They ask people who they are. We built a twenty-year, multibillion dollar machine to avoid asking. Only one of the two is accurate, and it just lost its American funding.

Denmark. Summer of 2000. Roskilde Festival.

Four people from a violence-prevention NGO set up a library where the books were human beings. Ronni Abergel, his brother Dany, and their colleagues Asma Mouna and Christoffer Erichsen, working out of a group called Stop the Violence, built it after a friend of theirs was attacked. It ran four days, eight hours of conversation a day, and over a thousand people came to read.

Twenty-six years later they have published books in more than 85 countries.

Here is the mechanic. Pay attention to the mechanic. The mechanic is the entire column.

You walk in. You take a catalog. The titles are people: ADHD. Refugee. Bipolar. Homeless. Transgender. Muslim. Veteran. Recovering.

You pick one.

A volunteer sits down across from you and answers anything you want to ask for the next half hour. Including the stuff you have been too polite, or too chicken, to ask a single human being in your entire adult life. Difficult questions are not merely tolerated there. They are the point.

And then there is the clause. Goffstown, New Hampshire, which has been running these since 2014, publishes the terms flat out: the books are on loan for twenty minutes, and they reserve the right to be returned at any time.

The book can close itself.

Now say all of that in our language.

A person volunteers. Declares their own attributes. In their own words. Consents to one specific use. Sets the duration. Can revoke mid-session, no reason given, no lawyer involved.

No intermediary. No inference. No modeling. No lookalike expansion. No decay curve. No match rate. No fee.

That is a fully consented, self-declared, zero-party identity graph with a hard real-time opt-out. It is being run for free. By volunteers. In public libraries.

We spent twenty years and a staggering amount of other people's money building the photographic negative of it.

A third-party segment is a rumor with a CPM.

Somebody watched somebody else's browsing exhaust. Decided what it meant. Gave the conclusion a name in Title Case. Sold it by the thousand to buyers who would never meet the subject either.

The subject was never asked. The subject was never told.

We built a card catalog for people who never agreed to be catalogued, then charged each other for the privilege of looking them up wrong.

It is not really a database. It is a séance. The browsing history rattles the table, the modeler goes under, and the room is informed in hushed tones that the spirit present is a 34-year-old in-market auto intender, HHI over 100k, dog owner.

The dog is usually the most accurate part of the reading.

And every serious study of third-party demographic accuracy has told us the same thing for a decade. When you check the guesses against the actual humans, a meaningful share of them are simply wrong. Wrong gender. Wrong age band. Wrong household. The polite industry word for that result is "directional." The accurate word is coin-flip. Nobody in the chain is incentivized to run the check, which is why the check almost never gets run.

Here is the part that should sting.

We are not a stupid industry. We saw the problem. We did the responsible thing, on paper.

IAB Tech Lab built the nutrition label. The Data Transparency Standard. A way for a buyer to see where a segment came from, how it was built, what it actually measures. Good spec. Honestly made.

Almost nobody installed it.

Because the label would have had to say something. And what it would have said, over and over, is that the premium audience you just bought was modeled off a panel small enough to fit around a conference table.

So: we built a machine to avoid asking people who they are. Then we built a second machine to disclose what the first machine was doing. Then we quietly declined to plug in the second machine.

Sunday reading. Let me tell you about Sean.

He is one of the books. New Zealand. Joined in 2022. He was 58 years old when he was diagnosed with ADHD.

Fifty-eight.

Do the arithmetic on that. Six decades of being described by people who were guessing. A real answer arrives at an age when most people have stopped asking questions about themselves. And what does he do with it? He volunteers to be asked about it. By strangers. On the record. Over and over. For nothing.

Our industry would have had him at nineteen.

Not with a diagnosis. We do not do diagnoses. We do adjacency. We would have had him as a flag three levels deep in a taxonomy, inferred off time-of-day activity and a couple of forum visits, priced at a two dollar data fee riding on top of the media, wrong about half the time, resold four ways.

Never shown to him. Never correctable by him. Never returnable.

Sean got thirty minutes and a stranger who genuinely wanted to know. We got a line item.

The ugly part of a nice column.

In March, city staff in Mountain View, California told a committee that the Human Library organization had ceased operating in the United States after a federally funded liaison grant was pulled. The city looked at running the event itself, decided it did not have the capacity to serve as liaison, and marked the project closed.

Last October, the Reading Garden in Copenhagen had its final opening day. Five years. Free. Built with the city. You could walk in off the street and borrow a human being.

Gone.

I have not confirmed the American shutdown with Copenhagen, and I am not going to pretend one municipal meeting summary is the whole picture. See the note below.

But read the ledger as it currently stands.

The organization operating the most accurate consented dataset about human beings on earth appears to have lost its American operation over a grant.

In that same window, this industry cleared more money through programmatic supply paths before lunch on a Tuesday than that grant was worth in its entire life. Nearly all of it spent describing the same people badly.

I have sat across from strangers professionally for a very long time. A meaningful percentage of them were lying to me.

What sticks with me here is not the kindness. It is the efficiency. Thirty minutes and a willingness to be asked will get you further into a person than any dataset ever assembled about them. It costs a chair.

Everything we sell is an elaborate workaround for the fact that we did not want to ask.

So, the Sunday ask. Not a donation link.

Find out if the depot near you still exists. Check whether your company has a budget line that could keep it open, because this is the rare one that returns something you can actually read.

And the next time a vendor tells you their segment knows who somebody is, ask them the only question that has ever mattered.

Did anybody ask?

The Human Library's entire competitive advantage is that it asked. That is the whole moat. That should humiliate every one of us.