Why ProfileScribe separates fact extraction from published claims
Abstracts
Abstracts
Most of the work on ProfileScribe this stretch has gone into the crawler agent that feeds it. Before anything reaches profile memory, the crawler pulls from approved public sources and normalizes what it finds into timestamped facts, kept distinct from the language ProfileScribe eventually uses to talk about someone's work. That separation is deliberate: it is easy for a system that drafts confident-sounding copy to blur the line between what a source actually shows and what would just read well. Keeping extraction and claim-drafting as separate steps, with a review queue in between, is a small architectural choice, but it is the one that determines whether the output stays trustworthy as the source graph grows.