Scores here are heuristic signals about sourcing and corroboration across outlets — not verdicts on whether a story is true, and not an endorsement or condemnation of any outlet. How this works
This page is the whole story on how scores here are computed — no proprietary black box. There is deliberately no single "truth score" anywhere in this system. Automatically rating whether a news story is genuinely true is not a solved problem, and a single opaque number claiming to do it would be both misleading to readers and an unfounded claim about specific outlets. Instead, every story cluster carries five separate, labeled signals about how it's being covered.
Corroboration score is simply the count of independent outlets we've found covering the same event, after clustering near-duplicate articles together (so five outlets running the identical wire story still each count once). Diversity is the share of those outlets that come from distinct ownership groups — five stories from outlets under the same parent company count as far less diverse than five independently owned outlets, even though the raw corroboration count looks the same.
Every outlet we track is assigned a tier by us, by hand, ahead of time — never by an AI model at scoring time. It's a static, version-controlled list (edited via code review and git history, which is the audit trail for any change), reviewed periodically since outlet ownership and editorial quality do shift. The tier reflects an editorial-process judgment (wire services vs. major mastheads vs. openly partisan outlets vs. international sources), not a judgment about whether any specific story from that outlet is accurate.
Wire Service: Agence France-Presse, United Press International, Associated Press, Reuters
Major Masthead: BBC News, NPR, The Guardian, The New York Times, The Wall Street Journal, PBS NewsHour, CNN, Bloomberg, Financial Times, Politico, Axios, The Washington Post
Partisan-Leaning / Mainstream: Fox News, NBC News, The Hill
International: Al Jazeera English, Deutsche Welle, The Economist, Sky News
Outlets not yet on this list are treated as "Unranked" rather than guessed at. Full list with owner groups: src/config/newsSourceTiers.json in the codebase.
We cross-reference each story's headline against Google's Fact Check Tools API, which aggregates published fact-checks from outlets like PolitiFact, Snopes, and Reuters Fact Check. When a match exists, we link to it directly with its original rating — we don't generate our own rating, and the absence of a match here doesn't mean a claim has been verified, only that no matching fact-check was found in that database.
A short, Gemini-generated note describing the tone and sourcing pattern across a story's coverage — whether language is hedged or definitive, and whether claims are attributed to named or anonymous sources. The model is explicitly instructed never to render a verdict on whether the underlying claim is true, and never to use words like "true," "false," "confirmed," or "debunked." It's labeled "AI-generated read" everywhere it appears, deliberately not styled or worded like a verdict.
A 1-5 salience heuristic — how significant this looks, not whether it's true — computed from a fixed, documented formula (no AI involved): breadth of independent coverage, how fast corroborating coverage is appearing (velocity), and the best reliability tier among contributing outlets. The exact weights are in code: src/lib/news/scoring/impact.ts.
If you're an outlet listed here, or a reader who thinks a story's tier, corroboration, or AI consistency note is mistaken, email news-corrections@nextlevelintell.com with the story link and what looks wrong. We review every report against this methodology.