The runbook is paper. The rehearsal is the work.

Every failed cutover we have been asked to autopsy had a runbook. Good ones, some of them. Numbered steps, named owners, rollback section, the works. Paper does not fail. Execution fails, and execution had never been tried.

Rehearse it. All the way through, against a production-shaped copy, with the people who will run it live, at the hour they will run it. Yes, including the hour. A step that takes fifteen minutes at 2 p.m. takes forty at 2 a.m., and the difference is your window.

What a rehearsal buys, in the order the findings arrive:

1. The timings. Every runbook we have rehearsed was optimistic by 40 to 100 percent on total duration. Every one. The plan gets rewritten against measured numbers, and suddenly the window conversation with the business is honest.
2. The missing step. There is always one: the cache that needed warming, the license server that noticed the new hostname, the confirmation email that goes to a mailbox nobody owns. Finding it on a Tuesday costs nothing.
3. The rollback truth. Rehearse the abort path too, from the worst checkpoint. If rollback takes longer than the remaining window, your go/no-go times just moved, and better to know while it is still ink.
4. The handoff gaps. Watch what happens when step 14's owner finishes and step 15's owner starts. The silence between them is where 2 a.m. incidents live.

Cost: two days of the team, once, plus the environment copy. Set against the going rate for a blown cutover weekend and the retail price of the second attempt (reputation not included), it is the cheapest insurance in this business.

One rule for the rehearsal itself: no fixing. Log every snag, keep walking. The rehearsal that stops to repair things tests your repairs, not your plan. Fix Wednesday, rehearse again Thursday if the snag list was ugly.

Then tape the measured timeline to the wall next to the runbook. Paper plus proof. That pair has never let us down.