Provenance

Provenance is the product. Every fact/observation includes:

"provenance": {
  "src_id": "src_kpi2_be16edbfefe50d35",
  "source_provider": "sec",
  "source_url": "https://www.sec.gov/Archives/edgar/data/1543151/000154315126000019/0001543151-26-000019-index.htm",
  "accession": "0001543151-26-000019",
  "quote": "Gross Bookings grew 25% YoY to $53.7 billion, and 21% on a constant currency basis.",
  "content_hash": null,
  "extracted_at": "2026-07-20T10:18:42.762293+00:00"
}

• quote is present for LLM-extracted facts (KPIs, extracted facts) and null for structural facts (XBRL, prices).
• src_id is stable across reprocessing — key it, and it won't churn.

Resolving a source

Every cited fact already includes its direct source_url. GET /v1/src/{src_id} currently validates and describes the stable identifier but returns "resolvable": false; the reverse index needed to recover a URL from a bare src_id is not yet available. ?redirect=true therefore returns a not-covered problem response. Store the citation's source_url alongside the src_id if you persist facts.

The serving gate

A serving-time allowlist guarantees no vendor-sourced row can appear in any response. Only first-party and licensed-redistributable providers pass: sec, signals_transcript, signals_ir, signals_rss, issuer pages, senate_efd, house_clerk, FINRA, and licensed price feeds.