Decide which layer owns each statement
A stable rule governs repeated work. Examples include using confirmed details only, requiring independent review for material claims, and preserving a correction history.
Job content belongs to one assignment. It includes the selected notes, audience, offer, first attempt, finding, and corrected artifact.
Ask one question for every statement: should this remain true when the subject changes? If yes, it may be a stable rule. If no, keep it with the job.
Map separation to context selection
This decision belongs to “The right context for this job.” The component assembles the reusable instructions and job-specific sources needed for one task without treating every stored file as current input.
In an AI harness, that separation keeps owned files, task definition, independent review, and run history connected without mixing their roles. CommonMark provides a readable format for both layers. W3C PROV-O helps describe relationships among their artifacts. The NIST AI Risk Management Framework supports documented governance around AI use.
The sources support the structure, not the fictional workshop facts.
Use separation where rules recur
This method fits jobs that share an evidence standard, review method, or artifact pattern. It adds ceremony to a disposable task with no reusable instruction.
Stable does not mean permanent. Change a rule deliberately when the operating decision changes, and avoid rewriting it merely because one job has unusual content.
Position the rule boundary
Begin with Owned AI Workbench for the category decision. Assign Responsibility in an AI Workbench names who maintains each consequential choice. Choose the First Job for an AI Workbench supplies a bounded task for testing the separation.
Separate the lasting rule from the assignment
In the synthetic job, confirmed details only is a reusable rule; the workshop audience and offer remain job-specific content.