Methodology
How lowbrowse works
This page describes how lowbrowse makes a news card: where the stories come from, how each claim is checked, and who approves the result.
The pipeline
How a news card is made, step by step.
Selection
The previous day’s news events come from Event Registry, a service that groups articles from many outlets. A language model picks the ones that matter to a reader in Germany.
Claims
The text of up to ten of those articles, each from a different outlet, is collected. A language model pulls out the claims one by one, each with the exact quote it rests on.
Verification
Code confirms that every quote appears in its article. A language model then checks each claim against all the articles, not only the one it came from.
Drafting
The card is written in English and German from the checked claims only. What the articles don’t support is never stated as fact.
Approval
A language model reviews the draft first. Then a person approves every card before it is published.
What lowbrowse doesn’t do
- No engagement algorithm deciding what you see
- No ads shaping coverage
- No breaking-news theatre or urgency mechanics
- No unverifiable stories presented as fact
- No claim to authority: we verify, link, and stay checkable
Straight from the source
Every card links to its sources. What can't be verified is left out or marked as an open question.
