Chapter 2

It takes her four days to get from that's odd to that's wrong, and the four days are, in every practical sense, the last ordinary days of her life.

Because here is the problem with the deceased pile: the code doesn't say anything. It says deceased, and that's a fact about a letter, not about a person. It can mean that the recipient died last year and the register hasn't caught up. It can mean the postal worker on the route knew the house and put a code on it. It can mean the letter went to a nursing home and came back with a stamp somebody chose from a drop-down at four in the afternoon on a Friday.

So the number nineteen means nothing at all until it can be put next to something.

That's what she thinks about on the bike home on Tuesday, and it's what she is still thinking about at eleven that night, and it's the closest thing to happiness she has felt in a month, which she is aware is a diagnosis of some kind.

So she builds herself a pipeline. Nobody asks her to. It's just that it's a mess, and messes can be sorted, and that is one of the few things in this world that can simply be done.

She builds it over four afternoons, in the way things get built when they're your own. The first version is a count. The second gets columns for week and batch, then a reconciliation step, then joins, then checks on the joins, because a join that nobody has checked is a rumour. In the third version she introduces colours — it's a Thursday, late in the day, the boxes have been collected, and it isn't a decision, it's laziness: instead of writing verified in a column she marks the row green, and unverified rows red, and from that moment the thing has what all good systems have, which is a conscience. You can see how much is missing. The red looks at you.

The colours are the only part of it she does for her own comfort rather than for correctness. Everything else in the pipeline she would defend to a stranger.

The first step towards the real version is getting hold of the merge file — the file that paired the population register extract with the biobank's own sample register when the letters went out. It sits with the print vendor, a firm in Glostrup with a name like a toothpaste, and the correspondence takes three emails and a week. The first email is answered with a friendly question about who she is. The second with a form she has to fill in, which asks about "purpose of data access", and where she writes quality assurance of returned-post processing, which is entirely true and looks entirely harmless, and both of those things are worth remembering. The third email contains a link that expires in 72 hours, and an access code sent in a separate email two minutes later, which is the kind of security that would make an auditor bite her pen.

And then it's there: 1,104,882 rows.

She pulls it into a working environment on the institute's machine, and the machine sulks the way institutional machines do, and she waits it out, and while she waits she reads the schema. It's the first thing she always does, and it's the only real skill she has that nobody has ever paid her for: not reading the data, reading the shape of it. What the fields are called. What they permit. What they cannot express.

Twenty-two columns. Ten minutes with the field definitions tells her more about the institution than four weeks in its basement has.

There is a status column with a small, closed set of values, and one of them is `04`.

There is no column for withdrawn. There is no column for objection, no restriction, no free-text field for anything a person might once have said about their own sample. There is a field called consent_v2, added recently, in a different naming convention from everything around it, the way a room gets an extension built onto it — and the older part of the schema has nowhere at all to put a no.

She notes it, on the black notebook's third page, without weight, as a thing about a system: no field for refusal. Where did they put them?

Then she runs the reconciliation.

Twelve thousand returned. Against the merge file, against the return codes, against the sample register's own status field — three sources that should agree, and mostly do. Category 1: 4,102. Category 2: 3,517. Category 3: 3,901. And category 4, the deceased, 340.

Three hundred and forty.

Out of twelve thousand returned letters that's a completely reasonable number. Out of 1.1 million recipients it's nothing at all. It is, statistically, exactly the sort of number a country produces without anyone doing anything wrong.

Except that in the sample register — the institute's own, the one the letters were generated from — all 340 of them carry the status `04`.

`04` is mors. Deceased. And it isn't a postal code applied by a courier at the door; it's a register value inside the biobank, set at some point by somebody or something, and it is the reason the letters were generated with a flag that the print vendor's system read and passed on, and the reason the post came back with the code it came back with.

Which means the letters didn't come back marked deceased because the recipients had died.

They came back marked deceased because the biobank had said so first.

Maja sits with that for a while. It is nineteen minutes past five, and the building above her has gone quiet, and the fluorescent tube is doing its thing.

Then she does the only honest thing there is to do with a hypothesis, which is to try to kill it.

She takes a sample of eight. Not the first eight — she takes them spread across the file, across birth years and postcodes, because a sample chosen for convenience is a sample chosen to agree with you. And for each of the eight she goes to the public register interface she has legitimate access to as part of the returned-post job, and looks up the one thing she is allowed to look up: whether the person is registered as living.

The first is living. The second is living. The third is living.

By the sixth she has stopped feeling the pleasant, cold interest of a puzzle and started feeling something else, in the chest and the arms, which she would not have been able to name.

All eight are living.

She checks the birth years, because eight living people out of 340 could still be a coincidence if the 340 were mostly ancient. They are not. The oldest of the 340 was born in 1979. The youngest in 1996. It's her own generation, give or take, and the reason is obvious the moment she thinks about it: this is a heel-prick biobank, and the letters went to people whose samples were taken at birth in the years the programme has existed.

Three hundred and forty people, none of them over fifty, all of them registered dead in a national biobank, and the eight she has checked are alive.

She writes the four numbers in the notebook, and under them the eight, and under those, because she is a person who writes things down so she can sleep:

Not a postal error. The register said it first.

It's past six, and the building above her has stopped making noise. She sends Dorte the email with the three tidy categories and leaves the fourth where it is until she knows what it is. Then she packs her bag, turns off the fluorescent tube — it flickers once more in the dark, like a full stop having second thoughts — and goes up the stairs and past Poul, who nods.

Dorte's reply is there next morning, sent at 7.42, and consists in its entirety of: Fine. Good work with the numbers. D. Four words and a letter, and Maja reads them three times and carries them around with her most of the morning, which she is well aware is sad, and which does not make it less good. Good work with the numbers. It is the first professional acknowledgement anyone has given her since Aalborg, and it comes from a woman with a calendar, for four columns in an email — and somewhere in the system the fifth column sits waiting, unannounced, like a guest in the hall.

That evening she sees Sofie, which has become their Wednesday, and halfway through the second beer she nearly tells her. She can feel the story sitting ready in her mouth — there are 340 people registered as dead, and they're alive, I've checked — and she can also hear exactly how it would sound across a café table: like an anecdote. Something about crazy systems. Sofie would laugh in the right place and say that's insane, and then the story would be used, converted into entertainment, finished. And there is something in Maja that doesn't want it used yet — not because it's a secret, but because it isn't done: it isn't a story, it's a table with an open end, and you don't tell people tables.

"What?" says Sofie, who can see there's something.

"Nothing. Work stuff. I found an error in a register."

"Look at you," says Sofie, raising her glass. "Four weeks in, and you've already found a fault in the Danish state. Cheers to that."

They drink to it. And the sheet stays down there in the dark under the institute with its 340 rows, and nobody in the world yet knows that it exists.