What is the difference between a source and a confirmation?

A source tells you something. A confirmation is a second, independent path to the same fact that does not depend on the first. Two people repeating one rumour is one source.

Why is provenance harder than accuracy?

Accuracy asks whether a claim is true. Provenance asks where a file came from — who recorded it, when, and where — and a file can be entirely genuine and still be years old and somewhere else.

Does speed necessarily cost accuracy?

Not by itself. What costs accuracy is publishing before the check rather than fast; the archive below is largely a record of newsrooms learning to run both at once.

Is a correction an admission of failure?

It is the opposite: a visible, dated correction is the only externally checkable evidence that a publication verifies anything at all.

Verification in journalism: the checks before publication

Every newsroom describes verification as a value and practises it as a checklist. The checklist has three items that are genuinely distinct, and most published errors are a failure of exactly one of them while the other two were done properly.

Independent confirmation

The oldest of the three, and the one most often faked by accident. A confirmation has to reach the same fact by a path that does not run through the first source. Two accounts repeating a claim that both read in the same place is one source wearing two coats, and this is precisely the failure mode a platform timeline manufactures at scale: volume reads as corroboration when it is only propagation. The archive below covers the first newsrooms to learn that the hard way, on breaking stories where the timeline moved faster than any desk.

Provenance

Where a file came from: who recorded it, when, and where. Provenance is a separate question from accuracy and it is the harder one, because a photograph can be completely genuine and still be from another country and another year. The techniques that answer it — reverse search, shadow and weather cross-checks, metadata, finding the earliest upload rather than the loudest — became newsroom skills rather than specialist ones during the period this archive covers, and they are now the first thing a desk does with anything it did not record itself.

The correction record

The third check is the only one a reader can audit from outside. A publication that corrects visibly, with a date and a description of what was wrong, is making a claim about its process that can be tested; one that silently edits is not. That is why a corrections page is a better signal of editorial seriousness than any statement of values, and why quiet edits are more damaging than the original error.

What changed, and what did not

What changed is the material. When copy arrived from named correspondents, confirmation was about people and provenance was rarely in doubt. When it arrives as a file from an account, provenance becomes the first question and confirmation becomes a problem of distinguishing independence from repetition. Adding readers as contributors — the subject of citizen journalism — multiplied both problems and also supplied most of the tools that solved them.

What did not change is that verification is a cost paid before publication and never recoverable afterwards. Machine drafting has made the first draft nearly free, which moves the scarce resource decisively toward checking; that is the argument running through AI in journalism, and it is the same argument in newer clothes.

Precedent, from the archive

These pages covered the earlier rounds of the same problem.

Verification in journalism: the checks that decide what gets published