Evidence-first methodology

From reports to one event

1. Discover reports from open data, RSS, and official feeds.
2. Normalize URLs, timestamps, language, and source identity.
3. Remove exact and near duplicates, then cluster reports describing the same event.
4. Extract claims with their source identifiers.
5. Compare corroborating and contradictory claims.
6. Generate a story only from retained evidence, then verify every sentence.

Confidence

Confidence reflects corroboration, source independence, primary evidence, contradiction severity, and timeliness. It is not certainty.

A developing label does not mean a story is unimportant or stale. It signals that material facts may change, that independent confirmation is still arriving, or that the central report currently comes from one clearly identified source.

Source independence

Ten websites repeating one wire report are not ten independent confirmations. We track publisher identity and ownership groups, distinguish discovery systems from original reporting, and give primary documents and reputable direct reporting additional weight.

Breaking-news publication

Trusted single-source reports may publish promptly with explicit attribution and a developing label. Nuclear or chemical incidents, uncertain head-of-state deaths, coups, mass-casualty claims, terrorism attribution, election reversals, war-crime allegations, and directly contradictory central claims can be held for an editor.

AI use

Models assist with extraction, comparison, drafting, verification, and social adaptation. Model agreement is never treated as source corroboration.

Updates and corrections

New evidence updates the existing event record where possible. Material versions are preserved, visible correction notices are not silently removed, and retracted stories remain auditable rather than disappearing without explanation.