TruthGraph: The right rule depends on when—and for whom

TruthGraph is a new evidence-based AI agent designed to navigate complex rulebooks by accounting for temporal and contextual variables. Developed for the Sanity Challenge, the tool addresses the limitation of standard LLMs that often ignore specific match conditions or historical rule changes. By utilizing Sanity Context to query structured collections of sports rules—including football, cricket, and chess—TruthGraph ensures that answers are grounded in official documentation rather than generic model training. The system evaluates claims against specific dates, match types, and relationship hierarchies, providing users with transparent citations and a comparison against keyword-based baselines. Built with Next.js, TypeScript, and the Vercel AI SDK, the project demonstrates how structured data and deterministic evaluation can improve the reliability of AI-generated responses. TruthGraph is currently open-source, allowing developers to explore its architecture for managing evolving knowledge bases and ensuring factual accuracy in domain-specific queries.