Our method
How we label claims.
Every factual claim in a Reconstructed story carries one of four labels. The label tells you how strong the evidence is right now, not whether a claim is true forever. When the evidence changes, the label changes, and we log the change on the story. We decide in this order: a credible contradiction makes a claim Disputed; independent verification makes it Confirmed; otherwise it is Claimed when only someone involved says it, or Reported when one credible source does. If none of these fits, we leave the claim out.
Independently verified, and nothing credible says otherwise: an official record shows it (a court filing, official data, an authority's statement about its own action), or at least two independent news organizations each confirmed it with their own reporting.
One credible news organization's own reporting, or one official source, that nobody has independently confirmed yet. We always say who reports it.
Said by someone with a stake in the story, such as a suspect, a lawyer, a prosecutor, a company or a government talking about itself, and not independently verified. We always say who claims it.
Credible evidence or another credible source contradicts it, and it isn't settled. We say who disputes it and why.
Where facts come from
Official records and statements (investigators, regulators, courts, agencies), wire services, established newspapers and broadcasters, and primary data such as flight-tracking or weather records. Aggregators, rewrites, Wikipedia, partisan sites and social posts can point us to a story, but never count toward a label. We save a copy of every source we use. Each claim links to the sources behind it, and any claim that isn’t Confirmed is attributed in the text (“according to…”, “the airline says…”).
Two things never count as independent confirmation: the same wire story republished by many sites (that is one source), and several outlets repeating the same person’s statement (that is one statement). A company, a government or a lawyer speaking about its own case is a party to it, so what they say stays Claimed until someone independent verifies it.
How a story is checked
We use AI tools to help gather and organize sources, and every story then goes through checks before it’s published:
- Every claim must quote its evidence, and software matches that quote word for word against the saved source. A claim whose evidence can’t be found is removed.
- Software checks each label against the evidence rules above and lowers any label the evidence doesn’t support.
- A second, independent AI review, on a different model, tries to prove every label wrong using the same saved sources. Where the two disagree, the more cautious label wins, and anything it finds contradicted is held for a person to decide.
- The causes of crashes are never Confirmed before an official finding, guilt never before a conviction, and death tolls only from an official count, with the time it was given.
- Every new story is read and approved by a person before it is published. People accused of crimes are presumed innocent, and we say so.
Our narration audio and illustrations are AI-generated, and we say so. Our 3D reconstructions are illustrations of what sources describe, not footage.
Corrections
If we get something wrong, we fix it on the story and add a dated note under “Corrections and label changes”. If you follow the story, you’ll get the update by email. To report an error, email info.reconstructed@gmail.com.
What we don’t claim
We don’t claim to be unbiased or to cover every source, and a label is not a legal finding. We show you what the evidence supports today, and where it comes from, so you can judge it yourself.