A regulator asks for evidence that your exit plan from your primary cloud provider is credible. The plan is a twelve-page document. What will they find, and what should you do?

What they will find

A document describing an intention. Supervisors have moved from accepting exit plans to asking what has been tested, precisely because most plans describe a migration nobody has attempted and whose duration nobody has measured.

Expect three specific challenges: has any part of this been executed; what does it cost, including egress; and what provider-specific capabilities does the architecture depend on that the plan does not mention.

The dependency inventory is the substance

Go through the estate and classify each component by portability:

• Portable — containerised workloads on open runtimes, open table formats, open protocols. Movable with effort.
• Substitutable — a managed database or queue with an equivalent elsewhere, at the cost of a migration and some behavioural differences.
• Provider-specific — proprietary serverless runtimes, managed identity, bespoke analytics services, and anything whose semantics have no direct equivalent. These require redesign rather than migration.

The third category is what determines the timeline, and it is what a twelve-page plan usually omits.

Test something

Even a narrow test changes the conversation entirely. Restore one dataset into a second provider and run a real workload against it. Time it. Record what broke.

That produces a measured number, a list of gaps and evidence of capability — which is what "credible" means in this context. A full rehearsal is better and is rarely proportionate; a partial one is far better than none.

Cost it honestly

Egress charges are frequently the largest single line and are routinely omitted. Include data transfer, dual running during migration, engineering effort, and the retraining and tooling cost of operating elsewhere.

Say what you have decided

The strongest position is not that everything is portable. It is a conscious, recorded decision per component about how much lock-in has been accepted and why — because full portability forgoes the managed services that make cloud worth using, and a supervisor understands a deliberate trade better than an unexamined one.

The weak position is lock-in accumulated by default with a document asserting otherwise, which is what the question is testing for.