Researchers have developed a system called LEDGER to help review and audit the outputs of large language models. The system creates layered graphs that show how the model arrived at its conclusions, including which actions it took, what artifacts it created, and what validation steps it performed. This can help reviewers understand whether the model's outputs are trustworthy. The system preserves detailed execution records while grouping them into nodes that represent evidence and workflow decisions. --- Why it matters: This matters to researchers in AI because large language models are becoming increasingly powerful and widespread, but their outputs need to be audited for trustworthiness. LEDGER provides a tool for doing this efficiently and effectively. Source: https://arxiv.org/abs/2608.18398