Policy-as-logic for robust reasoning over rules

TL;DR - A hybrid system converts written policies into formal logic, uses language models to extract facts, and delegates rule-based decisions to an answer set solver. This separation improves interpretability, robustness, and efficiency for policy-grounded question answering.

• Outperforms policy-as-prompt and policy-as-code approaches in most evaluated cases.
• Reduces token usage by approximately 10×.
• Produces auditable answers by separating predicate grounding from symbolic reasoning.
• Remains robust under input perturbations when applying objective policy criteria.