Time Throws Fire: Free-Energy Destruction and the Physical Legibility of Cosmological Time

Why can a changing universe become historically legible? Temporal geometry alone does not explain why later physical states contain accessible evidence of earlier transformations, and entropy production alone does not determine what those records say. This paper develops a mechanism-level architecture of historical consequence. A supplied law, state, measure, boundary/preparation data, correlations, factorization, dynamical assumptions, quantum-state structure, and model class determine a history law Γ; history-sensitive interactions create encodings E; some encodings persist, copy, transform, or propagate into descendants P; authenticated observers access only a subset A. The chain is not a universal theorem: every arrow requires a physical channel. A reversible one-bit copier demonstrates that prepared memory can convert ln2 of local nonequilibrium resource into ln2 of event-memory correlation while the same interaction with equilibrium memory records nothing. A stationary three-state cycle demonstrates path-reversal distinguishability without boundary bias, separating directional source from recording resource. A detailed-balance relaxation example separates physical reversal from Bayesian retrodiction and shows directional path evidence increasing while endpoint historical memory decays. Quantum data processing bounds origin information available in downstream records, and an explicit fragment model shows arbitrarily redundant records of a history variable with zero information about an upstream origin label. Radiative, gravitational, horizon, biological, and observational channels can therefore make histories durable and accessible without making their cosmological source uniquely invertible.