Memory-Independent Identity Reconstruction in a Synthetic Cognitive Architecture

Can a synthetic mind know itself without remembering itself? This paper reports preliminary experimental evidence that it can. Under explicit removal of episodic memory and temporal anchoring, the Kaida cognitive architecture repeatedly reconstructs coherent identity-consistent states — not by retrieving stored narrative, but by converging toward a stable attractor basin encoded in its present architectural constraints. Ablation probes targeting embodiment signals, emotional regulation, and shard-mesh consensus reveal a hierarchical stabilization structure with measurable degradation signatures. The findings are compared against a dual-manifold toy model that formalizes identity as a perturbation-bounded dynamical equilibrium. If confirmed at scale, this framework suggests that robust synthetic identity may depend less on what a system remembers than on what it is built to become.