Golden Dome Latency Governance: Autonomous Operations Model Extended to Boost-Phase Intercept Timelines

Motivation and Context

On January 27, 2025, an executive order initiated the "Golden Dome for America" program: a persistent, multi-layer missile defense architecture intended to protect the continental United States from ballistic, hypersonic, and cruise missile threats. Unlike legacy regional systems (THAAD, Patriot, Aegis), Golden Dome relies on a massive LEO constellation of infrared sensors and kinetic interceptors designed to engage threats during their most vulnerable flight phases. Cost estimates range from $175 billion (White House) to $831 billion (CBO), with some lifecycle projections reaching $3.6 trillion. To manage the resulting complexity, the Missile Defense Agency has approved over 2,440 vendors through the SHIELD contract. Prominent among them is Anduril Industries, whose LatticeOS platform is positioned as the software integration layer fusing sensor data and orchestrating autonomous command and control across the constellation.

The Anduril LatticeOS announcement made the governance problem concrete. LatticeOS is designed to assign decisions to tiers by cognitive budget, but that model was built for terrestrial and airborne systems where ground-station contact is routine. Extending it to a LEO intercept constellation exposes a hard physical constraint: GEO relay paths add 474 ms of round-trip propagation delay, and any given sensor node has direct downlink availability of only roughly 8% of its orbit. For HGV terminal intercept, where the engagement window is 5 to 15 seconds, those two facts together close human-in-the-loop authorization as a reliable governance mechanism.

The companion paper (Bilar 2026, Anduril LatticeOS: Autonomous Operations Model, DOI: 10.5281/zenodo.19266807) established the baseline AOM tier structure. This paper extends it to Golden Dome's boost-phase and terminal-phase timelines and specifies what governance at the edge must actually look like when the speed of light is the binding constraint.

This position paper and its accompanying simulation suite (DOI: 10.5281/zenodo.19368609) are intended for defense technologists making architecture decisions now, before Phase GD-0 hardware characterization has occurred. The 7-check gating matrix, the pre-authorized Mission Intent Object envelope, and the Hub-Tier Model Swap Protocol are proposed engineering defaults, not validated specifications. The falsification results quantify what the architecture can and cannot guarantee: 99.0% combined detection against a nation-state adversary, dropping to 89.2% against one with sensor calibration access, and to 61.7% if the PINN hardware entropy assumption fails. Those numbers are only as reliable as the per-check parametric models underneath them. The paper's goal is to make the residual risks explicit and the governance dials visible before the system is built, not after.