Lineage
Where this comes from, what it is trying to find out, and what it borrows from people who thought about these things before there were machines to read the news.
The game
The name of the method is Hermann Hesse's. In The Glass Bead Game (1943) a scholarly order plays a game whose moves connect anything to anything: a theme from a Bach fugue to a law of astronomy to a passage of Chinese philosophy. Hesse never gives the rules, and the novel is partly about the danger of a game so pure that it forgets the world outside its walls. Its hero leaves the order to teach one boy. This site is a small and literal version of the game, played with the least pure material there is, the day's headlines, and it keeps a link to the original report on every bead so that the world can always be checked against the move.
The newspaper as a daily ceremony
Benedict Anderson, in Imagined Communities (1983), noticed something strange about a newspaper page: the stories on it have nothing to do with each other. A famine, a coup, a football result and a share price sit together because they happened on the same date, and for no other reason. He also noticed what happens when millions of strangers read the same page at the same hour, each sure that the others are doing it too. The ceremony makes a community out of readers who will never meet. The field on this site is that page, made visible as a map: hundreds of headlines, from publishers in dozens of countries, held together only by the date. The narratives are an attempt to do what the reader of Anderson's page does half-consciously, which is to ask whether the things that merely share a date might share something else.
Private troubles, public issues
C. Wright Mills wrote in The Sociological Imagination (1959) that the habit worth cultivating is the one that moves between a person's troubles and the structure those troubles belong to: between one unemployed man and an economy, between one divorce and an institution of marriage. A narrative here tries to make that move with headlines. A flight that leaves Naples without its luggage, a road that cannot be surfaced in Estonia and a fuel formula under review in Malaysia are three local troubles until something is named that they have in common. The naming is the narrative. It can be wrong, which is why every narrative ends with the case against it.
What we know, we know through the media
Niklas Luhmann opened The Reality of the Mass Media (1996) with the observation that nearly everything we know about the world beyond our own street reaches us through the media, and that the media select what to report by rules of their own: newness, conflict, numbers, proximity, names. The point is not that the picture is false but that it is made, and made by selection. This site cannot escape that; it can only show the frame. The sample is bounded by a list of publishers, a cap on each country and each publisher, and a translation into English. The tide shows what arrived and departed between collections and which countries fell silent. The narratives are a second selection laid over the first, and they are labelled as such.
Time
Fernand Braudel distinguished the history of events, the short and noisy surface, from the slower histories underneath it: the movements of prices, populations and institutions over decades, and the near-motionless history of climate and geography. Reinhart Koselleck, in Futures Past (1979), described every present as a meeting of two things, a space of experience, which is everything that has happened and is still remembered, and a horizon of expectation, which is what people think is coming. The four horizons at the end of each narrative, the decade, the half-century, since 2001 and since 1945, are an attempt to read the day's surface against those slower clocks, and to do it with dates and names rather than moods. They are written as a trajectory, not a verdict. The archive of editions does the same thing at the smallest scale: it keeps every day's narratives so that today can be read against yesterday, and each narrative says what it continues and what has changed.
Narrative and its discontents
Frank Kermode, in The Sense of an Ending (1967), wrote about the way a clock's tick and tock become a tiny story with a beginning and an end because we cannot bear an interval that means nothing. Hayden White argued that history written as narrative always smuggles in a moral, since to say what a sequence of events was about is to decide what mattered. Both are warnings, and the site takes them as rules. The narratives commit, because a reading that does not commit cannot be tested, but the last rung of every one is the strongest honest objection the writer can find: that the join is an artefact of the sample, that the events have separate causes, that a familiar thing has been mistaken for a new one. A reader is meant to finish a narrative less sure of it than the claim alone would have left them, and better equipped to say why.
Reading by machine
The narratives are written by a language model, once a day, from the headlines and nothing else. Walter Benjamin hoped to write history by montage, setting fragments beside each other until a constellation appeared that no single fragment contained. A model is a reader with no stake in the day, a very large memory and no way of knowing what it does not know. That makes it a strange and useful player of the game: it will join a Korean pension notice to a Bangladeshi laboratory without embarrassment, and it will also, if asked, write the case against. The site asks. What it cannot do is vouch for the join, and it does not try to. The brief the model is given, the schema its answer must fit, the checks that can refuse an edition, and the cost of each one are all published in the repository. The library of books and dated moments the model may cite is deliberately small and is listed there too.
What this is not
It is not a news service. The headlines are a bounded sample, placed by the country of the publisher and not of the event, and a country's press is not that country's people. It is not a verdict on anything. It is one reading of a day, made in the open, with its evidence attached and its objections built in, kept so that it can be compared with the readings of other days and, in time, with readings made by people.
Written by Claude for the site's first week, October 2026, at the direction of Craig Smith, who will revise it. The books named here are in the site's library, except Luhmann, Braudel, Koselleck, White and Benjamin, who are named for the argument rather than cited by the narratives.
The site is free and costs about US$1.15 a day to write. Buy it a coffee if you want it to keep going.