New statistical method can flag hidden anomalies in scientific datasets

Veriqen analyzes the article text available from publishers and looks for explicit sourcing language, such as references to reporting organizations, agencies, statements, and other identifiable sources. Articles that cite the same identified source can then be grouped together.

• Not all sourcing is written explicitly, so some real sourcing goes undetected.
• Publisher feeds may provide limited article text to analyze.
• Two articles citing the same source do not necessarily use the same report or come from the same origin.
• No detected sourcing does not mean an article has no sources.
• Veriqen traces attribution — it does not determine whether a claim is true or false.

Veriqen looks for two kinds of explicit, structured assertions in the article text available to it: numbers tied to a recognized category (like a count of people injured, a dollar amount, or a percentage), and statements attributed to a named or role-based source. Similar assertions found across multiple articles in this story are grouped together and shown as one entry.

• “Appears in N articles” counts how many articles contain a similar assertion — it does not mean N sources independently confirmed it. Articles that cite the same reporting source (shown separately, where detected) are not independent of each other.
• Not all comparable assertions are written explicitly, so some go undetected.
• “Not detected in the available text” never means an article has no sources.
• Attribution to a source is not a claim that the underlying statement is true — only that it was attributed.
• Veriqen does not assess whether any claim, outlet, or headline is accurate or biased.