Note: This work has been integrated into and superseded by a unified theoretical framework. For the most current version, including the finalized 7-Bridge taxonomy and implementation protocols, see: https://doi.org/10.5281/zenodo.18120609 Current approaches to Artificial Intelligence rely on private, opaque datasets that prioritize volume over veracity. We argue that this enclosure of training data poses a systemic risk to public epistemology. While recent advances in "World Models" and "Embodied Cognition" attempt to ground AI in physical laws, they fail to account for the abstract, high-stakes domains of human civilization - law, finance, and ethics. This paper bridges that gap. We propose a General Theory of Incentive-Compatible Semantics (ICS), positing that whatever can be empirically known - whether through physical mechanics, biological survival, or legal liability - can be harvested to ground Artificial General Intelligence (AGI). We present an exhaustive "Atlas" of fourteen "Truth Bridges" that unite the physical (e.g., Meteorology) with the institutional (e.g., Solvency). We assert these domains as a New Public Domain - a shared, open-source infrastructure of "Hard Data" designed to protect the quality of AI training and ensure that the growth of Superintelligence is rooted in the open, verifiable reality of the public commons. This paper presents concepts adapted from a forthcoming book by Gregory Caldwell Beier.