Follow people, not platforms

Several outlets reporting one casting announcement is one event, not several headlines. We cluster the reporting and show you the occurrence, with every outlet that carried it listed underneath.

Every update carries the reports it was built from, by name and by link. If we say something is confirmed, you can see which outlets confirmed it and go read them. If it is one outlet reporting alone, we say that too, and mark it as reported rather than confirmed.

We separate how important something is from how sure we are it is true. Collapsing those into one score is how a rumour ends up on a push notification.

Official APIs, RSS and Atom feeds, and licensed providers — nothing else. We do not scrape pages, and we identify ourselves on every request. Publishers can read exactly what our crawler does, and ask to be removed, on the crawler page.

The prose on every update is written by AI from the reports listed underneath it — and a check that is not an AI reads every name, number and date back against those reports before it can be published. If one of them is not there, the whole summary is thrown away rather than edited into shape.

What the AI does not do is decide anything: who a report is about, which reports are one story, how important it is and how sure we are are all rules you can read. Exactly which parts are a model, and which are not.

Corrections and retractions stay at their original address rather than disappearing. If we emailed you a link to something we later withdrew, that link takes you to the withdrawal — not to a 404. A dead link reads as evasion; the correction should reach whoever got the claim.