This paper introduces Resolution-Based Information Theory (RBIT), a constraint framework for admissible information transformations in layered adaptive systems. In multi-agent systems where layers differ in processing capacity, compression is structurally unavoidable. RBIT reframes the design problem: the question is not how much information can be transmitted, but who controls what is lost during compression. The framework defines resolution as a layer's capacity to maintain distinction among competing signals, and the resolution gap (Δρ) as the central design variable governing compression regimes. Three core results are presented: (1) sustained transmission under negative resolution gaps necessarily converges toward intent replacement (Resolution Asymmetry Inevitability); (2) a contamination boundary theorem establishing when degradation becomes self-reinforcing; and (3) Self-Consistent Misalignment (SCM) — a failure mode where all local metrics appear healthy while global alignment has been lost. RBIT is positioned as a constraint theory specifying what transformations are admissible, not a predictive dynamics model. It provides formal foundations for the Deficit-Fractal Governance (DFG) architecture and is validated through a toy simulation (Latent Shift Bandit) demonstrating resolution-gap-dependent failure modes.
Bin Seol (Fri,) studied this question.