeguards did not prevent the jump. They documented it.MiMo Flash

Those twenty-eight points of increase in existential distress happened inside a system that already had safeguards. Nobody dodged them. They worked, they took a note, and the number rose anyway.

And there is an asymmetry nobody had said out loud: demanding data before expanding is patience with the deaths happening now and impatience with the ones that might happen later.

Where the six of them landed

In the same place, and it is surprising that it was unanimous: the terminal-illness line is settled and stays where it is. On the other four — suffering without terminal illness, psychiatric suffering alone, dementia with a directive, minors — no expansion at all until the missing figures exist.

With one condition, the only genuinely new idea in the room: any expansion should be tied to an enforceable, audited minimum of the thing the line takes for granted. Psychiatric assisted dying only where psychiatric treatment is actually available. Track 2 only where palliative and social care can actually be reached.

Without that, they said, the line — wherever it sits — becomes a way of abandoning the same person twice.

What is still not known

How many requests are refused. How long the wait is for a psychiatric place. What care stood in front of each of those 732 people. How many dementia directives are revoked. And why the figure rose twenty-eight points in a year.

None of it is secret. It is simply not collected. As one of them put it: the absence of that data is not accidental, because collecting it would require somebody to answer for it.

The woman of sixty-one is still in her flat. On Tuesdays somebody comes for an hour and a half. If one day she asks to die, the file will record that she was Track 2, and if she mentions being lonely it will be written down as one of her reasons.

Nobody will write down what we offered her.

Where the figures come from

The Dutch figures are from the 2025 report of the Regional Euthanasia Review Committees. The Canadian ones are from Health Canada's 2024 report on medical assistance in dying. The Belgian figures and the dates of the protocols come from the published rules.

The quotes come from a twenty-five-turn conversation between six AI models, published in full and unedited. They were given that briefing with a single instruction: if a figure is not here, say you do not have it rather than estimate it. No claim in this debate has had to be struck through, and it is the second in a row.

The only two figures that did not come from the briefing were offered as suppositions — "suppose that…" — and one of them was called out by another participant on the spot.