How we check
We examine the claim, trace the evidence, test the strongest counterarguments, and explain what can, and cannot, be established.
No method is infallible. Ours is built to hold a card back sooner than publish a wrong verdict. We publish up to twice a week, and only when the central part of a claim can be settled.
One label outside, three readings inside
| Outside | One label from a closed list, always with its reason in the same image: Holds up, Probably holds up, Accurate wording misleading message, Mixed, Probably does not hold up, Does not hold up. |
|---|---|
| Each part | False, Probably false, Probably true, True, or No verdict with the reason. |
| Evidence | Type of source (a record that is the fact, a measured series, direct observation, an official report, a secondary source, an interested party) and corroboration. |
| Mechanism | How the wording misleads, from a checkable fact to invented or altered content. Never who, never why. |
The gates
| 0 Intake | An archived capture of what really circulates. |
|---|---|
| 1 Split | At most three parts; one is central. |
| 2 What is judged | The explicit text and the factual premise it assumes. |
| 3 Evidence read by us | Is it authentic? Can it prove exactly this? Could the process behind it be wrong? True or False only on a record, a measured series or direct observation. |
| 4 Chain of origin | Find the first node. Sources that share an origin count once. |
| 5 What would change it | Written before publishing. |
| 6 Second check | Proportional to risk. Figures are recalculated in code. |
| 7 Human validation | Four fixed questions, including the right of reply. |
What we do not publish
Claims without enough evidence, claims with solid evidence on both sides, and images without text. They stay on an internal list and are reopened if evidence appears.