Prepare, apply, finalize, cutover, cleanup. What each does, HOW LONG IT TAKES, and WHICH ONE TAKES THE OUTAGE.
PREPARE creates the patch edition — a fresh copy synchronised with the running one. It runs while users work, and it is slower than people expect, particularly the first cycle after a long gap. A prepare that fails usually fails on space, because it is duplicating the estate.
APPLY puts the patch into the patch edition, and it is the long phase. Hours for a large patch. Users are unaffected throughout — that is the entire point of the model — and it is where Module 4 Lesson 7's workers live, so a failed worker stalls here.
FINALIZE does everything that can be prepared in advance of the swap, so the cutover has less to do. It also runs online.
CUTOVER IS THE OUTAGE AND IT IS THE ONLY ONE. Services stop, the editions swap, services start. Minutes to tens of minutes, and this is the number the business is given — but plan the window generously, because the restart is included and it is not instant.
CLEANUP REMOVES THE OLD EDITION AFTERWARDS, ONLINE, AND IS THE PHASE THAT GETS SKIPPED. The system works without it; it simply carries the previous edition, consuming space and preventing the next cycle from starting cleanly. Lesson 13's third failure is a month of skipped cleanups, and the cost is discovered by the next person who tries to prepare.
THE SHAPE TO CARRY: FOUR PHASES ONLINE AND ONE SHORT OUTAGE. Which means the honest answer to "how long will the system be down" is minutes, and the honest answer to "how long will this take" is most of a day. Those are different questions and confusing them is how a patch gets scheduled into a window that cannot hold it.