Extract Facts

Given web content or a URL, extract a set of discrete, independently verifiable facts. Each fact must be atomic (one claim per fact), attributed to its source, and accompanied by a confidence indicator.

Procedure

1. Receive source content (raw text, URL, or list of URLs).
2. If URLs are provided, fetch each one using web_fetch.
3. Parse the content and identify factual statements. Distinguish:
   • Hard facts: specific figures, dates, names, events with a clear source
   • Soft facts: general statements of consensus that are widely accepted
   • Opinion/editorial: explicitly exclude these
4. For each hard fact, record:
   • The fact statement (one sentence, active voice)
   • The source URL
   • The source date
   • A confidence level: high, medium, or low
5. Cross-check facts against other sources when possible. Upgrade confidence if confirmed by a second independent source.
6. Return the extracted facts as a structured list.