A Similarity–Complexity Scaling Law in Late-Time Cosmological Residuals

I report an empirical scaling law relating ordering-sensitive residual complexity to cosmological scale in Type Ia supernova distance data. Using a compression-based structure metric applied to guardrailed residual streams, I find that the excess complexity ∆K obeys a stable, monotonic scaling with redshift aggregation scale and truncation depth. The law persists across admissible preprocessing choices and null constructions, and collapses only when similarity structure is explicitly suppressed. This Similarity–Complexity Scaling Law is a property of the inference stream itself and exists independently of any cosmological model. Its existence implies that late-time distance inference generically admits scale-dependent bias channels even in the absence of new physics.