Every source, scored by what it actually did
Media-reliability ratings usually come from a panel's opinion. Ours come from a ledger: every claim a source feeds our pipeline is extracted, verified, human-reviewed, corroborated against independent outlets, and sometimes later contested — and the source's score is computed from that record, nothing else. It updates daily, it's append-only, and the formula is published below. State-affiliation and ownership labels are editor-reviewed before they appear.
These sources have fewer than five desk decisions in the pipeline, so no behavior score is shown — a neutral starting number is not a quality verdict. Their identity and class labels apply as usual; the score appears once real history exists.
Positive weight: claims the desk approved, plus half weight each time the source corroborated another outlet's reporting or was independently corroborated itself. Negative weight: claims rejected, double weight for approved claims later contested, and verifier source-concern flags. The blend is shrunk toward a neutral 50 until a source has real history, so nobody buys a high score with three good days. Duplicate reports folded by the clustering pass are neutral — reprinting a fact isn't a sin, it's corroboration, and it's counted as such. Identity, ownership, and state-affiliation labels are machine-drafted with cited evidence and published only after editor review — the same rule as everything else here. The ledger is append-only: yesterday's numbers never change, and the full history is public on each source's page.
Two different things are shown and should not be confused: source class — what an outlet is (news agency, broadcaster, aggregator, state media) — and the behavior score, which measures only its observed record inside this pipeline. A wire agency with little desk history has a fine class and no score yet; that is a sample-size fact, not a quality judgment.