How Clarity Today reads

How we work

Clarity Today is written by software. Language models draft every article, headline and summary from the reporting it cites, and no person reads each article before it publishes.

What is automated

Software does every step: finding reports, grouping the ones that describe the same event, choosing which events become stories, drafting, the claim ledger, the checks below, and the order of the front page. The model that drafts an article also writes each factual claim it makes into the ledger, with the source it came from. Newsletter editions are written by models from published stories, and other models check each edition against those stories before it is sent. Ask Clarity answers a reader’s question about a story with a language model.

Automated checks run on every draft before it publishes:

• Every factual claim goes into the article’s claim ledger with the source it came from. An article that cites a claim missing from its ledger does not publish.
• An editing model reads the draft for the habits of machine-written prose and may revise its wording.
• Our policy is that every quotation appears word for word in a cited source. A quotation that does not match is paraphrased or cut, and the article does not publish if it still fails. Where we captured too little of a source’s text to compare, the check cannot confirm the quotation.
• A long passage copied from one source without quotation marks is rewritten as an attributed quotation or a paraphrase, and the article does not publish if it still fails.
• A second model checks each headline against the claim ledger. A headline the claims do not support is rewritten once and checked again; if it still fails, or the check cannot run, the article does not publish.
• “N publishers” counts distinct publishers whose article text we could read, and leaves out an article that is a near-copy of another publisher’s. It never counts links or repeats.
• A daily check revisits cited source links, oldest-checked first. Once a link is confirmed dead the page stops linking it; an uncertain result never removes a link.

Bylines name house desks, not people, and Clarity Today, the organization, is the author in our structured data. A picture is either the publisher’s own photograph, credited and linked, or a generated illustration labeled as one.

How claims are checked

After a story publishes, a separate model reads its claims against the reporting we collected on the same event. For each source it records whether the source supports a claim, contradicts it or only adds context. It then gives each claim a state, shown in the story’s claim list: Supported, Contested where the reporting disputes it, or Insufficient where the sources do not establish it. A claim shows no state until this check has run.

A source that contradicts a claim stays in that claim’s record, and signed-in readers can open each claim’s evidence source by source. When more reporting on the event arrives, the check runs again, so a claim’s state can change after publication.

Corrections

If a story gets something wrong, send the story link and what is wrong to contact@theclarity.today. North Pulse LLC is accountable for these standards. When a story changes, its link still leads to it, and a story that changes on a later day than it was published shows when it was last updated.

Who publishes Clarity Today

Clarity Today publishes theclarity.today and its newsletters. For corrections or anything else, write to contact@theclarity.today.

North Pulse LLC also operates theclarity.us, Clarity’s personalised morning edition, and it is made the same way: language models write each reader’s edition and its podcast from the reporting they cite, synthetic voices read the podcast, and no person reads an edition before it reaches a reader.

We use automation so six desks can cover far more of each day’s reporting than anyone could read by hand, with every claim traceable to the outlet that reported it.

Readers in the EU and EEA see the mark “AI-written · How we work”, which links here, next to each article’s byline and in each newsletter’s footer, on theclarity.today and on theclarity.us. In ChatGPT, our story cards show it when ChatGPT places the reader in the EU or EEA. The Atom feed and the agent server carry it once.