Progressive Autonomy Architecture
Task-specific autonomy earned through evidence
PAA describes how automation can earn task-specific autonomy from evidence, with explicit oversight, promotion, and demotion.
Progressive Autonomy Architecture (PAA) describes how an automated system can earn more responsibility for a defined task. Evidence supports changes in autonomy, oversight remains explicit, and degraded performance can restore tighter control.
Earned and revocable responsibility
A worker performs a declared task, an external evaluator judges the result, and the implementation records behavioral and economic evidence. That evidence can justify changes to the operating configuration or eligibility for different oversight. Changing authority is a separate governance decision, granted through a declared approval or automatic rule.
Autonomy is task-specific and revocable. A better score alone does not justify broader authority or higher operating cost. Evaluation continues after an oversight change, and confirmed regression can restore the declared safer operating mode.
Architecture and implementation
The PAA website is the canonical architecture documentation. The runtime repository implements task declarations, the motion lifecycle, evidence, and an append-only autonomy event store; it is one implementation of the architecture’s contracts.
Scout is an implementation environment for examining these mechanisms. Its checked-in tasks are shadow or disabled, neither actively enforces runtime behavior, and none is deployed autonomously. Reference evidence demonstrates control-plane mechanics rather than earned production autonomy.