How we work
This desk does not report from the scene. It reads what outlets around the world file, groups the reports that describe one event, and writes a plain explanation with every claim marked according to how well it stands up. What follows is the whole of it, including the parts that are limits rather than features.
Six stages, and nothing skips one
- Ingest. We read published feeds from outlets in several languages, at a polite rate, honouring robots.txt. We never scrape a page that asks us not to.
- Cluster. Reports describing the same event are grouped, so nine outlets filing on one vote become one story carrying nine reports rather than nine entries.
- Extract. The specific factual assertions are pulled out of the reporting as separate claims.
- Explain. A plain-English account is written from the reports. It is original writing, not a copy of anyone's article.
- Verify. Each claim is checked against every other outlet in the cluster, and its status comes from what those outlets actually said.
- Publish. Sensitive subjects and unresolved claims wait for a person. Nothing about war, death, crime, elections, health, courts or children publishes itself.
What we do not do
We do not republish articles. We summarise, cite and link, and the outlet keeps its reader. We do not have reporters in the field, so we cannot confirm anything nobody has reported. And we do not present a single outlet's account as established fact: one report is one report, and the page says so.
Where this gets it wrong
Clustering can split one event into two stories when outlets describe it in very different words. A claim can be marked confirmed because several outlets carried the same wire copy, which is agreement rather than independent corroboration. And an explanation is only as good as the reports behind it: if every outlet repeats an error, so will we, until one of them corrects it and we follow.
When we get something wrong, the record stays up. Every change is listed on the corrections page, and each story keeps its full revision history.
Questions people actually ask
Do you write your own reporting?
No. This desk has no reporters and does not send anyone anywhere. It reads what other outlets publish, groups the reports describing the same event, and writes one plain account of it with every claim traced back to who filed it.
What does it mean when a claim is marked confirmed?
That the assertion is carried by more than one outlet reporting it independently. It is not a guarantee of truth, and it can be withdrawn: a claim that was confirmed is marked disputed or false if the reporting behind it changes.
Several outlets said the same thing. Why is that claim still unverified?
Because agreement is not corroboration. Outlets frequently run the same wire copy, so twenty versions of one agency report is one source, not twenty. A claim only moves once outlets have reported it separately.
Is any of this written by AI?
Yes, and the boundary is fixed. A model groups reports, drafts the summary and explanation, and proposes a status for each claim. It never decides what publishes. War, death, crime, elections, health, courts and anything involving minors cannot leave the queue without a person releasing it.
What happens when a story turns out to be wrong?
The record stays up. The story is corrected rather than quietly edited, the change is listed on the corrections page, and each story keeps its full revision history so you can see what it said before.