Choose a reviewable migration boundary
A migration rehearsal should answer whether a defined workload stage is ready to proceed. Starting infrastructure is only one part of that question. Applications, data flows, external interfaces and operating responsibilities may cross the boundary that appears convenient in an infrastructure plan.
Map the dependencies and identify which components must move together. If only part of the workload will move, describe how the remaining systems connect during coexistence. Treat the temporary architecture as something that needs owners and verification, not as an invisible interval between two steady states.
Use representative integrations and data behavior
Select checks that exercise the interfaces relevant to the stage. Record which inputs and assumptions the rehearsal uses, what it cannot reproduce and how those limits affect the decision. A synthetic request can be useful, but it should not be described as proof of every production condition.
Identify data owners and write behavior before cutover. A change can create records, messages or external effects that remain after infrastructure is restored. The rehearsal must make those effects visible enough to discuss recovery rather than treating them as a configuration detail.
Separate reversal from recovery
A configuration rollback returns a selected configuration to an earlier version. It does not necessarily undo data writes or actions taken by other systems. Data restoration, reconciliation and a forward repair are different options with different constraints.
For each option, identify the decision owner, required evidence and limits. If a recovery approach is untested or unavailable, record that condition before approving the move. Do not label a migration reversible simply because an earlier infrastructure definition is retained.
Write the proceed-or-pause contract
Agree acceptance criteria before running the rehearsal. Define which failed checks stop progression and who may approve the next stage. Include the change-window constraints and the continuing responsibilities of teams operating unmoved systems.