A04

• Input: A procedure graph and a set of actors/tools/resources.
• Operation: Keep domain reasoning steps invariant while separately assigning performers, messages, concurrency, retries, and resource controls; connect layers through typed interfaces.
• Output: One reasoning model and one coordination/execution model.
• Scope conditions: The same reasoning can be performed by different human or automated arrangements.
• Failure conditions: Coordination changes alter decision semantics unnoticed; interface types are absent.
• Predicted effect: Reassigning execution will cause fewer reasoning regressions than a single mixed workflow.
• Measurement: Semantic regression count after reassignment, interface violations, and migration time.
• Implementation test: Swap actor assignments while holding reasoning tests fixed.
• Decision rights: Domain owner approves reasoning; operations owner assigns execution.
• Uncertainty handling: Unresolved actor capability is represented separately.
• Verification: Invariant reasoning test suite before and after reassignment.
• Provenance: Cross-layer interface links.
• Implementation cost: Maintaining two synchronized models.
• Limitations: Non-additive benefit is untested. Direct: separation aspiration. Inference: regression prediction.