How INDB scores stories

Everything on a story page is counted from the outlets covering it. This page says exactly how, so you can check our work.

How articles become stories

INDB reads the RSS feeds and YouTube channels of the outlets listed on the sources page. It keeps the headline, the short description the outlet publishes in its feed, the link and the time. It never copies the article itself.

Each headline and description is turned into a list of numbers that describes its meaning. A new article joins an existing story when it is close enough in meaning to that story and was published within 48 hours of it. Sharing the same named people, organisations or places counts in favour; very common names count for less. An article that is close to nothing starts a new story.

Two articles from the same outlet need a closer match to land in one story, so an outlet's daily round-up does not pull unrelated items together. Stories that turn out to be the same are merged, and old links redirect to the merged story.

Coverage by lean

Every outlet has a lean rating from -1 (left) to +1 (right), or no rating at all. On a story page each outlet is counted once, however many articles it published:

• -0.2 or lower counts as left
• +0.2 or higher counts as right
• anything in between counts as centre
• outlets nobody has rated are shown as not rated, and are left out of the percentages

Lean describes the outlet, not the article. A centre outlet can publish a one-sided piece, and the reverse.

Coverage gaps

A story has a coverage gap when at least 5 rated outlets cover it, one side makes up 10% or less of them, and the other side makes up at least half. A story missed by the left is one the left is mostly not covering.

Source quality

Reliability is a 0 to 100 rating of how often an outlet gets facts right and corrects its mistakes. The factuality bands are: 85 and up very high, 72 to 84 high, 60 to 71 mostly factual, and lower scores mixed, low or very low. A story's average reliability is the average over the rated outlets covering it.

Every rating shown on INDB has a source. The outlet's page lists who rated it and when. Ratings inherited from INDB's earlier editions are labelled INDB editorial. Where an older list only held placeholder values, the outlet is shown as not rated rather than given a number nobody checked.

Outlets rated below 45 count as low credibility. State media means a government funds the outlet or controls what it publishes.

Truth score

Every story gets a truth score from 0 to 100. It measures how well the story is corroborated, not whether each claim in it is true. INDB does not fact-check individual claims; fact-checks from independent fact-checkers that match a story are listed in the misinformation radar.

• 40% from how many independent outlets carry it. One outlet adds nothing, 2 outlets a quarter, 4 outlets half, 16 or more the full amount.
• 35% from the average reliability of those outlets. When none is rated, 55 is used.
• 15% from how many sides of the spectrum (left, centre, right) are covering it.
• 10% from the share of outlets that are not low credibility.
• Then points come off for signs of a push: up to 25 for amplification, 0.15 for every point of average headline sensationalism above 40, and up to 10 for the share of state media.

The bands are: 80 and up strongly corroborated, 60 to 79 corroborated, 40 to 59 limited corroboration, below 40 weakly supported. A new story covered by one outlet starts low and rises as other outlets confirm it.

Warning flags on articles

Each article can carry warning flags. They describe the outlet and the headline, not the facts of the story.

• Low reliability source: the outlet is rated below 40. Mixed reliability: rated 40 to 54.
• High-risk content (60 or more) and elevated risk (40 to 59): content risk is half from the outlet's reliability (60 when not rated), a quarter from how partisan the outlet is, and a quarter from how loaded or sensational the headline is.
• State-funded media: a government funds the outlet or controls what it publishes.
• Possible propaganda: the outlet is low credibility and part of an amplification group, or it is state media whose headline uses loaded or sensational language, or it is low credibility and the headline uses two or more loaded words.

A story's propaganda risk is the share of its outlets with at least one article flagged as possible propaganda.

Amplification

A story is flagged as amplified when at least 3 low-credibility outlets publish nearly the same text within 3 hours. Copies of wire agency stories (Reuters, AP, AFP and similar) are normal and are not counted.

Ranking

Top stories are ordered by how many outlets cover a story, how many new articles it got recently, and how many sides of the spectrum are covering it. Scores halve every 10 hours, so new stories can rise. INDB does not rank by clicks.

When the list is built, a story whose lead outlet, lean, owner, country or section repeats the stories just above it is moved down a little. That keeps one outlet or one side from filling the page.

Trending

A name is trending when more outlets mention it in the last 2 hours than its usual rate over the past 3 days would predict, by a wide margin, and at least 4 outlets mention it.

Summaries

Stories with 3 or more outlets get a short summary. It is written by a language model running on INDB's own server, from the headlines and descriptions only, and says what the outlets agree on, what is disputed, and how the coverage splits. When the model is not available, INDB picks sentences from the outlets' own descriptions instead. Both are labelled. Summaries can be wrong; the original articles are always one click away.

Limits

• INDB only sees outlets that publish feeds, and only the parts they put in them.
• Lean and reliability ratings are judgements. We show where each one came from so you can weigh it.
• Grouping is automatic and sometimes wrong. If a story mixes two events, tell us on the contact page.