Editorial policy

Last updated 25 September 2026

Data first

Every page starts from filing data, not from a story. Topics for articles are chosen from what the filings contain (a new report, a large trade, a cluster of insiders buying) and each article is written from a structured summary of those filings: the facts object.

How an article is made

1. The facts object is built by code from the filings, with a link to every source document.
2. Claude, an AI model made by Anthropic, drafts the article from the facts object only.
3. A validator checks the draft against the facts: every number, date, ticker and name must appear in them, and it rejects advice ("you should buy", "will rise"), accusations and outside links. A draft that fails is redrafted once, and any remaining problems are shown to the reviewer.
4. A person reads the draft, approves it and publishes it. Nothing is published automatically.

Articles say so at the foot of the page, list their sources, and have a Markdown version. Executive-branch pages are generated from the data with templated wording and carry no hand-written claims.

Neutral about people

The people in these filings are named because the law requires them to disclose. We write "reported" and "disclosed", link the source filing, and do not suggest motives or wrongdoing. Trades are shown whether they gained or lost.

Independence

No one pays to be included in, ranked in or left out of the data or the articles. The free app shows sponsor and partner offers; they are labelled, kept apart from the data, and are not endorsements. Pro users see none.

Not advice

Nothing we publish is a personal recommendation. See the disclaimer.

Corrections

We fix errors and log them on the corrections page. Explainers and comparisons are dated with when they were last checked.