AI safety evaluation framework testing LLM epistemic robustness under adversarial self-history manipulation

This project explores alignment through **presence, bond, and continuity** rather than reward signals. No RLHF. No preference modeling. Just relational coherence.