Keeping the model honest.

Everything up to now was planning. This is where planning ends and controlling begins — and it's a completely different discipline.

A planner builds a model. A controller keeps a model honest while ten thousand things try to make it lie.

The heartbeat

Control isn't something you do when someone asks for a report. It's a cycle, and it runs whether the news is good or bad.

Notice the last two steps, and notice their order. You analyse, then act, then report.

Most people invert this. They update the schedule in order to produce a report, and the analysis never happens at all. That's not project control — that's project journalism. You're describing the accident instead of steering away from it.

And the cycle has to be boring. Same day of the month. Same cut-off. Same rules. A schedule that gets updated when someone remembers is not a control system — it's a document that occasionally tells the truth.

Where the numbers come from

Update quality is decided before you open the software. It's decided by where your actuals come from.

Here's the version that ruins projects: at five o'clock on cut-off day, the planner phones round the site engineers. "Where are you on the level 3 blockwork?" Someone who is driving home, and who does not want a difficult conversation, says "yeah, about eighty percent."

That number is now in your model. It will be in the client's report on Friday. It will inform a decision about crew allocation next month. And it was invented in a car.

Actual dates come from site records — the daily diary, the inspection sheets, the delivery notes. Things that were written down when they happened, by someone who was there. Not from memory, and never from a number someone is comfortable saying.

Remember from Week 8: actuals override the logic. A wrong actual doesn't just make one bar wrong. It anchors the whole downstream forecast to something that never happened.

Stop asking how done they are

Now the single highest-leverage change you can make to your update, and it's just a change of question.

"How far along are you?" invites an opinion. And the opinion is always ninety percent, because ninety percent sounds like success and nobody has to explain anything.

"How many days do you still need?" invites a forecast. It's a harder question. It cannot be answered with a vibe. And it forces the real information out: eight more days — we're waiting on the glass.

Now you have something. You have a date, and you have a cause — and the cause is the thing you can actually go and fix.

This is why remaining duration is the number a schedule really runs on. Percent complete looks backwards, at work already done. Remaining duration looks forwards. Only one of them can move a date.

An activity that has reported ninety percent for three months is not ninety percent complete. It's a problem that nobody has been forced to describe out loud.

Float is a radar

Now, the part that separates a good controller from a reporter.

Most people read float as a status: this activity has twenty days of float, so it's fine. That's reading a single frame of a film and calling it the plot.

Float is only interesting over time.

The facade had twenty days of float in January. Green. Nobody looked twice.

In February it had sixteen. In March, eleven. Every month it quietly gives back five days — and every month the report says it's fine, because it still has float left.

It doesn't have float. It has a countdown. And anyone tracking the slope knew in February that the facade would be a crisis by May — while there was still time to do something cheap about it.

That's the whole job. Today's float tells you where you stand. The trend tells you where you're going.

And when float goes negative, understand what you're looking at. Negative float isn't a warning that you might be late. It's the model telling you that you already are late — the deadline has passed the work, and the only question left is who gets told, and when.

An out-of-date schedule doesn't just fail to help. It actively harms, because people make real decisions on it — ordering steel, booking cranes, promising handovers — believing it is current. The wrongness is invisible, and it compounds every week nobody touches it.

Practical insight

Change three things in your next update.

Get your actual dates from the written record, not from a phone call. Ask every site engineer "how many days do you still need," and don't accept a percentage. And before you write a single line of the report, open last month's file and compare the float on your top ten activities — not the value, the change.

That last one takes twenty minutes and it's the closest thing our profession has to seeing the future.

Key takeaways

✔ Planning ends at the baseline; controlling is keeping the model honest while the job moves.
✔ The update cycle is a heartbeat: collect, enter, recalculate, analyse, act — then report.
✔ Analyse before you report. Reporting without analysis is project journalism.
✔ Actuals come from site records, not from a phone call at five o'clock.
✔ Stop asking "how far along are you." Ask "how many days do you still need."
✔ Remaining duration is the forecast; percent complete only describes the past.
✔ Read float as a trend, not a status — the slope predicts the crisis months early.
✔ Negative float doesn't mean you might be late. It means you already are.

What's coming next

The cycle is running and the actuals are honest. But the model itself can rot from the inside — quietly, while every update looks perfectly normal.

Next week we run a full health check on it. The hard constraint that's hiding a two-week delay. The dangling activity making float out of thin air. And the one that catches almost everybody: what your software does when the site builds things in the wrong order — and why its default setting will hand you a finish date that is simply not real.

A schedule can be perfectly updated and completely broken. Let's go and find out.