This technical position paper extends the Autonomous Operations Model (AOM) to Golden Dome boost-phase and terminal-phase hypersonic glide vehicle (HGV) intercept timelines. It makes three contributions. First, it quantifies the combined propagation-delay and ground-station-availability argument that closes human-in-the-loop (HITL) control for HGV terminal intercept: GEO relay adds 474 ms round-trip, and direct downlink is available only ~8% of the orbit. Second, it specifies a dual-axis latency-aware tier matrix and a unified 7-check Gating Check Matrix executable in under 5 ms on rad-tolerant edge hardware, combining operational checks (schema, MIO hash, physics envelope, staleness, geo-fence) with an adversarial-robustness layer (cross-modal Mahalanobis consistency, temporal coherence filtering, chaotic-quantized PINN seeded by a physically uncloneable hardware entropy source). Third, it specifies a Hub-Tier Model Swap Protocol for surviving political or legal severance of any single foundation-model provider. Under the parametric per-check models, combined detection is approximately 99% against a nation-state adversary (s=0.74) and approximately 89% against a nation-state with calibration access (s=1.0). With the PINN zeroed out, detection drops to approximately 62% at s=1.0. All quantitative thresholds are proposed engineering defaults requiring Phase GD-0 hardware characterization before operational use. Version 8.1 (August 2026) adds a program-office confirmation of the automation dial, records that the ISL mesh is not yet on orbit, adds a quantization-backdoor gate to the model-swap runbook, and strips false-precision confidence intervals from the parametric results. Architecture, thresholds, and simulation results are unchanged.
hypersonic glide vehicle · Golden Dome · Anduril · LatticeOS · latency governance · human-in-the-loop · kill chain · physics-informed neural network · edge autonomy · Monte Carlo · quantization · adversarial machine learning · SHIELD · PINN