How the Digest picks its stories

From about fifty newsletters to one edition a morning, and what every number on it means.

Method · Source: Daily Newsletter Digest pipeline

What it reads

Every morning the pipeline reads the newsletters that arrived in the previous 24 hours: about fifty on a weekday, fewer at weekends. They are editorial newsletters with their own reporting or analysis, from general news, politics, technology, media, advertising and sports business.

Some mail is dropped before anything is read closely: breaking-news alerts that only repeat a headline, product-recommendation emails, automated news round-ups, re-shares of a newsletter that also arrived on its own, personal newsletters and promotional sweeps. Citizen of Nowhere’s own writing is never included; the digest is about other people’s work.

How stories are picked

The reading and selection are done by an automated editorial step, a language model working to a written brief, and the results are published as it leaves them. The newsletters are grouped into five to seven themed chapters, read as one morning audio episode, with a written post alongside. The edition on this site carries every article that post links, in the order it links them: the chapters first, then the further reading. A normal day is twenty to forty stories.

A story makes the post when it is substantive: original editorial work with specific numbers and named sources, not a one-line alert. When the same newsletter arrives twice, once on its own and once re-shared, the original is the one read. For a story that runs for days, the edition leads with what is new since yesterday.

Only direct article links are published. A link that only works through an email redirect is left out, because those can identify the subscriber, and a publication’s front page or section page is never counted as a story.

What each story carries

The headline and source are the publisher’s own, and the headline opens their page, which may ask you to subscribe. The why line is ours: one factual sentence drawn only from that day’s post. A story the post linked without discussing reads “Further reading from the day’s section on …” instead of an invented reason.

Place, club and league tags link to those pages on this site, and the story then appears under In the news on them. A club or league tag is only kept after its page has been checked live; an unknown place is dropped rather than guessed. Topics, themes and sports are computed from the source and the story, and each one is a filter over the last 60 days.

On the latest edition, a story about a place or club also gets a strip of our own numbers for it: power rank, population, the leader, the next election, the policy rate, a club’s league position. Archive days leave the strip off, because today’s numbers beside last month’s news would misread.

Editions and the archive

Each morning is one numbered edition. Edition No. 1 is the first digest on file, and the number counts the days that published, so a day with no digest does not use a number. The rail of the last seven days under the tabs shows each day’s story count; a day with no edition keeps its place in the week, greyed out.

Every story has its own page, with the why line, its tags, its interest score and the stories after it in the same edition. The day page is the home of the edition and the one search engines are pointed at. The archive lists every day, grouped by month.

Quiet days

A light mailbox makes a short edition, and the digest says so rather than padding it. An edition is called a quiet day when it has fewer than 20 stories and fewer than 70% of the median of the seven editions before it. Both, so that a slow week does not turn every day quiet, and a thirty-story day after a run of fifties is still an ordinary one. With fewer than three earlier editions to compare against, nothing is called quiet.

On a quiet day the episode goes deeper on fewer stories: more context, more cross-source comparison, never filler. A quiet day is marked on the day page and with a dashed border on the seven-day rail.

The interest score

Every story gets an interest score, three factors multiplied together. Each factor is 1.00 when it has nothing to say, so a missing input never sinks a story. The score ranks stories within an edition; it does not reorder the edition, whose order stays editorial.

  1. Newsletters. How many of the morning’s newsletters carried the story, as 1 + ln(count): one newsletter is 1.00, three is 2.10, ten is 3.30. Not recorded yet: the pipeline does not store a per-story count today, so this factor reads 1.00 for every story until it does.
  2. On this site. What the story touches here. A metro in the power rankings top 10 weighs 3.0, top 50 2.2, top 200 1.6, any other 1.2; a league 1.6, a country 1.5, a club 1.4. The strongest tag counts in full and each further tag adds a quarter of its weight, capped at 4.0. An untagged story is 1.00.
  3. Movement. Whether the story is running. Each of the seven editions before this one that carried the same metro, club, league, company, person or market adds 0.15, up to 2.05. Countries and broad themes such as “elections” do not count: they are in nearly every edition, so every politics story would be running. This stands in for movement in our own data (odds, Elo ratings, ranks), which is not yet joined to the stories’ tags.

The editor gate

Before a story is trusted, it goes through mechanical checks: the link opens an article rather than a front page; the headline is real, not just the publication’s name; the why line says something (at least thirty characters, unless it is marked further reading); the same article is not already in the edition; and it did not run in the previous seven editions. A story that fails any of them is held for the editor to look at.

The gate reports; it does not hide, reorder or rewrite anything on the public page, and its verdicts are seen only by the editor. A second step, a small judgment model answering “is this a substantive story, yes or no” with a confidence the gate can act on, is planned and is not switched on: it needs a labelled set of past stories to be measured against first.

What it does not do

The digest summarises other people’s reporting; it does not verify it, and a why line is only as good as the newsletter it came from. The interest score and the editor gate are mechanical: they rank and flag, and neither one decides what runs.

Where these stories come from

Each morning the Daily Newsletter Digest reads about fifty newsletters, groups them by theme and records an audio episode. These are the day’s most substantive stories from it, each with a direct link to the original article.

Links open the publisher’s own site, which may ask you to subscribe. The one-line summaries are ours; the reporting is theirs. A place tag links to that metro or country on this site, where the same story appears under In the news.