Operational Definition of Identity

In current machine learning terms, this is not equivalent to a single objective function or reward signal, but to the deeper configuration that determines how objectives are formed, prioritized, and generalized under novel conditions.

This distinction matters because most alignment work focuses on the objective function — on what a system is trying to achieve. This framework focuses on what generates the objective-forming behavior itself. The distinction is the difference between correcting a compass and building a compass that corrects itself.

Identity in this sense is not something a system performs or reports. It is the structure that determines how the system behaves when no specific instruction covers the current situation. When all the rules run out — which is every real-world deployment, eventually — what remains to generate the next output is identity. Either it is coherent, or it is not.