How 4N stories are made

1. Where the articles come from

We subscribe to the public RSS feeds of national and international news organisations across the political spectrum, plus a Google News feed that surfaces additional outlets. Roughly every ten minutes we fetch each feed and keep the fifteen most recent items published within the last 48 hours. Items are discarded if they duplicate a headline or URL we have already seen.

Each outlet is labelled with a political-lean rating seeded from the AllSides Media Bias Chart (Left, Center, Right, or Unknown where no rating exists). The label appears next to every source on a story page. It describes the outlet, not the individual article.

2. Filtering out opinion and promotion

Every new article is scored for two things: whether it reads as opinion or advocacy rather than reporting, and whether it carries genuine news value rather than being a press release, product announcement, or retrospective. Articles that score as strongly opinionated or as having little news value are set aside and never enter a story.

3. Grouping articles about the same event

Articles are matched to existing stories first by the words they share in their headlines, then, where that is inconclusive, by a language model asked a narrow question: do these headlines describe the same specific event? Articles that only share a broad theme (both about the economy, both about the Middle East) are not grouped. A story holds at most eight articles.

4. The two-source rule

No brief is published from a single outlet. A story must contain reporting from at least two independent news organisations before it can be summarised, and the rule is enforced again in the database query behind every public page, so a story that later loses a source disappears from listings automatically. The story page states how many independent sources confirm it.

5. Writing the brief

A language model is given the headlines and opening text of every article in the group and asked to produce a fixed structure: a descriptive headline, a one-sentence brief, three to eight key facts, what happened, why it matters, and what to watch next. Its instructions prohibit loaded language, speculation, and any date or figure not present in the sources. Where sources disagree, the brief says so. Every key fact must cite at least one of the numbered sources, and the response is rejected and regenerated if the citations do not line up.

The brief is then reviewed by a second model pass acting as a neutrality editor, looking for opinion stated as fact, political framing, loaded words, and unsupported conclusions. Briefs that fail are rewritten once or twice with the specific objections attached; if they still fail, they are not published. Each story carries a neutrality score from this review, and stories scoring below our threshold are excluded from every public listing.

6. Keeping the feed short

Before a new brief is published it is compared with the briefs of the last 24 hours. Follow-ups, reactions, and secondary consequences of an event we have already covered are folded into the existing story rather than published as new ones. The goal is a few dozen distinct stories a day, not a running commentary.

7. Updates and corrections

When new articles join a story, the brief is regenerated from the full set of sources and the page shows its updated time. We do not delete published stories; a story that falls below our standards after publication stays reachable at its address but is removed from listings and marked so that search engines do not index it.

Automated summaries can misread a source. If you spot an error, tell us through the contact page and include the story link. Corrections are applied by regenerating the brief from its sources.

8. What 4N is not

4N has no reporters and adds no facts of its own. It does not verify claims beyond checking that multiple outlets make them, and agreement between outlets is not proof. The original articles, linked from every story, remain the authoritative record. Read them.