# AEG Architecture Evaluation Checklist

Version 1.0 · August 28, 2026  
Source: https://jacobpallattu.com/architecture/aeg/kit  
License: CC BY 4.0 — adapt with attribution to Jacob George and the AEG Architecture Model.

This checklist is a design-review aid, not a certification. Every checked item should point to inspectable architecture evidence.

## Intent boundary

- [ ] Natural-language requests are normalized into a typed intent.
- [ ] The intent identifies target, scope, constraints, requester, and expiry.
- [ ] Executors cannot act directly from model output or conversational text.

## Independent governance

- [ ] Policy evaluation is outside the component that generated the intent.
- [ ] Allow, deny, and approval-required are explicit decision outcomes.
- [ ] Approval is bound to the exact intent and cannot authorize a broader action.

## Controlled execution

- [ ] Execution adapters accept only authorized, unexpired intents.
- [ ] High-impact operations have limits, rollback conditions, or compensating actions.
- [ ] Replays and retries cannot silently expand the original authority.

## Observable evidence

- [ ] Decision and execution events share a durable correlation identifier.
- [ ] The final outcome can be reconstructed without relying on model reasoning.
- [ ] Denied, expired, failed, and rolled-back outcomes are observable.

## Evidence notes

For every unconfirmed item, record the missing artifact, owner, and next design decision.

| Check | Evidence or gap | Owner | Next decision |
|---|---|---|---|
| | | | |
| | | | |
| | | | |