Methodology: how we source and date

Primary sources first

Every number traces to the agency that published it: Federal Reserve releases, Treasury statements, CBO reports, BLS releases, Congress.gov, FRED. We link the release, not a summary of it.

We do not invent Fed rates, FRED values or quotations. A claim that needs a source and does not have one is left out. That is a harder rule than it sounds, and it is why some stories that circulate in coverage never appear here as figures.

Two visible dates on everything

Publish date and last-updated date appear on every article and glossary term. Standing explainers are revised in place; the update note says what changed.

Data stamped with an “as of”

Live figures carry the reference period and the retrieval date. A figure that has not been retrieved is not displayed — we show the slot as awaiting its first reading rather than filling it with an estimate. Where a series is revised by the issuing agency, we show the revised figure and note the revision.

A failed fetch never invents a value. It holds the previous one with its original stamps and says so. That rule is enforced in code, not by convention.

Projections labelled as projections

CBO baselines and Fed projections are forecasts under stated assumptions. We name the assumptions when we cite them. A baseline is current law, not a prediction that Congress will restrain itself.

Corrections are public

Errors of fact are corrected in the article with a dated note, and listed on a single corrections page.

Edited, not all hand-written

Every page is signed by the publication, with the editor one click away, and “edited” is the accurate word for what he does. Some of the text on this site is drafted or maintained with the help of algorithms and artificial-intelligence systems, and every figure is fetched and formatted by code. The editor is answerable for all of it. We say this plainly because a reader deciding how much to trust a page should know how it was made.

Two kinds of page are made two different ways, and the difference matters:

Explainers, glossary terms and data pages are reviewed before they publish. A person reads them and decides they are ready. They are revised in place rather than replaced.

Daily briefs and week-in-review pages publish automatically, after a machine check. They are researched with an AI system, which files each item as structured data: the headline, a summary, and every figure as its own field with the source it came from. Before anything appears, an automated pass fetches that source and looks for each figure in it. An item publishes only when every one of its figures was found there. An item whose figures could not be found — because a source blocks us, publishes only a PDF, or simply does not carry the number — is held back and not published at all, and the page says how many were held. Every item shows what was looked for and where it was found; you can open that record on any item.

The check confirms that a figure appears at the source it is credited to. It does not confirm that the source is right, and it is not a judgement of how an item is framed. Framing on this site is editorial: money and fiscal facts are treated as choices with named tradeoffs, not as weather. That part is reviewed after publication on the automated pages, and before publication on the explainers. Anything wrong is corrected in place with a dated note and listed on the corrections page.

Not everything here is a fact

A figure with an “as of” stamp is a fact as the issuing agency published it. The sentences around it — what a number means, why it moved, who pays, what usually follows — are interpretation, offered in good faith and open to being wrong. We keep the two visibly distinct, and nothing on this site is advice. The terms of use set out what that means.

Interpretation on the economic pages starts with mechanisms: the quantity of money, how a deficit is financed, which vote actually spends. It does not start with a search for temporary factors that would leave every institution blameless.