How we check an article
A PINTRU verification does not say whether an article is true. It says what the sources confirm, what they contradict and what they are silent on, claim by claim. Here is the whole process: what we choose to check, how we assess it and where we stop.
The rating
Every verification ends with a score from 0 to 100, simplified to a mark out of 10, and with a sentence that sums up the assessment.
Green is reserved for texts that pass almost every check: a 7 is a good result, but it does not mean everything was confirmed. Two cases stay off the scale: content that is plainly fictional, and content that cannot be verified.
Largely verified
from 85 to 100 pointsAlmost every claim is confirmed by independent sources.
Mostly verified
from 70 to 84 pointsMost claims hold up; a few are inaccurate or unconfirmed.
Partly verified
from 60 to 69 pointsPart of the text holds up; the rest doesn't or can't be confirmed.
Barely verified
from 40 to 59 pointsFew claims are confirmed, or there are no sources to show it.
Largely contradicted by sources
from 20 to 39 pointsSources contradict much of what the text claims.
Entirely contradicted by sources
from 0 to 19 pointsSources say the opposite of almost everything the text claims.
Fabricated content
off the scaleFacts, places or people that don't appear to exist, or a text that presents itself as fiction. It is not a low score: it is a different category.
Not checkable
off the scaleAn opinion piece, or a text in which most claims have no sources that could confirm or contradict them.
How sources are chosen
For each claim we look for sources independent of the text under review: news agencies, institutions, official data, other outlets, original documents. Every source comes with its publisher, date and address, so anyone can check it.
Each claim receives one of six outcomes. The words are chosen to describe what we found.
Confirmed by sources
The sources consulted confirm the claim.
Confirmed, with inaccuracies
Right in substance, but a verifiable detail (a figure, a date, a name) doesn't match. If the detail changes the meaning, it is no longer an inaccuracy.
Only partly confirmed
Sources confirm part of it; the rest doesn't hold or can't be found.
Not verifiable with the sources consulted
The sources don't say enough either way. It is not an accusation: it is a gap.
Accurate facts, misleading framing
The facts check out, but the text leaves out context that the sources report and that changes their meaning. The note always says which context is missing: if it can't be said in one sentence, it isn't misleading.
Contradicted by sources
The sources say the opposite.
Where the score comes from
The score is not an overall impression. It is the sum of four parts, each with a fixed weight that is the same for every article, and every verification shows the breakdown: you can see where each point comes from.
The outcome of the claims
This is the main component of the rating. Each claim counts according to its outcome: in full if confirmed, a little less if inaccurate, half if only partly confirmed, little or nothing if misleading or contradicted. A confirmation with no sources to show is worth less, because nobody can check it.
How many different sources
The more independent publishers support the verification, the more solid the result.
How the reasoning is built
Something is taken off for each recognisable flaw: a coincidence presented as a cause, partial data used as if it were complete, context left out, a headline the text does not support.
The tone of the language
Vague attributions («according to some experts»), emphatic tone, figures nobody stands behind. It weighs little on purpose: a text can be emphatic and true, or sober and false.
If a part cannot be measured, such as the language of a very short text, it is left out and the score is calculated on the others.
Two rules come before all the others. Style does not make up for facts: a text contradicted in every claim cannot score above 1, however well written and full of quotes it may be. No sources, no full verification: if there is no source to consult, the score cannot go above 5.
A verification follows the text over time
Every verification refers to the text as we read it. If the article is corrected, the verification is flagged for review; if it is rewritten, it no longer applies.
We also check that the text really is the one that was published. The text of a paywalled article, moreover, stays private: the verification is available only to the user who requested it.
A verification can also grow. When someone requests it again, the search picks up from the claims still open and looks for new sources; the outcomes are reviewed and the score is recalculated. On a story that is still unfolding, it is normal for the assessment to change.
What a user can do
Request a verification
On any current-affairs article. The outcome is written in the language of the article and has its own public page, with a stable address to cite or share.
Widen the search
If a verification already exists, requesting it again doesn't duplicate it: it extends it, with more sources for the claims still open. It is the most useful way to come back to a story that is still unfolding.
Say whether an outcome convinces you
You can give your own judgement on the verification and indicate whether the individual checks against the sources seem relevant to you. These are readers' opinions and are kept separate: they do not affect the score, which depends only on the sources.
Dispute it
Anyone who thinks a verification is wrong can dispute it, without needing an account.
FAQ
Who writes the outcome of a verification?
The claims, the sources and the note on each outcome are produced by an artificial intelligence model, following fixed rules that are the same for everyone, and every verification says so. Everything else is calculated by code: the language measures, the score, the band, the summary. This keeps the part that interprets separate from the part that counts, and the second can be redone by hand.
Why do you never say «true» or «false»?
Because that is not what a verification can establish. We can say what we found in a certain number of sources, at a certain moment.
Are verifications by the PINTRU newsroom any different?
No. Some of the verifications on news outlets are prepared every day by our newsroom, with the same rules and the same calculation. In addition, before publication, every contradiction is reread by a further check.
Does a low score mean the article is false?
It means the sources consulted do not support it, or contradict it. On very recent events the sources may simply not exist yet: that is one of the reasons a verification can be extended and its outcome can change. The score should always be read together with the claims and the sources behind it.
How precise is the score?
As precise as the claims it is built on allow. With four claims the possible results are barely enough, which is why the number of claims is always shown next to the score. Extending a verification increases the number of sources, and with it the quality of the result.
Is it an official assessment?
No. It is a documented, repeatable check, not a legal opinion or a certification, and it does not bind the publisher of the page.
Put it to the test
Paste the address of an article you read recently and see, claim by claim, what the sources say about it.