The project told you in month two.

Four choices. There is no fifth.

Your telemetry shows a real variance. From this moment, there are exactly four things a project can do.

You can accept it, correct it, change the plan, or stop the project.

That's the list. And the value of the list is that it forces a decision into the open. Because in practice, most projects choose a fifth option that isn't on it: they look at the red number, feel uncomfortable, say something reassuring, and quietly do nothing at all.

That isn't accepting the variance. That's accepting it without admitting you did — and the difference matters enormously, because one of those is a decision somebody owns and the other is a decision nobody will remember making.

When is doing nothing legitimate?

Sometimes it genuinely is. But only under two conditions, and "I think it'll come good" is not one of them.

First: the variance is inside a tolerance you defined in advance. Before the project started, someone said what "normal" looks like — ten percent, twenty, whatever your organisation can stomach. Inside that band, noise is noise.

Second: there is no trend. The number is wobbling around, not walking steadily toward the edge. A variance that self-corrects is weather. A variance that grows a little every month is a diagnosis.

You will notice that this is exactly the float trend from Week 16, generalised. Never read a single month's number. Read the slope. A comfortable variance heading in one direction is not comfortable — it's early.

The fifteen percent rule

Now the hardest fact in this entire series, and the one I'd most like you to remember.

Large-scale studies of major programmes found something brutal: a project that is over budget or behind schedule at the 15% completion mark essentially never fully recovers. Not sometimes. Effectively never.

Think about what that means. On a two-year build, the ending is largely determined in the first three or four months — while everyone is still saying it's early days.

And the reason is not mysterious. It's Week 19's arithmetic. Your efficiency is a rate, and rates persist. The team that converts a dollar into 79 cents of building in month three is the same team, on the same site, with the same subcontractors, in month fifteen. Nothing about the situation improves on its own — and by month fifteen there's less time left to change it.

"We'll make it up later" is the most expensive sentence in construction. It assumes a future version of your project that is faster and cheaper than the one you have evidence for. That project does not exist. It has never existed.

So the whole discipline reduces to something almost unfairly simple: watch early, and act while acting is still cheap. The intervention that costs a fortnight in month three costs a claim in month twenty.

What "correcting" actually costs

Say you choose to correct. Be honest with yourself about what that requires.

Time, cost, scope and quality are tied together. You cannot pull time out of the equation without pushing something else in. To recover the schedule you must spend more money, deliver less, or lower the standard. There is no fourth lever, and any senior person who insists there is one is asking you to find the time in the only remaining place: your people's unpaid evenings.

Which brings the trap that catches good managers. Unpaid overtime looks free. It isn't. Tired crews make mistakes, mistakes become rework, rework consumes exactly the hours you were trying to save — and the best people, the ones you can least afford to lose, quietly start taking calls from other companies.

Recovery bought this way costs more than the delay it was hiding. It just puts the cost somewhere the report doesn't show.

Zero is the most dangerous number

One last thing, and it's the reason a good controller never trusts a summary.

The project dashboard reports a variance of zero. Green. Nothing to discuss.

Underneath: the structure package is a hundred thousand over, on the critical path, in genuine trouble. And the landscaping package is a hundred thousand under — because it hasn't started.

They cancel. Perfectly. And they have nothing to do with each other. One is a crisis. The other isn't a saving at all; it's a delay wearing a saving's clothes. Add them together and the number you get is worse than useless, because it is confidently, precisely wrong.

This is why aggregation is where control goes to die. The summary is for communicating. The package level is for managing. Any controller who only looks at the top line will keep being surprised by problems that were visible for months, one level down.

Every technique in this series exists to answer one question: where are we, actually? Not where we hoped, not where the report implies, not where everyone would prefer. And the reason that question matters is not so you can describe the problem elegantly at month eleven. It's so you can turn the ship at month three, while turning it still costs almost nothing.

Practical insight

Take your project and work out what percent complete it is. Genuinely — earned value over budget at completion.

If you're under 15% and any of your numbers are red, you are standing at the only moment where a real recovery is still possible. Behave accordingly, and be as loud as the situation deserves.

If you're past it and the numbers are red, stop writing recovery plans. Start writing options: what it costs, what we cut, or when it lands. That's not defeatism. It's the last useful thing anyone can do.

Key takeaways

✔ Four choices for any variance: accept, correct, re-plan, stop. Doing nothing quietly is not one.
✔ Accepting is legitimate only inside a pre-agreed tolerance, and only with no trend.
✔ Read the slope, never the single month — a comfortable number moving one way is early warning.
✔ A project behind at 15% complete effectively never recovers. The ending is set early.
✔ Efficiency is a rate, and rates persist — "we'll make it up later" assumes a team you don't have.
✔ To recover time you must spend money, cut scope, or lower quality. There is no fourth lever.
✔ Unpaid overtime isn't free: fatigue becomes rework, and your best people leave.
✔ A zero at project level can hide a crisis and a delay cancelling each other out. Manage at package level.

What's coming next

We know how to read the instruments and when to act on them.

Next week we turn the same rigour on ourselves. Project reviews: how to run one that produces the truth rather than a performance, how to measure whether your own schedule is actually any good — and why the healthiest question a team can ask is not "who caused this" but "how would we know sooner next time."

The project gets audited. Now the process does.