Distributed Exergy Commons: Metrics, architectures and protocols for the AI-native extraction of diffuse ambient exergy

Diffuse ambient exergy (thermal, pressure and humidity gradients, gravitational head and temporal oscillations) is largely ignored not because it is scarce, but because the cost of collecting it, dominated by heat-exchange conductance and aggregation overhead, exceeds the value of any single node. This preprint proposes "Exergy Commons", a single engineering framework to screen, design and aggregate such sources. Part I (framework) introduces: a scaling law P = ξ·G·ΔT²/(4T̄) and a break-even conductance cost κ* that screens sources before they are designed; a theory of temporal exergy, in which an inertial store acts as a high-pass filter on environmental signals (Spectral Exergy Density); a Lift-over-Convert principle (below a temperature difference of about 30-40 K it is better to raise the working temperature of existing machines than to generate electricity); the Firm Exergy Stack, a layered design for constant electricity governed by a Spectral Firming Principle and a Sink Margin rule (in an illustrative simulation, at least 95% of target power is delivered in 97-99% of hours with a 2-4 h battery); an AI-native aggregation layer (urban exergy cadastre, open resource descriptor protocol, exergy-aware dispatch); and a numerical screening of solid-state converters, showing that magnetohydrodynamic conversion in water is three to six orders of magnitude too weak and that piezoelectric, triboelectric and magnetostrictive devices are sensor supplies, not generators. Part II (gravity-driven columns) develops a screening figure of merit for working-fluid columns based on the standard real-fluid adiabatic gradient Γ = gTβ/cp. It shows that extra turbines and "stacked" gradients add no energy, that fluid choice trades tower height against source temperature, and, through an exact Gouy-Stodola exergy balance, that a closed air loop does not beat an open updraft tower. A finite-conductance criterion identifies the best fluid at maximum power per unit exchanger conductance. Several results are explicitly acknowledged as prior art (notably US 8,640,461) and are included only in quantified form. A synthesis joins the two parts: 27 falsifiable hypotheses, an implementation plan, and a discussion of limitations and prior art.