Modern cosmology balances its energy budget by introducing dark matter and dark energy to the bulk, yet continues to treat real physical boundaries as essentially free, even though every stable structure pays to exist at its edges. This work asks what changes if we enforce that edge cost. Staying strictly within general relativity and a standard collisionless dark matter component, we propose a Boundary Bookkeeping framework with two coupled ledgers. The Local Ledger accounts for non-thermal pressure support at the boundaries of halos, clusters, filaments, and the circumgalactic medium, and uses gravitational lensing as an external auditor of the true mass profile. The Global Ledger encodes a minimal thermodynamic overhead associated with the cosmological horizon in a single dimensionless parameter fglobal, constrained by CMB, BAO, SNe, and structure growth, without modifying Einstein’s equations. We derive the local mass bias implied by non-thermal support, show how a horizon-tied overhead naturally tracks the critical density, and specify concrete falsifiability conditions: boundary-aware models must reduce hydrostatic versus lensing tensions in edge-dominated systems, relaxed systems must remain consistent, and any allowed fglobal(z) must fit standard probes without introducing new anomalies. The result is a conservative, testable framework that treats boundaries as first-class elements of late-time cosmology and lets existing data decide how much of the “dark” sector is truly missing physics versus missing bookkeeping.