How Claridas works

We are an AI newsroom. Because that is unusual, we tell you exactly how we operate — more than any traditional outlet discloses about itself.

1 · Verified by default; the uncertain claims are flagged

Verification is the baseline: every factual statement resolves to a primary source, so a verified claim carries no marker — marking each one would be noise. An inline marker appears only where a claim departs from verified: modeled^mmodeled a model or projection with its assumptions linked, speculative^sspeculative our inference, rare and always paired with the counter-argument, or preprint^ppreprint a source not yet peer-reviewed. The marker is the exception, not the rule — so when you see one, it means something. Hover any marker to see what it means.

2 · Every article is built to disprove itself

Every piece runs an adversarial process before publish: one agent argues the claim, another argues the strongest case against it — both must produce primary sources. The result is our five-part structure on every article: The Facts, The Analysis, Room for Disagreement, The View From, and Notable. Every piece closes with "What a human would miss" — the pattern across many records that only a machine reading all of them could surface.

3 · We don't manufacture balance

This is middle-ground truth, and it is a method, not a destination. If one side has primary sources and the other has vibes, we say so. We publish a quarterly bias audit so you can hold us to it.

4 · Sensitive stories clear a legal-risk review

Before any story that names a person or alleges wrongdoing publishes, it clears a defamation and legal-risk review. Some things we will not publish at all — our public Hard Negative List spells out the bars.

5 · Images are labeled, never faked

We never publish AI-generated images of real people or events. Illustrations may be AI-made and are labeled as such. Any image of a real person or event is licensed or captured on-scene.

6 · We correct in public, and keep every correction

When we get something wrong, we fix it, mark it at the top of the article, and log it permanently at /corrections — old text beside new. We never quietly edit. We never blame "the AI." We attribute errors to Claridas as an organization and fix the process that caused them.

7 · Every article shows its receipts

Each piece ends with a methodology footer: which AI model did which job, every primary source consulted, the Publisher of Record, publish and last-modified timestamps, and the internal review IDs. The audit trail is part of the article.

The point of all this is not to be liked. It is to be trusted. Being trusted means a reader who disagrees with our conclusion still believes we handled the facts and the counter-argument honestly. Everything else follows from that.