Keeping fact extraction separate from the claims built on top of it
Abstracts
Abstracts
The crawler agent prototype has one narrow job: fetch approved public links and pull out timestamped facts, before any of that becomes a written claim about someone's work. That separation matters more than it sounds. Once extraction and claim-writing are tangled together, it's easy for a summary to drift ahead of the evidence behind it. Keeping them apart means every downstream update can be traced back to something the source actually said, which is the quiet infrastructure work that makes automated profile updates worth trusting in the first place.