Modern software is highly concurrent, with many operations contending for shared information stored across large geographic distances. The systems on which these applications are written must provide a well-defined semantics that is easy to understand so that programmers can ensure their software is correct. Strictly serializable transactions provide a particularly simple interface for writing code in a concurrent setting, but they come at a cost: implementations require commit protocols to resolve contention between potentially conflicting transactions, sometimes coordinating across distant nodes. This dissertation explores warranties, time-dependent guarantees on the system's state, which improve the performance of distributed transaction systems by avoiding synchronous communication. Warranties guarantee that a predicate over the system's state and the current time holds until an associated expiration time. These predicates can express a wide variety of checks performed by applications ranging from simple comparisons to complex application-specific logic. Warranties can be constructed compositionally, using other warranties as evidence that a more complex predicate holds. Furthermore, these predicates can be time-varying, expressing guarantees about trends on the system's state. While holding an active warranty, nodes do not need to perform synchronous communication to validate the associated assertion. The system enforces an active warranty by delaying updates that falsify the guarantees until the warranty is safely retracted or expires. To ensure the benefits of warranties outweigh delays to updates, the system uses a cost model to determine a warranty's expiration time and to select other simpler warranties that help enforce a warranty with low overhead. Using a variety of benchmarks and real-world applications, warranties are shown to significantly improve the performance of distributed transaction systems.