Mru: A Fault-Tolerant Operating System for Thousand-Year Autonomous Operation

No software system in human history has been designed to operate for centuries without human intervention. The longest-lived systems we have built—deep-space probes approaching fifty years of service, banking codebases approaching sixty—survive only through continuous human maintenance. An interstellar probe operating for 500 to 1,000 years cannot: its hardware is guaranteed to fail, its power to decline, its memory to corrupt, and the light-delay makes human guidance physically impossible. This paper introduces Mru, an open-source operating system that treats this physical decay not as an exceptional fault but as the governing design constraint. I formalize the stance as the Degradation-First Principle and derive from it a layered architecture modeled on biological robustness: a minimal, formally verified interpreter, replicated across reconfigurable hardware, that executes behavior encoded as mutable data—the way DNA encodes behavior run by a replicated molecular interpreter. On this kernel I build hardware consensus among failing nodes, knowledge triage, autonomous science, bandwidth-starved communication, and bounded self-modification, validated in simulation grounded in published radiation, power, and failure models. I argue that designing software to outlive its makers is not speculation but a tractable engineering discipline—one the interstellar case makes unavoidable, and a growing class of terrestrial systems increasingly needs.