The month it stops being your estimate.

Two lives, one number

The two things a forecast can be look identical on a page and do opposite jobs.

Which is worth separating before going further, because the rest of the week depends on the distinction and it isn't one anybody states out loud.

A prediction is made to inform a decision. Its value is accuracy, its natural behaviour is to move whenever information arrives, and a prediction that never changes is failing.

A commitment is made to be relied on. Its value is stability, other people plan against it, and one that moves imposes cost on everybody who used it.

Both are legitimate and a project needs both. The trouble is that the same figure, in the same cell of the same report, is asked to be both at once.

What happens to a number once it is a target

Follow the incentive and the mechanism is plain, and it has nothing to do with anybody's character.

A forecast that moves reads as evidence that the forecasting was poor. A forecast that holds reads as evidence that the project is under control. Neither reading is correct — a number that tracks reality is doing its job whether it moves or not — but both are how the movement gets interpreted.

So the pressure on the person producing it runs towards keeping it still. Not by falsifying anything: by resolving genuine ambiguity in the direction that requires no explanation, by waiting one more month before accepting that a trend is real, by treating a recovery plan as though it will work because it might.

Each of those is defensible on its own, and each would be the right call in isolation. Together they produce a number stiller than the thing it describes.

Nobody in this has decided to mislead anybody. That is what makes it hard to address: there is no point in the sequence at which somebody could be told to stop.

The sawtooth

The consequence has a shape, and it is the shape a series of monthly forecasts takes when this pressure is acting on it.

The number holds, holds, holds, and then jumps. It jumps by more than any single month's news could justify, because what moved wasn't that month's reality but the accumulated divergence that could no longer be absorbed.

And the month it jumps in is chosen by capacity rather than by cause. It moves when the gap has grown past what a recovery plan can plausibly cover, which depends on how much room the plan had, not on anything that happened that month.

And the jump is read as a failure of forecasting, which produces more pressure towards stillness, which lengthens the next flat period and enlarges the next jump. The mechanism reinforces itself and requires nobody to behave badly at any point.

What follows the jump commercially — the point at which a forecast crossing the price forces a loss to be recognised in full — is Cost & Cash Week 23, and it is the sharpest treatment of that on this site.

Neither direction is free

It would be tidier if the resistance were only to bad news. It isn't.

Revising upward requires explaining what went wrong and exposes the person revising. Revising downward invites the question of why the number was wrong before, and can look like a figure being managed. So both movements cost the forecaster something, and standing still costs them nothing at all.

Which explains something that otherwise looks like inconsistency: the same person who resists reporting an overrun will also resist reporting that a recovery worked. The resistance isn't to the direction. It's to the movement.

That is the same shape as the approval asymmetry in week 24: acting is recorded against a name and not acting is recorded nowhere. When the same structure turns up twice in one phase it is worth treating as a property of governance rather than a coincidence.

When the transition happens

Here is the useful part, because the moment isn't mysterious and can be anticipated.

The forecast becomes a target the first time it leaves the team and gets used by somebody else — the month it appears in a report going to a board, a client, or a parent company. Before that it's a working figure. After that, somebody outside has planned against it.

And “planned against it” is literal. A parent company has set an expectation with its own board. A client has told somebody when the asset will be available. Those commitments are made of your number, and moving it now moves theirs.

Which means the transition has a date, and the date is on your reporting calendar. The last unencumbered forecast is the one before it, and it is the last chance to move the number without the movement itself being the story.

Knowing that in advance is worth more than any technique. A forecast can be revised in month four at the cost of a sentence, and in month fourteen at the cost of a meeting with people who have made commitments against it.

So the thing to do with the last unencumbered month is spend it. Put the worst defensible number in, while putting it in is still free.

System design

Row one asks for a series rather than a value, and it is the cheapest thing in this table because the months already happened. Overwriting last month’s forecast with this month’s destroys the only evidence of whether the number tracks reality or trails it.

Record	Produced by	Required quality	Verified against	Feeds
Forecast as issued	You, every month	Kept as a series, not overwritten by the current view	What was known that month	Whether the number tracks or lags
Date first believed	You, at the time	The month you thought it had moved, recorded then	What you did about it that month	The measurement of the gap
Date reported	The reporting cycle	Already exists; the pair is what matters	The report itself	Whether the gap is widening
Cause of each movement	You, in a sentence	What changed in the world, not what changed in the model	The event it names	Whether the same cause recurs
Assumptions carried	The forecast, restated monthly	Restated each month rather than referenced, since they expire quietly	Whether each still holds	What the number rests on

Row five is the quiet one. A forecast rests on assumptions that were true when it was built, and they expire without announcing it. Restating them monthly costs four lines and is the difference between a forecast and a habit.

Practical insight

Keep two dates against every movement in your forecast: when you first believed it had moved, and when you reported it.

That is the same instrument week 23 asked for on a change, pointed at a different object, and it works for the same reason. One date you already have. The other is the one nobody records, and the gap between them is the only measurement of the pressure this week describes.

It isn't a record of concealment and you should not treat it as one. The gap exists for good reasons — a trend takes weeks to become distinguishable from noise, and reporting every wobble is its own failure. What the two dates give you is the ability to tell the difference between a delay that was analysis and a delay that was reluctance, and only you can see that difference at the time.

Look at your own gaps after six months. If they are getting longer, the number is becoming a target and you are the last person who will notice from the outside.

Key takeaways

• A forecast has two lives — a prediction made to inform a decision, and a commitment made to be relied on.
• A prediction should move whenever information arrives; a commitment imposes cost on everybody who planned against it when it moves.
• Both are legitimate, and the trouble comes from asking one figure in one cell to be both.
• Once it is a target, movement reads as poor forecasting and stillness reads as control, and neither reading is correct.
• So the pressure runs towards stillness, achieved by resolving genuine ambiguity in the direction requiring no explanation.
• Each of those resolutions is defensible alone, and together they produce a number stiller than what it describes.
• The result is the sawtooth: long flat periods and jumps larger than any month’s news could justify.
• The jump reads as a forecasting failure, which increases the pressure towards stillness, which enlarges the next jump.
• Revising upward exposes the forecaster and revising downward invites the question of why it was wrong before, so standing still is the only free move.
• The forecast becomes a target the first time it leaves the team, which is a date on the reporting calendar rather than a mystery.
• Recording when a movement was first believed, against when it was reported, is the only measurement of the pressure that exists.

Records born here. The forecast as issued, each month, kept rather than overwritten · the date each movement was first believed · the date it was reported · what changed in the world to cause it · the assumptions the forecast rests on, restated each time · the reporting cycle at which the number first left the team.

What is coming next

Phase E ends here, and so does the middle of the project. Everything from now is about stopping.

Finishing turns out to be a boundary rather than an event, and four families of contract draw that boundary in four different places. Which one your contract uses decides how much unfinished work changes hands, who is responsible for it afterwards, and what the word complete is going to mean in every conversation for the next year.