Trace is a provenance-tracing tool. Paste a URL and it reconstructs where an article's claims actually come from: what it cites, what cites it back, what's a near-verbatim copy elsewhere, and — when the corpus already has enough to tell — the earliest occurrence of each claim.
Nothing here is invented. Every confidence level, timeline date, and "unverified" flag is derived from evidence already on the page or already in the corpus — Trace never fabricates certainty it doesn't have.
- Article extraction — title, author, date, full text, and outbound links pulled from any URL via Readability, with SSRF protections against internal and private targets.
- Claim detection — statistics, attributions, and cited facts are pulled out sentence by sentence and tied to their paragraph.
- Citation & copy graph — resolves
citesandcopied_fromrelationships across every article in the corpus, including which side is chronologically earlier when dates allow it. - First-seen tracking — finds the earliest dated occurrence of a claim across the corpus, explicit about the difference between "earliest found" and "confirmed origin."
- Confidence, never invented — every claim and relationship gets an evidence-based level (high / medium / low / unverified) with the reason attached, never a made-up percentage.
- Divergence detection — compares a claim's wording against its resolved source's actual text to flag number mismatches or claims the source doesn't actually support.
- On-demand source chasing — Brave Search-backed source discovery and depth-limited chain following, triggered per claim, one request at a time.
- Optional Firecrawl fallback — for sources a plain fetch can't reach past bot protection, available once signed in.
- Interactive provenance graph — an auto-laid-out, pan-and-zoom graph of the whole trace, with an inspector panel and a timeline of every dated event found.
- Per-account history — signed-in users get a private, real trace history; anonymous visits are never recorded.