How we verify

Four statuses

The examples below are invented and only show the form.

Confirmed. The claim is supported by a primary official source, or by at least two independent outlets, and there is no credible refutation. It is written as fact.

The central bank raised its key rate to 7%. (Announced by the bank itself.)

Developing. There are reports, but not enough confirmation yet: for example, the event is less than an hour old, there is only one source, or a second comparison with the source did not agree with the first. Written as "according to preliminary reports", marked "developing", with the time of the update.

According to preliminary reports, about 2,000 people were evacuated after the flood. Developing, updated 14:20 UTC.

Unconfirmed. Someone's statement or assessment that cannot be checked as fact. Written only with who says it, and never as fact in a headline.

The minister said the reform would cut prices by a third.

Disputed. A credible source refutes the claim. The story shows both positions and the sources of each, without taking a side.

The army said it had taken the village; the other side's authorities deny it.

Independent sources

• Five reprints of one report are one source. If an outlet retells another one or publishes nearly the same text, WORA counts them as one origin.

• Outlets with a common owner are one origin too. For this WORA keeps an ownership map: who owns each outlet, who funds it and how it is tied to a state. Each entry links to a document that supports it.

  • An example from the map: TASS and RIA Novosti belong to the same state, so their reports are one origin, not two independent ones.
  • Agency copy (Reuters, AP, TASS and others) reprinted by many outlets also counts as one source.

Source tiers

• T1 — primary official: government bodies, companies about themselves, courts, regulators, scientific journals, documents. One T1 report is enough for "confirmed".
• T2 — established media and agencies. Two independent ones are needed.
• T3 — public accounts and posts. Only as a statement with the author named. They do not raise a status.
• T4 — unknown or anonymous. Never the only source and never raise a status.

Honest limits

We measure accuracy and publish what we find. The latest measurement is from 11 October 2026.

• Sample. 30 stories and 150 claims that the system had not seen before.
• Reference. The correct status of each claim was set in advance. The labelling was done by AI assistants, and part of it was checked by a person.
• Wrong "confirmed". In three independent runs the system marked a claim "confirmed" when it should not have for 4, 5 and 6 claims out of 150 (2.7–4.0%). Our target is at most 1%. It has not been reached yet.
• Caution. The system misses a lot that is true: it confirmed 10–13 of the 45 claims that could have been confirmed, and the rest stayed "developing". We chose to miss a confirmation rather than confirm too much.
• Main known error. Two outlets retell the same press release, and the system takes them for two independent sources.
• Grouping reports into stories (measured 7 October 2026). Reports of different events are almost never merged: precision 97%. But one event is often split into several stories: 43% of the links were found.

• Texts (assessed 11 October 2026, 17 stories in both languages, 374 text units).

  • Units without support in the claims: 0.
  • Errors in numbers, dates and names: 16 units in 5 stories. All 16 were in English texts written from Russian sources.
  • The English and Russian versions often differ in which points they list.

This is why an editor reads every story before it is published.

Errors happen. If you find one, write through the "Contact" page: which story, which claim, what is wrong and, if you can, a source with the correct information. A confirmed error is corrected openly, by the rules on the "Corrections" page.