Method
Tally. Compare. Filter. Predict.
How a verdict is reached, and why a measurement stands on its own while a claim needs a person.
Four steps
From a claim to a verdict.
Tally
Count the coverage carrying the claim from outlets graded credible. Thin coverage is reported as thin.
Compare
Six independent signals weigh in. No single one can make a claim verified; the one input decisive on its own is a published false rating.
Filter
What nothing corroborates cannot be verified. It is flagged, unverified at best, and only what a person verified reaches the customer feed.
Predict
How fast a narrative is growing, how soon it doubles, and the precedent for that kind of narrative show where a claim is heading.
Measured, or reported
A reading stands on its own. A claim needs a person.
Measured
A seismic reading, a fire detection, an airspace count. It is kept as an observation with its source and its revisions, is never given a claim verdict, and never enters the claim queue.
Reported
Someone says a thing happened. The six signals weigh it. Before it reaches the customer feed a person verifies it with a confidence score; dismissing it takes a reason, and each decision is audited.
The verdict format
Every verdict shows its working.
| cross-source corroboration | SUPPORTS |
|---|---|
| the fact-check record | NEUTRAL |
| spread velocity | NEUTRAL |
| prior-narrative correlation | NEUTRAL |
| known manipulation patterns | NEUTRAL |
| models weigh, humans verify | SUPPORTS |
| VERIFIED | The signals weigh clearly for the claim, and at least one corroborates it. |
|---|---|
| UNVERIFIED | Nothing corroborates it, or the signals do not settle it either way. |
| DISPUTED | The signals lean against the claim. |
| LIKELY FALSE | The signals weigh clearly against it. |
| FABRICATED | Fact-checkers have rated it false, or the signals weigh overwhelmingly against it. |
A check that could not run, because a source was down or a key is not configured, is listed as unavailable. It counts neither for nor against the claim.
Next