# Progressive Autonomy Architecture > A vendor-neutral architecture for moving agentic tasks from human review to monitored autonomy through a task contract of five orientation groups: boundary, evidence log, evaluator (owning oracle and position policy), promotion rule, and demotion (owning fallback implementation). ## Start here - [Framework](/framework): The canonical task model — five orientation groups (boundary, evidence log, evaluator, promotion rule, demotion) with eight commitments; oracle and position policy are evaluator-owned, fallback implementation is demotion-owned — and the illustrative task declaration rendering. - [Manifesto](/manifesto): The founding argument for earned, scoped, and revocable autonomy, and why the evaluator is the primitive. - [Evaluators](/evaluators): The evaluator contract (target, technique, oracle, position policy), evaluator-owned commitments explained, the technique catalog, and the maturity curve. - [Flows](/flows): Five reference flows, each a complete instance of the task model. - [Grounding](/grounding): The prior-art basis (SAE J3016, Osprey, Safe-SDL, Malik) and what PAA adds. - [Contact](/contact): Discuss a workflow, review an evaluator, or contribute a pattern. ## Reference flows - [PR review agent](/flows/pr-review): Cheap deterministic checks can stack under an expensive semantic judge, and the artifact boundary itself is the thing being reviewed. - [Refund approval agent](/flows/refund-approval): Worker and evaluator are different roles, and selective autonomy means the same task can be autonomous for one slice and gated for another. - [Document routing](/flows/document-routing): The live evaluator and the promotion evidence are different things, and both are required if the automation is going to stay safe. - [Transaction / trade execution gate](/flows/transaction-gate): Risk-adjusted failure cost sets the bar, and irreversible actions stay under deterministic gating rather than learned judgment. - [Support response agent](/flows/support-response): Eval-the-eval is the point here, because the evaluator is validated, promoted, demoted, and eventually made cheap enough to own the gate. ## Full reference - [llms-full.txt](/llms-full.txt): A compressed, self-contained version of the whole framework for agents that cannot fetch every page.