You learn very little about a storage system while it is running normally. Normal operation is what it was designed for and what it was tested against. The interesting properties surface during a migration, because a migration is the one time two systems must simultaneously hold a coherent opinion about the same data.
The questions a cutover forces
Who is authoritative right now, and how does every participant find out? "Both, briefly" is the answer that causes incidents.
What happens to a write that arrives during the transition? Not the happy path — the one that arrives after the source stopped accepting and before the target started.
How do you roll back after the first byte has been written to the target? If the answer is discovered during the rollback, it is not a plan.
How do you prove afterwards that nothing was lost? Not "the tool reported success" — an independent check, ideally one that would fail loudly if the tool lied.
Why this is worth practising deliberately
Every one of those questions is a distributed-systems question with a storage accent. Teams that have done a real cutover reason about consistency differently afterwards, because they have been forced to say precisely what they mean by it.
Rehearse the rollback, not just the migration. The migration gets rehearsed because it is the plan. The rollback is what you will actually be doing at 3am.