How News Unframed works

Methodology version 0.5 (draft) · September 26, 2026

News Unframed is a short daily briefing for readers in the U.S. who want to know what happened without reading several competing accounts of the same event. Each story has one factual headline, a short summary, and links to the original reporting underneath.

We aim for accuracy and restraint, not perfection. Choosing stories, grouping reports and wording summaries all involve judgment, and we can't guarantee neutrality. What we can do is show our sources and correct our mistakes in public.

What's in the briefing

Two sections, each ranked from most to least important: up to 10 U.S. stories, then up to 10 world stories. A story goes under U.S. when it happens in the U.S. or the U.S. is the main actor abroad; everything else that happens abroad goes under World, which includes only major events. On a slow news day a section is shorter rather than padded. The briefing is published around 7 AM and 4 PM Eastern time. The afternoon edition updates the morning one rather than starting over. Every reader sees the same briefing. There's no personalization and no endless feed.

How stories are chosen

Each candidate event is weighed on five questions:

• How many people are affected, and how seriously?
• Does it change something concrete: a law, a market, a policy, a scientific finding?
• Is it a new development, rather than a repeat of what was already known?
• Does it matter to readers in the U.S.?
• Is it time-sensitive?

How often a story was shared and how many clicks it got are not part of the decision. Coverage by major wire services and by outlets across the political spectrum counts as a signal of importance, but a large number of articles alone can't put a story on the list. Each story shows a category label, but categories never set quotas: there's no requirement to include a story from any category or political perspective.

Where the reporting comes from

We draw on a fixed list of news organizations chosen to cover a range of perspectives, plus official records where they help establish a fact such as a vote count.

Perspective labels are provisional and will be checked against a current independent rating before launch. Some outlets are paywalled; when only an excerpt of an article was available to us, the story's source page says so.

What “sources” means

“12 sources” means 12 different news organizations whose reporting was used for that version of the story. It doesn't mean 12 independent confirmations: many outlets republish the same Associated Press or Reuters story, so a republished wire story is counted once, as the wire service. Articles we found but didn't use aren't counted.

Perspective labels

On each story's source page, articles are grouped as Left-leaning, Center or Right-leaning. Every outlet, including non-U.S. outlets such as the BBC, is placed on the same scale. The label describes the outlet overall, not the specific article, and says nothing about accuracy. An outlet we haven't rated is shown as Unrated, never as Center.

Some stories are covered mostly by one side of the spectrum. We show that as it is rather than hunting for balance that isn't there.

How summaries are written

The goal is simple: take one event, read what dozens of outlets reported, and state just the facts: what happened and why, before anyone's reaction to it. Undisputed facts are stated plainly. Claims, estimates and figures that outlets disagree on are attributed to whoever made them. We don't adopt any outlet's angle, descriptive language or implied motives, and we prefer primary documents such as official fact sheets, court orders and transcripts where they exist.

An AI system drafts each headline and summary from the collected articles, following written rules: state what happened, attribute claims to whoever made them, keep uncertainty visible, and leave out analysis, predictions, motives and loaded language. Code then checks that every number, quote and outlet name in the draft appears in the sources, and a second AI system fact-checks the draft against the source articles. A story that still fails after two rewrites is dropped. Editions are published automatically, twice a day, without a human editor.

Changelog

• 0.5 (September 27, 2026): separate U.S. and World sections, up to 10 stories each.
• 0.4 (September 27, 2026): U.S.-first: up to 15 stories, at most three foreign; slow days get a shorter list.
• 0.3 (September 26, 2026): editions publish automatically with no human review; story labels removed (uncertainty is stated in each summary).
• 0.2 (September 26, 2026): one ranked list of 15 stories; categories are labels, not sections.
• 0.1 (September 25, 2026): first draft.