How we verify

Every story on TheTrueNews is written and checked by software. This page explains each step, what the verdicts mean and which model versions did the work. Nothing is published until every key claim is verified.

From source to story

A story passes through these steps in order. If a step fails, the story stops there and is never shown.

  1. CollectReports are collected from the public source list. Nothing is taken from unlisted sites or social media.
  2. GroupReports about the same event are grouped into one story.Root AI - model b1t cluster v1
  3. WriteA short report is written in neutral language: facts, attributed quotes, no adjectives of judgement, no speculation.Root AI - model b1t writer v1
  4. Extract claimsEvery checkable factual statement in the draft is listed. Claims the story depends on are marked as key claims.Root AI - model b1t claims v1
  5. Check each claimEach claim is searched for in independent sources. Each piece of evidence is stored with the source, a quote, the link and the time it was retrieved, and marked as supporting or contradicting.Root AI - model b1t factcheck v1
  6. Publish or holdA fixed rule, not a model: the story is published only if every key claim is verified. Other claims that are not verified are removed from the text first. Otherwise the story is held and re-checked.
  7. TranslateThe published text, claims and evidence quotes are translated into every enabled language. Fact-checking happens once, in English.Root AI - model b1t translate v1

What the verdicts mean

Each key claim in a story carries one of these marks. Each has its own shape, so it can be told apart without colour.

  • Verified

    At least two independent sources state the same fact, and no listed source contradicts it. Only verified key claims are published.

  • Partly verified

    One source confirms the claim, or sources agree on the substance but differ in a detail (a date, a rounding). Shown only for non-key claims.

  • Unverified

    No independent evidence was found. A story with an unverified key claim is held and not published.

  • Contradicted

    At least one listed source states the opposite. The story is held, or if already published, corrected or retracted with a public entry.

  • Not checked yet

    The claim is in the queue. Never shown on a published page.

Independent sources

Two sources are independent if they are different organisations and neither only cites the other. In practice:

  • Outlets with the same owner count as one source.
  • An article that reproduces agency copy counts as the agency.
  • A primary source (the organisation the news is about) counts, but on its own never verifies a key claim.
  • Evidence shows the owner next to the source name so you can judge for yourself.

Corrections and versions

Every change to a published story creates a new version. The story page lists its versions, and the public corrections log lists every correction and retraction across the site. Nothing is silently edited. Corrections

Model versions

Each AI feature has a name and a version. The version that produced a story, a claim list or a verdict is stored with it and shown on the page, so a change in behaviour can be traced to a change in the model.

  • Root AI - model b1t cluster v1

    Groups reports about the same event into one story.

  • Root AI - model b1t writer v1

    Writes the report from grouped source reports.

  • Root AI - model b1t claims v1

    Extracts atomic, checkable claims and marks key claims.

  • Root AI - model b1t factcheck v1

    Finds evidence for each claim and assigns a verdict.

  • Root AI - model b1t translate v1

    Translates published text into every enabled language.

What this does not do

Verification means that listed sources agree on a fact, not that the fact is beyond doubt. Sources can be wrong together. The system does not evaluate opinion, prediction or motive, and it does not cover events that no listed source reports.