Agree the decision before collecting evidence
Migration acceptance is a decision about a named workload stage, not a celebration of a successful deployment command. Before cutover, write down which behavior must work, which systems remain outside the move and who is authorized to approve progression. This guidance proposes a review structure; it is not a guarantee of availability or a substitute for an organization's change controls.
Use a small acceptance record with explicit fields: workload boundary, candidate revision, representative checks, known gaps, data-state assumptions, stop conditions and receiving owner. An empty field is a visible question. Filling it with an unsupported assumption makes the record less useful, not more complete.
Bind each result to the candidate that was checked
Record the application revision and infrastructure configuration used for the review, together with the environment and relevant dependency versions. Avoid copying credentials or customer data into the record. A reference to an approved evidence location is often safer than duplicating raw logs.
If the candidate changes after testing, identify which conclusions remain valid and which checks must be repeated. A green result from a different environment or an earlier configuration should not silently become approval for the current cutover. The decision owner needs to see both the result and the scope it actually supports.
Make an acceptance record that can be rechecked
Separate the candidate record, evidence rows and approval decision. The candidate record identifies the application revision, infrastructure plan and intended environment. Each evidence row needs a check ID, expected behavior, observation time, result, evidence reference and the candidate identity it actually tested. The decision record names the reviewer, decision time and accepted limitations. These records answer different questions; a passed evidence row is not itself permission to cut over.
Use explicit states such as passed, failed, blocked and not run. A blank result should not silently mean success. When a dependency or configuration changes, mark the affected evidence stale and identify the rows that need repeating. Keep immutable references to the earlier observations rather than editing their timestamps. This small acceptance manifest makes the approval inspectable without copying sensitive logs into the review.
Test the dependency boundary, not only the homepage
Choose representative paths through authentication, data access, scheduled work and external interfaces that matter to the selected workload. Identify who owns dependencies that will stay in the original environment. A synthetic request can establish a narrow fact without demonstrating every production condition.
For each check, record the expected behavior, observed result and limitation. If a third-party sandbox cannot reproduce production behavior, carry that limitation into the decision. A missing check is not a pass. Agree who can accept the remaining uncertainty and when a failed check stops the move.
Keep data-state limits separate from infrastructure rollback
Restoring an infrastructure definition does not necessarily undo records, messages or actions created after cutover. State which system accepts writes during coexistence and how reconciliation would work if the team pauses or reverses direction. Name the person authorized to choose among rollback, restoration and forward repair.
Do not label a migration reversible merely because the old environment still exists. The evidence should explain the point at which data divergence, an external side effect or an unavailable dependency changes the recovery options. An untested option should remain explicitly untested.
Close with a bounded proceed-or-pause decision
The reviewer should be able to explain why this candidate can proceed, what remains unresolved and who will operate it afterward. Record the decision time and the evidence version. Keep continuing support duties separate from project acceptance; a handoff does not automatically establish incident-response coverage.
Bring the acceptance questions into scope discussions before implementation. Architecture advice, cloud foundations and a workload migration are different possible engagements. The useful starting point is the decision that needs evidence, not a presumption that every platform must move.
compare cloud migration and broader Cloud delivery scopes in Canada
