Cognitive Nutrition for Foundation Models: Empirical Validation of the PEST Framework for Provenance-Verified, Semantically Dense Training Data

In The Entropy of Recursion, we introduced the Provenance-Embedded Semantic Transport (PEST) framework as a theoretical response to model collapse and context blindness. This paper presents the empirical realization of that vision. We report on a production metadata factory that processed 10,097 cultural heritage artifacts into provenance-verified, semantically enriched "Aeternum Assets," each carrying 2,000–6,000 tokens of structured context across 111+ fields organized into four ontology layers. Central to our approach is the Translation Paradigm: enrichment is machine-assisted articulation of existing human knowledge. The pipeline employs multi-model Oracle Synthesis, orchestrating complementary frontier models to achieve depth, nuance, and structural precision. We propose a mathematical framework for evaluating Cognitive Nutrition using topological data analysis (persistent homology) and causal disentanglement (back-door adjustment). We introduce the Amendability Framework—a living curation model where originating institutions retain governance over their enriched collections.