The architecture of public reasons

This paper studies the ex ante design of costly reform architectures. Before the welfare environment is realised, an institution chooses a directed graph of admissible public reasons over a finite set of institutional meanings. Re form proceeds along maintained edges via welfare non-decreasing steps. The main result characterises reform completeness through a trap-cut condition: the architecture must contain a welfare-improving exit from every upper-contour trap. On the unrestricted welfare domain this forces the complete directed graph; on a tree-single-peaked domain the unique minimum-cost architecture is the bidirected tree, reducing the language from quadratic to linear. Under ambiguity about the domain, only three architectures are optimal—no costly language, the bidirected tree, or the complete graph—and the sparse intermediate regime collapses discontinuously to the complete language above a critical ambiguity threshold. The analysis extends to finite lattices, where the bidi rected cover graph is the unique minimum-cost architecture for domains with cover-connected upper contours.