VerifiedSignal evaluates documents through eight intelligence lenses—logical fallacy exposure, factuality confidence, AI authorship likelihood, pseudoscience signals, fiction vs. fact separation, provenance, semantic relevance, and collection-level analytics.
Single-number “trust” metrics collapse distinct failure modes. A document can be factually tight yet manipulative, or well-sourced yet machine-generated. Dimensions let operators route review, compare sources, and explain decisions to stakeholders.
Users group documents into collections to compare scorecards and watch aggregates shift over time—useful for monitoring outlets, issuers, or internal document classes.
The review experience is explicitly side-by-side: proposed extractions and scores stay anchored to passages so humans can correct low-confidence fields without losing provenance.
Long-running analyses stream stage progress via Server-Sent Events so scorecards feel responsive even when backend work spans minutes.