The arXiv preprint 'Computable Fairness: Boltzmann-Softmax Control for AI Resource Allocation' (submitted April 2026) introduces Computable Fair Division (CFD), reframing the Boltzmann-Softmax function as a tunable allocator for scarce resources like GPU cycles rather than a mere selector. By treating the inverse temperature β as a real-time control variable, the framework creates a Pareto frontier where efficiency and fairness trade off along a 'Stability Corridor' of near-constant total loss. In dynamic tests, the AHC++ controller uses dominance-error feedback to adjust β, suppressing concentration spikes under shocks while preserving throughput. This work is a preprint and has not undergone peer review. Methodology consists entirely of simulations with no specified number of independent runs or statistical power analysis; scalability was probed via a 100x agent increase yielding ~5.5x runtime growth. Limitations include absence of real-world deployment data, sensitivity testing to noisy dominance measurements, and validation against heterogeneous agent utilities common in production clusters. Prior coverage focused narrowly on throughput gains, missing the deeper link to resource ethics: as AI training clusters grow, unchecked dominance mirrors historical patterns in cloud markets where early movers captured disproportionate capacity, echoing findings in related work on multi-agent systems. Synthesizing with Barocas et al. (Fairness in Machine Learning, 2023) on allocative harms and Parkes & Seuken (Algorithmic Game Theory for Resource Allocation, 2022) on mechanism design, CFD offers a computable bridge between static fairness axioms and online control, yet risks entrenching target-setting power in operators who define 'acceptable' dominance thresholds. The approach thus reframes fairness not as post-hoc auditing but as an embedded, adjustable feedback loop—an angle underexplored amid efficiency hype.