A number in a story is a claim with an issuer, and you are the person in the newsroom who knows it. WorldbyFlow surfaces figures with each one's source and interpretation and carries a fact ledger grading provenance on every read, so the numbers you check are already labeled by how well they are grounded.
Written for Works numbers and datasets inside a newsroom.·Runs in the General domain
Data journalism is half analysis and half provenance, and the provenance half is what the newsroom relies on you for. The scans here carry it: the figures with sources and interpretation, the money behind a story with who gains and who pays, polling with method and sponsor named, a graded explanation when a headline statistic moves, and the tests that would prove a popular claim wrong.
The figures on a topic with each one's source and interpretation stated. The numbers, labeled by where they came from and what they count.
Who is spending, who gains and who pays, mapped from disclosures. The financial structure under a data story.
A poll read naming pollster, dates, sample, method and sponsor, spread shown rather than averaged. Polling as it was measured.
A headline statistic moved and the explanations are flying. Candidate causes graded against the record.
The tests that would prove a claim wrong, named before the analysis. The method section, written first.
The order matters. The first read gives you the structure; the next ones fill the parts that are hardest to source by hand.
Check the source on each figure against your own. If it names the same issuers and separates the same measurement kinds, it can start the next one.
The candidate explanations graded. The story is usually in which one the record does not support.
What would show the claim is wrong and how you would know. Pre-registration for a newsroom.
Export the source list to check each figure at the original, and cite it in the methodology note.
Every figure and quote links to where it came from. A claim the scan could not source is marked as such rather than dressed up.
Copy the figures and sources as Markdown into the methodology note, so readers can trace each number back.
Ask a follow-up question of any result, or run a Red Team pass that tries to break its own conclusions before someone else does. How the grades work →
The AI-attributed layoff figure is a measure of what companies choose to disclose, not an independently audited count of jobs automation replaced, because Challenger's reason field depends entirely on employer language with no external verification step.
The fastest way to judge the result is to pick a subject you know cold and read it against what you know. If a colleague sent you here with an invitation, the credits land on your account when you sign up.