LeanHallucinationGuard is a deterministic enforcement layer for AI reasoning. It intercepts model outputs, validates geometric and semantic invariants, blocks operator invention, and records drift‑collapse telemetry to ensure grounded, invariant‑preserving computation.
spectral-analysis semantic-validation drift-detection lean4 hallucination-detection ai-interpretability structural-ai latent-manifold geometric-interpretability reasoning-invariants operator-typing modal-adjacency collapse-surfaces provenance-enforcement safety-enforcement
-
Updated
Oct 14, 2026 - Python