TRUST / REGISTRY PROVENANCE

Every claim. Mapped.

For each company-level fact, the registry that issues it and the cadence we recompute. Anchored vs enrichment is explicit.

7 + 50
Federal + state anchors
Weekly
Recompute cadence
Anchored
vs enrichment, explicit

Claim → registry map

Claim Anchored to Status Recompute
Company exists State SOS filing + IRS BMF Anchored Weekly
Legal name State SOS filing Anchored Weekly
EIN IRS BMF Anchored Quarterly (IRS)
LEI GLEIF golden Anchored Daily
UEI / CAGE SAM.gov Anchored Daily
CIK + filings SEC EDGAR Anchored Daily
FDIC cert FDIC institutions Anchored Weekly
NPI NPPES Anchored Weekly
Registered address State SOS or SAM.gov Anchored Weekly
Operating address Commercial + bidstream consensus Enrichment Weekly
Domain WHOIS + Common Crawl + Delivr Enrichment Weekly
Phone Commercial sources Enrichment Weekly
Industry (NAICS) SAM.gov primary; commercial fallback Anchored where SAM-listed Weekly
Employee count Commercial estimate Enrichment Weekly
Revenue SEC for public; commercial estimate otherwise Anchored for public; enrichment for private Quarterly
Parent / subsidiary GLEIF L2 + SEC Exhibit 21 + Wikidata Anchored where reported Daily / Annually
Provider specialty NPPES + state board Anchored Weekly
Bank charter type FDIC + NCUA Anchored Weekly / Quarterly
Federal contractor flag SAM.gov Anchored Daily
501(c) status IRS EO Anchored Monthly

What "recompute weekly" means

The weekly NebulaGraph rebuild reads the latest Iceberg snapshot of every Tier 2 registry feed. If a registry publishes daily (SAM.gov, GLEIF, EDGAR), the snapshot used is at most six days old. If a registry publishes monthly or quarterly (IRS BMF, IRS EO), the snapshot reflects the most recent publication.

Reconciliation runs after the registry feeds land. The output table companies_verified holds the consensus row plus pointers to the underlying claims.

Anchored vs enrichment

Anchored claims have a single authoritative issuer that we can re-fetch independently. If the registry says it, the graph says it. If the registry retracts it, the next rebuild retracts it.

Enrichment claims come from commercial datasets, behavioral inference, or third-party APIs. They are useful but not authoritative. Confidence on enrichment claims is weighted lower and is reported alongside the value.

Where the boundary matters

For compliance use cases (KYB, sanctions adjacency, OFAC pattern matching), only anchored claims should be relied on. For commercial use cases (account scoring, intent ranking, ABM targeting), enrichment claims add lift and the confidence field should be honored.

Related