Hierarchical Predictive Intelligence (HPI):Support-Induced Ternary Projection

This paper introduces the Hierarchical Predictive Intelligence (HPI)  framework—a theoretical blueprint for autonomous agents capable of scientific discovery. We unify latent predictive world modeling (JEPA) with rigid symbolic inference. The core mechanism is the transition from continuous representations to discrete structures via non-invertible coarse-graining operators and Brusentsov’s ternary logic. We define epistemic lacunae as structural voids in the world model that trigger the formation of structured knowledge. Finally, we prove the asymptotic consistency of the conflict resolution operator based on the Minimum Description Length (MDL) principle under identifiability and coverage constraints.