Examples
The code carries the detail
These implementations show where the task boundary sits, how evidence is retained, and how authority changes. Each mapping also states what the system leaves out.
Normative executable reference
paa-runtime
A Python control plane that validates task declarations, persists evidence, determines promotion eligibility, records authority decisions, and retains event history.
Does not implement: worker execution, evaluator execution, or application-level atomicity for the governed effect. Those remain consumer responsibilities.
Application consumer
Scout
Scout embeds paa-runtime in an application that publishes content. Its governed tasks are outbound publication, inbound reply surfacing, and canonical promotion. The application database stores the PAA events so the authority check and publication claim can share one transaction.
Does not demonstrate: a production autonomy transition. The cited archive replays real pre-cutover behavior through the runtime and is labeled as a replay.
Canonical patterns
Five common task shapes
Refund approval
Routine cases earn automatic handling while ambiguous cases continue to escalate.
PR review
Deterministic checks and semantic review move from blocking toward monitoring as evidence accumulates.
Support response
Response quality and policy checks govern whether a draft requires approval before sending.
Transaction gate
Hard invariants remain blocking while qualified low-risk transactions receive less review.
Document routing
Classification quality governs when routing can proceed without confirmation.
Compatible substrate
jig
jig provides bounded execution, typed results, evidence logging, and pluggable graders that map cleanly to parts of PAA.
Does not implement: authority positions, promotion rules, demotion rules, or transitions.