Separate three incomplete outcomes
An unfinished stage can lead to different responses:
- Wait for a named piece of information.
- Revise the artifact using the evidence already available.
- End the job because its objective cannot be met within scope.
These outcomes are not interchangeable. The workflow should say who chooses among them and what record each choice leaves.
Preserve the incomplete state
The synthetic exception preserves the first brief and identifies missing schedule evidence before a person chooses correction or further research.
Preserve state before choosing the alternate path
Record the current artifact, completed stages, missing information, and allowed next choices. This gives the accountable person a bounded decision instead of an open-ended recovery problem.
The NIST AI Risk Management Framework provides a structure for managing AI risk across a system's use. W3C PROV-O can describe how an artifact was generated and revised. RFC 8493 defines BagIt packaging for groups of files when preserved materials must travel together.
Keep the exception at workflow level
This page decides what the wider process does when a stage cannot finish. It does not diagnose a missing retrieval result or a failed execution method. Those component-level questions belong elsewhere.
The approach fits recurring work where incomplete stages can be anticipated. It offers less value for disposable exploration that creates no downstream dependency.
Rejoin the workflow deliberately
Use Map an AI Workflow for One Recurring Job to place alternate paths in the full sequence. Review AI Workflow Approvals when resumption needs an explicit decision. Continue to AI Workflow Metrics to observe how often incomplete stages create rework.