When to export
Export the packet when the cycle needs to leave the review UI.What is in the JSON
The export is the machine-readable version of the audit packet section. It includes user-relevant structured data and excludes raw internal optimizer state.
This makes the packet useful for software, not only humans. An external reviewer can parse the same evidence path the UI shows, compare candidate selection and stage provenance, and decide whether the cycle is ready to approve, edit, rerun, or reject.
Use with external AI agents
An external AI agent can read the audit packet and return a structured recommendation for your workflow. Common checks include whether the trigger matches the intended issue, whether enough candidates were explored, whether regressions were filtered, whether the provenance is credible, and whether the cycle is ready for deployment through Adaline or an external release path. Exporting the packet does not approve, reject, edit, or deploy the cycle. It only gives another system the review context it needs to make or explain a decision.Before sharing externally
Audit packets can include prompt context, production-derived behavior labels, evaluator and dataset summaries, timeline details, and project metadata. Review your data-sharing policy before sending the JSON outside your environment. Remove or mask any information your external system should not receive, especially customer identifiers, private metadata, credential values, raw IDs, or sensitive production examples.Review a Cycle
Understand the review page that contains the audit packet.
Auto Prompt Optimization
See how Adaline explores and scores candidate prompt changes.