Inside our verification graph: how Ethur reasons over conflicting records
A look at the entity-resolution pipeline that powers identity, employment, and education verification at scale.
When you verify a candidate, you're rarely working from a single source of truth. You're reconciling claims (what the candidate told you) with evidence (what data sources confirm) — and those sources contradict each other constantly.
The data we reconcile
- Identity: government ID, SSN trace, address history, biometric match.
- Employment: payroll, employer-of-record APIs, direct verifications, tax records.
- Education: National Student Clearinghouse, registrar APIs, document verification.
- Licensing: 4,200+ state and federal license boards.
The graph model
We model every claim and every piece of evidence as nodes in a typed graph. Edges carry weight (source reliability), recency (when the evidence was generated), and provenance (cryptographic chain back to the source). Conflicts surface as cycles in the graph, which our resolver scores against learned priors.
Why graphs win over rules
Rules engines collapse under the weight of edge cases. By the time you've encoded the 200th 'if license expired in NY but renewed in NJ' rule, you've built an unmaintainable mess. Graphs let us reason compositionally and explain every decision back to the source evidence.
Every verification result Ethur returns can be traced back to the exact byte of source data that supports it. That's the bar for modern background checks.