graduated-autonomy
The graduated-autonomy primitive for production classification. Rules graduate to LLM, then to in-process ML, through statistically-gated phase transitions. The rule stays as the safety floor
