This paper studies how to write contracts when the party offering the contract cannot fully commit to future actions. The authors find small, standard sets of contracts that are enough to describe every possible equilibrium outcome in a wide class of models. Their result gives a way to analyze contracts even when the agent’s private information is continuous and preferences are general.

The researchers work within a classic principal–agent setup but remove strong commitment assumptions used in much prior work. They identify two minimal “canonical” contract spaces. For public contracting, where all contracts and messages are openly known, they show there is a contract space called G** that is minimal and fully captures every equilibrium that could arise from any complicated contract. For private contracting, where contracts or messages are hidden, G** is not enough; they identify a different minimal space G*** that plays the same role.

Technically, the paper extends earlier canonical-mechanism ideas in three directions. It allows agent types to be infinite or continuous (unlike Bester and Strausz 2001). It drops quasilinear utility assumptions used by some later work (unlike Skreta 2006). And it avoids adding extra commitment devices such as information design that rely on the principal being able to commit to how messages are processed (unlike Doval and Skreta 2022). The authors describe a “taxation-principle” style approach that narrows the set of contracts needed for analysis, making equilibrium problems tractable.

Why this matters: many real contracts involve limited commitment in practice—governments, long-term vendors, and partners routinely retain some discretion over future actions. Previous theory either could not handle continuous private information or depended on payment-linear preferences or extra commitment devices. By giving minimal contract spaces that work under general preferences and continuous types, the paper supplies a unified tool for studying a wide set of economic contracting problems. The authors also show how their canonical spaces can be used to solve motivating examples and fully characterize equilibrium outcomes in those cases.