A Compressibility Selection Principle for Physical Laws

This work proposes a selection principle according to which physical laws can be modeled as if selected to maximize long-term predictive compressibility under energetic constraints. Rather than modifying established dynamics, the framework defines a variational functional over possible histories, favoring universes in which prediction is energetically viable. Evolutionary simulations demonstrate a critical compressibility threshold above which learning systems discover true environmental symmetries and below which they overfit noise. The principle offers a unified explanation for fine-tuning, the effectiveness of mathematics, and the emergence of intelligence, while remaining compatible with known physics. Empirical validation remains open.