ULMA: A System Overview and Evidence Boundary for Distributed Agent Memory

ULMA is a distributed memory system for long-horizon agent workloads. This technical note records the current public system boundary: implemented benchmark control paths, monitor-based evidence outputs, and the claim-to-evidence mapping used by the project. It serves as a citable overview record rather than the final manuscript of any single conference paper, and it supports several focused research views, including distributed consistency and recovery, runtime memory abstraction, and benchmark methodology. Researchers interested in collaboration or discussion of the system may contact the author through the public contact information associated with this record.