The verification of 'synthetic text'—documents produced by Large Language Models (LLMs)—requires a shift from checking facts to auditing linguistic and rhetorical signatures.
Generative models often exhibit 'Low Linguistic Variance.' This means the text relies on highly probable word sequences and repetitive rhetorical structures. Veridex analyzes the complexity and perplexity of document fragments to identify these machine-generated patterns.
Identifying unnatural sentiment consistency typical of synthetic corporate or legal text.
Extracting atomic assertions and verifying them against первичный source records.
Unlike human error, AI 'hallucinations' often appear as highly confident, grammatically perfect statements that are functionally impossible or chronologically inconsistent. Verification involves cross-referencing named entities (actors, dates, treaty names) against verified temporal archives.
A low Verity Index on a document indicates a high density of 'low-perplexity' text segments combined with factual claims that lack primary-source correlation.
View Platform Methodology