Out of Ten — give a stranger a mark, then keep looking until the mark cannot hold them

Everyone resolves a stranger at one glance and carries the answer around as though it were the person. It is not a moral failing, it is a rendering budget: a face at four metres is all the data there is, so a number gets made out of it and the number gets treated as the thing.

What is underneath is not more of the same. It is different in kind at every level. A face becomes a mouth becomes a lower lip becomes twenty years of chewing it on one side becomes the sentences that got stopped there. A body becomes a foot becomes a second toe becomes several hundred silent decisions about sandals. None of that is available at four metres, and all of it is them.

The structure keeps going because people do. You are not shown a person and then the truth about the person; you are shown that there is always another level, and that the level you stopped at was chosen by your patience rather than by them.

The last thing on the page is the only instruction: you have done this to everyone you have ever met.

Every detail carries a charge between −1 and +1, and the mark you give at the start becomes one too. The readout is the mean of everything you know, placed on a one to ten scale, with an ordinary 95% interval — 1.96 standard deviations — around it. A mark out of ten is ruled out when the interval misses its bucket entirely.

Surface details were written mild and deep ones were not, because that is how it is in people: nobody has ever left over a lukewarm detail and nobody has ever stayed for one. The measured standard deviation rises with depth — about 0.36 at the surface, 0.52 a level down, 0.64 below that, and 0.81 at the involuntary ones.

Two things fall out of that, and neither was designed in. The centre converges on the middle of the scale and stays there: know enough about anybody and they come out average, because everybody does. And breadth is safe while depth is not — across three hundred generated people, a visitor can open all thirty-six things visible on a stranger’s surface and still hold a working number, every time, from any starting mark. Pulling those same threads to the bottom kills it every time: three of them is the usual price if you move around the person, and twelve is the worst case in three hundred. Both results are asserted in the test suite, so the page cannot quietly stop being true.

Two hundred and forty lines, written by hand, arranged as six territories with three regions each and three details in every region, and a thread under each detail that runs three steps further down before admitting there is no bottom.

One rule held the whole thing together: nothing in it is a compliment or an insult. Every line is an observation. Whether a callus is tender or repellent, whether a hand that goes up first when something falls off a table is love or reflex, is the reader’s work — which is the only thing this page has ever been about.

A person’s seed decides which two of the three details in each region are theirs, and fills in the small variable parts — which hand, which year, which door. Everything else is human-universal on purpose. The material is shared. What differs is where somebody looked.

No. Nobody on this page exists, no photograph is used and nothing is generated by a model at the moment you look at it. A stranger here is a seed number and a fixed body of writing, arranged into a structure. The reason it feels like somebody is that the details were written to be true of people in general, which is the entire claim being made.

Every line is written to be true of somebody, and most of them are true of a great many people. None of them is a measurement of the person in front of you, because there is no person in front of you. Read them as the kind of thing there is to know about anybody at that depth, which is what the page is actually asking you to consider.

Because you already did it, in the first second, before the question appeared. A page that pretended otherwise would be flattering you rather than showing you anything. The mark is asked for openly so that it can be entered as one observation among the rest and then left entirely alone. It is never argued with. It just gets outvoted.

Every observation carries a charge between −1 and +1. The mark you gave is one of them. The readout is the mean of everything you know, on a scale of one to ten, with an ordinary 95% interval around it, and a mark is ruled out only when the interval misses it entirely. That interval is the whole story: the middle barely moves, and the edges eat the scale.

The arithmetic can wobble back over the line if a run of new evidence happens to agree with itself. A readout that resurrected would be a lie about what looking at somebody does to you, so the collapse latches. You do not un-know things.

Almost never, and it says when. There is no account and no cookie for this, and nothing about your route leaves the browser unless you ask it to: pressing share puts the route in a link and hands the link to you, and buying the poster sends it to Stripe so the poster can be rebuilt when you come back. In that one case the route is held for a day against the checkout session and then gone. What travels is a seed and a list of digits describing a person who does not exist — never anything about you. Counters move for arrivals, marks given, going past the surface, reaching the bottom of a thread, shares, checkouts and payments. None of them is attached to anybody.

A poster: the stranger you met drawn at print size, everywhere you went in them, the mark you gave with what became of it, and the deepest single line you reached set underneath. It is A2 at 200dpi, it downloads the moment the payment clears, and it is drawn in your own browser by the same renderer that drew the page. Because it already exists, a button that only counted intent would be measuring politeness. The studio’s own playbook says the honest test in that position is money, so it takes money.

Yes, with some determination. The seed and the route are both in the URL, and the drawing happens in your browser, so the lock is soft and pretending otherwise would be silly. Making it hard would mean rendering on a server and keeping a record of what everybody looked at, which costs more honesty than four euros is worth.

You get the same stranger they had, and their route through that stranger is drawn as cool dashed rings you can see but have not been to. Where your route and theirs land on the same thing, it goes warm. The count at the end is how much of one person two people actually shared, and it is usually humbling.