Nobody decides to keep a bad assumption. It keeps itself. The rate line in the plan was reasonable the week it was written, the person who wrote it moved on to the next model, and from then on it is defended by the simple fact that nobody was assigned to question it. Six months later it is wrong by enough to matter and the first anyone hears of it is a variance in the April review. Assumptions do not fail. They expire, and the only question is whether the expiry date was written down in advance or discovered after the damage.
Why assumptions rot
Three reasons, and they compound. The first is that an assumption is usually written by someone who is not the person who will suffer when it breaks. The treasury analyst writes the rate line; the regional president eats the margin. The second is that the moment of writing is the only moment anyone looks at it. Plans get built once and tracked many times, and tracking looks at outputs, not inputs. The third is social. Re-opening an assumption after the plan is approved looks like second-guessing the plan, and nobody wants to be the person who asks whether the number the board signed off on is still true.
This cycle has a clean example. The Fed raised its target ceiling to 4.00% on September 17, and by November the market had a December hike priced near 97%. Most 2027 plans were drafted in early September. Their rate line was reasonable on the day it was written and is going to the board stale, not because anyone was careless, but because nobody owned the date.
Register them with a sunset
The fix is a single register, maintained by finance or by the Predictionist if the company has one, with one row per load-bearing assumption. Five to ten rows for most plans. Each row carries six things: the assumption as a sentence with a number in it, the date it was written, the public or internal signal that would tell you it broke, the value of that signal on the day of writing, the owner who brings it to the review, and the sunset date.
The sunset date is the date on which the assumption is treated as expired unless someone re-argues it, in front of the people who approved the plan, with the signal in hand. Not revisited if convenient. Expired. A rate assumption gets sixty days in a tightening cycle. A demand growth rate gets a quarter. A long-run input cost assumption might get six months. The horizon depends on how fast the signal moves, and the notes on each series in the Signal Library are a reasonable guide to how fast that is.
When the date arrives, the owner either defends the assumption with the current signal value and gets a new sunset, or the assumption gets rewritten and every commitment that leaned on it goes back on the table. There is no third option where it quietly rolls forward.
Who reviews, and where
Expired assumptions get argued at the monthly Convergence Review. That is the meeting where every active scenario is sorted into converging, holding, or collapsing, and it is the right place because an assumption that breaks almost always moves a scenario. The owner brings the row and the current signal. The team decides whether the assumption holds, drifts, or is broken, and for anything broken, the Stake Sizing Matrix gets re-run on the commitments built on it. Ninety minutes, once a month, and the register is the first item on the agenda.
Two rules keep it honest. The owner of an assumption is never the person who wrote it, because the author has a stake in being right. And the register is visible to the board, so re-opening a line is not second-guessing the plan, it is the plan working as designed.
The biggest assumption of all
Every annual plan carries one assumption that never gets a row in anyone’s register: that the annual planning cycle should still exist. It consumes something like ninety days of senior attention a year, the largest line in the consulting budget, and the company’s primary forward-looking conversation with its board, and nobody has ever been asked to defend that on a date. The 18-Month Sunsetis the book’s name for putting the expiry date on the cycle itself.
It is a sequenced plan, in four phases of roughly equal length, for retiring the annual marathon while Woodring’s Loopstands up beside it. Months one through three, the Loop runs in shadow mode while the signal layer builds. Months four through nine, dual-run: the Loop’s outputs start driving Stake decisions, the planning marathon compresses, and the board sees its first scenario brief. Months ten through fifteen, the marathon is formally retired, audit controls are mapped into the Loop, and compensation is restructured. Months sixteen through eighteen, full operation and a retrospective. Each phase has an owner and a date, which is the whole point. A transition without a sunset date is a permanent shadow system, and a future CEO will quietly dismantle it.
The Sunset and the assumption register are the same idea at two scales. Nothing in the plan, and not the plan itself, gets to live without a date on which someone has to argue for it again. The Cycle Audit is the step before the Sunset, for the same reason the signal value on the day of writing goes in the register: you cannot retire what you have not measured.
What goes wrong
The register fails when the sunset dates are set too far out to matter, when the owner is the author, or when expired rows are allowed to roll forward because the review ran out of time. The Sunset fails the same ways, larger. The company keeps both systems indefinitely and never pulls the trigger. The executive sponsor changes mid-transition and the new one inherits a project instead of a commitment. Finance treats the Loop as a threat to the function the old cycle defined. Or the board re-imposes the annual ritual under a new name in month fourteen. The defense in every case is the same: name the date up front, name the owner, and make the commitment outlast any single person.
What to do Monday
Build the register for the plan that is about to be approved. Five rows minimum. For each one, write the sentence, the date, the signal, today’s value from the Signal Library, the owner, and the sunset. The budget stress test walks the first pass in one afternoon. Put the first review on the calendar for December. Then, once the register has run for one quarter and the team has felt what it is like to have assumptions with dates on them, open the conversation about the date on the cycle itself. The full Sunset, with the phase milestones and the four things the old cycle keeps in mutated form, is in Chapter 12 of the book.