The method
How we work
Our goal is to give the news traditional media don't. We follow every story across all the outlets to surface what's left off the agenda: the independent, the real, what isn't paid media. Here's how.
Two separate engines
We publish our own bylined articles, like any outlet. And we follow stories across what others publish. They're different things and they're signposted differently. We never mix them.
What a model does and doesn’t do
A language model helps us cluster articles on the same topic and propose which questions went unanswered. It doesn't decide what's true, doesn't rank frames by credibility, and doesn't write the articles. Everything it proposes is reviewed by a person before publishing.
Facts and frames, always apart
In each story we separate what the sources share from how each one frames it. On one side, only what’s verifiable against more than one source. On the other, the frames, with the warning that a frame is not a fact.
We don't republish others' content
From other outlets' articles we show the title, outlet, date and a link. Nothing more. The link always goes to the original and traffic returns to whoever did the work.
The source list is public
Anyone can see which outlets we follow and by what criteria. The same list you see feeds the system. We can't change it without it showing.
Errors are corrected on top, with a date
When we get it wrong, the correction goes on top of the article, dated. We don't delete what was there. Making it visible what changed is part of the deal.