This deposit contains the full technical report and complete open-source reproducibility suite for Project Lacia — a sovereign cognitive architecture that achieves autonomous navigation of AGI benchmarks (high-entropy labyrinth traversal, combinatorial list-sorting, ARC-AGI abstraction, BEHAVIOR-1K long-horizon planning, and Shadow Hand stabilization) without Reinforcement Learning from Human Feedback (RLHF). Lacia replaces traditional reward-seeking with Sovereign Training, an internal homeostasis drive that minimizes spectral dissonance κx\kappa_xκx computed in O(1)O(1)O(1) time via a modular-residue graph over Z/24Z\mathbb{Z}/24\mathbb{Z}Z/24Z. The system is embodied in a self-healing 16×16 ultrasonic aerohaptic manifold (focused ultrasound for mid-air tactile feedback) and demonstrates: 53% average computational savings via the "Redundancy Cliff" Emergent escape from local minima in JEPA-style world models through non-Markovian internal tension accumulation Universal heuristic applicability across discrete and continuous tasks using the same spectral core Hardware-ready deployment on Jetson Orin with SPI serialization, diagnostics, and self-healing for damaged transducers The report details the full stack from theoretical UFT-F foundations to hardware integration. All Python scripts are organized by architectural layer and include virtual/hardware-agnostic abstractions for immediate reproduction on macOS/Linux. This work advances embodied AGI by grounding abstract reasoning in physical invariants and providing a verifiable alternative to RLHF-based alignment through internal geometric homeostasis. This work serves as the empirical validation for the theoretical foundations established in the UFT-F Spectral Framework series (Lynch, 2025). It is recommended to cite this report alongside the 'Redundancy Cliff' and 'Modular Fingerprint' datasets included in the related identifiers.