What problem does it solve?
This skill performs a rigorous, point-by-point factual audit of a synthesized document's key assertions. It is a critical quality control and validation step that operates by deconstructing the synthesis output into its constituent factual claims, tracing each claim back to its cited source material, and performing a detailed comparison to establish veracity. The physical process involves parsing the synthesis text to identify discrete propositions that are presented as factual statements, each typically accompanied by one or more citation identifiers (e.g., [1], [source_a]). For each identified claim, the skill locates the corresponding source content using the provided citation mapping. It then executes a deep cognitive analysis, comparing the semantic content of the claim against the explicit information, context, and nuance present in the source. This is not a simple keyword match; it involves evaluating if the claim is directly supported, indirectly inferred, partially accurate, exaggerated, contradicted, or unsupported by the source text. The skill applies specific data transformations: it ingests a synthesis object and a sources object, extracts claims, maps them to source snippets, runs a verification LLM process, and outputs a structured verification report that annotates each claim with a detailed verdict and evidentiary reasoning.
Core Features & Use Cases
- It identifies discrete factual claims and associates them with exact citations from sources.
- It compares claims to source content and assigns verdicts such as supported, contradicted, or unverifiable.
- It outputs a structured verifications payload suitable for downstream QA, editing, or governance reviews.
Quick Start
Provide the synthesis output and the corresponding source documents to the verification pipeline to obtain per-claim verdicts.
