Editorial standards
These standards describe what the desk accepts as evidence, how it checks a claim, and what it refuses to publish.
What counts as a source
A source is a named publication, a regulator, or an official project channel with a feed the desk can read. Outlets that only republish another outlet are treated as the same voice, not as extra confirmation.
A story needs three independent publisher websites, or two publishers plus a separate primary source such as a regulator notice or official project post. At least one publisher must be in the desk’s higher trust tiers. Website identity is a practical check, not proof that reports were produced independently.
How claims are checked
Software splits source text into individual claims, each with an entity, a field, a value and a date where available. A second model pass checks each claim against its source evidence. In this site’s fact sheets, verified means the claim passed those source checks; it does not mean an independent human investigation confirmed it.
Every number that appears in a story is compared with a stored claim or a stored data value before publication. If a number does not match, the story is rejected and the reasons are logged.
The writing must cite a confirmed claim or data value in each paragraph. A similarity check compares the text with the source articles and with our own recent stories to keep the wording original.
What the desk will not publish
Price predictions, trading advice, invented quotes, unnamed sourcing, or causal claims that the data does not support. Sponsored posts and press releases are not used as the basis of a story.
If the desk cannot verify an event, it records the event on the relevant profile timelines without creating a story page.
When we get something wrong
New information is attached to the same story page as a dated update. When a claim was wrong, the page carries a correction that states what changed and when. The corrections page lists how this works and the log stays public on the story itself.