This is not a measure of how many people use AI. It tracks whether AI is moving from advice and drafting into consequential action with less transaction-by-transaction human approval.
Delegation is the bridge between capability and real-world consequences. A highly capable model has a different social impact if a person must approve every important action than if it can spend, deploy, negotiate, modify systems, or operate for long periods on its own.
AI mostly proposes, drafts, or recommends; humans execute consequential actions.
AI executes bounded workflows inside meaningful permission and approval constraints.
AI routinely carries out consequential, multi-step work with limited contemporaneous human approval.
The state changes when admitted evidence moves the underlying indicators across the frozen scoring rules. The model should not move because a story is prominent in the news; it moves when the measured condition changes.