Methodology
How we monitor and analyse
How coverage is gathered, summarised and classified, what "verified" means here — and the limits of what the observatory can claim.
Four steps
From a published article to this site
Collect
We read the public feeds of the outlets listed on the Sources page. We store the headline, the date and the link — never a copy of the article.
Summarise
A language model writes the summary, the key points and the extracted figures. It is a machine reading, and we label it as one.
Check
A person reviews the analysis against the original article. Nothing reaches this site until someone has approved it.
Attribute
Every item links to the outlet that published it. If you are going to quote a figure, follow the link and quote them, not us.
Are these summaries written by a machine?
Yes. The summary, the key points, the sentiment reading and the extracted figures are produced by a language model. A human reviewer approves or rejects each one before it appears here, but the words are the model’s.
Why can I not read the full article here?
Because it is not ours to republish. The outlet that wrote it owns it. We publish our own summary and analysis, and send you to them for the text.
What does a "verified" figure mean?
That the figure we extracted matches what the original article actually says. It does NOT mean we have independently confirmed the underlying claim — only that we quoted the outlet correctly.
What does the source credibility score measure?
How often that outlet’s figures matched their own cited sources when we checked them. It is a record of our own checks, not a rating of the outlet’s journalism.
How does the chat assistant work?
It searches only the articles published on this site, then writes an answer from what it found and cites the articles it used. It cannot see anything we have not published, and it will tell you when it found nothing.
Found something wrong?
Tell us which article and what is wrong with it. Corrections are the point.