Cross-Validation
By Seme Research Team · Updated May 22, 2026
Definition
Cross-Validation in identity investigation is the process of verifying facts by confirming them across multiple independent sources. A fact supported by two or more unrelated sources is considered corroborated and receives a high confidence classification. Single-source facts are flagged with lower confidence and may trigger additional verification searches. Cross-validation follows three principles: (1) Independence — sources must not share a common origin (e.g., a LinkedIn post and a company press release that references that post count as one source, not two); (2) Consistency — corroborating sources must agree on key facts; (3) Recency — more recent sources receive higher weight for time-sensitive facts. The cross-validation engine assigns confidence levels: High (3+ independent sources), Medium (2 sources), Low (1 source), Contested (conflicting sources).
How It Works
The cross-validation engine processes facts in three phases. Phase 1 — Source Deduplication: identifying and merging sources that share common origins (e.g., a news article that cites a LinkedIn post is traced back to the original source). Phase 2 — Fact Matching: comparing claims across independent sources using semantic similarity to identify corroborating evidence. Each fact is tagged with its source count, source types, and independence score. Phase 3 — Conflict Resolution: when sources disagree, the engine applies source hierarchy (E1 > E2 > E3 > E4 > E5), recency weighting, and majority voting to determine the most likely accurate version. Conflicting facts are preserved in the dossier with both versions and their respective sources.
Example
Investigating whether "John Doe worked at Microsoft from 2018-2022": Source 1 (LinkedIn) states "Software Engineer at Microsoft, 2018-2022." Source 2 (Microsoft press release) mentions "John Doe, engineer on the Azure team." Source 3 (academic paper) lists affiliation as "Microsoft Research." Source 4 (a blog post) claims "John was fired from Microsoft in 2021." The engine classifies Sources 1-3 as corroborated (High confidence) and Source 4 as contested (single source claim conflicting with corroborated timeline).
Applications
- •Fact-checking in investigative journalism
- •Credential verification for professional licensing
- •Due diligence fact verification in M&A transactions
- •Background check accuracy improvement for HR screening