Selin has 18 working days to move a ceramics supplier in Izmir into a new warehouse without stopping orders, and most of what is on her list has no effect at all on when it opens. Three tasks decide the date. The rest can slip days before anybody notices, and telling those two groups apart is the arithmetic under every cluster on this page.
Her list, with the durations from the people doing the work. Fit the racking, 12 days. Count and label the stock, 9 days. Move the stock, 4 days, and it cannot start until the racking and the count are both finished. Switch the address on the invoices, the site and the courier accounts, 2 days, and only after the move. Photograph the new studio, 6 days, wanted for the spring catalogue and needed by nothing else here. Chain the tasks that have to follow each other and the date falls out of the list: 18 days, which is 12, then 4, then 2.
The critical path is the longest chain of dependent tasks, and it sets the end date. Counting the stock is not on it. Nine days of counting run beside 12 days of racking, so the count can start three days late and the warehouse still opens on day 18. Those three days are float. Float is how many days a task can slip before the end date moves, and the racking has none: a task on the critical path has zero float, which is the whole of what being on it means.
The number decides where help goes. Selin can borrow two people from the shop for a week. Put them on the count and the opening does not move, because adding people to a task with float does not move the date; the count finishes early and then waits. Put them on the racking and every day they save is a day off the opening. The same holds for pressure: leaning on the person counting stock buys nothing and is remembered for a year.
A plan earns its keep on the Tuesday after something moves
A plan is not a record of intentions. It is the thing that answers two questions on a Tuesday: what moved since Friday, and whether the date changes. The racking fitter says he is two days behind, and that is two days on the opening, said out loud in week two rather than discovered in week five. The photographer says she is a week behind, and the answer is that it does not touch this date at all, which is worth saying just as plainly.
The chain is also what makes a change survivable. Somebody asks for a mezzanine over the packing bench. The question is not whether it is a good idea but whether it sits on the chain: if it needs the racking crew it is more days on the opening and somebody has to agree to those days in writing, and if it does not, it costs money and no time. Saying which of the two it is turns an argument into a decision.
Estimates, and the number somebody will hold you to
The racking fitter first said about two weeks, and two weeks on a plan is 10 working days. He meant 12. Nobody lied. A fortnight is a unit of calendar and a plan is a unit of working days, and the gap between the two is where a sixth of Selin's schedule went missing before anything had been built.
An estimate is a range, and the single number asked for is its optimistic end. Making that usable takes one extra sentence: ask what would make it the bad number, and when that will be known. The fitter's answer is 9 days if the floor is level, 16 if it is not, and he will know on the second morning. That turns a guess into an appointment. On the second morning Selin has a number instead of a feeling, a week before anything depending on it is due to start.
The order that pays, and the one nearly everybody takes
Scope first, and written down: the four things that have to be agreed first are what a date is meaningless without. Then turning an intention into tasks with owners, because a piece nobody owns is a piece nobody estimates, and the chain cannot be drawn out of pieces that have no durations. Then the chain itself, then the weekly work of keeping it on track, and only then the two things cut first, and paid for later.
The order people take starts at the other end. A tool is chosen, the work is typed into it, and an afternoon produces a plan-shaped object that feels like a start. What breaks is specific: a board holds tasks and states and not dependencies, so a project run on one has no chain and no date it can defend. The first slip lands on something with float and nothing appears to happen. The second lands on the racking. The opening moves and nobody can name the week it moved in, which is why the conversation about it is always the worst one on the project.
The order
Five treads, and the tool is not one of them
Each tread stands on the one below it. A schedule built before the scope is agreed is a set of dates for work nobody has named yet.
- 01Agree what is in and what is outIn writing, before any date is said out loud
- 02Break it down until a piece can be estimatedThe unit is a piece one person could put a number on
- 03Find the chain that sets the dateEverything off it has float and can wait its turn
- 04Run the week against the chainWhat moved, what changed, and whether the date holds
- 05Name what could go wrong, on paperWhile it is still a sentence rather than a week
Tools and certifications are not treads. Both are where people start, and neither of them answers what is in and what is out.
The stall is a plan nobody has opened since week one
Here is where people stop and do not know they have stopped. The plan is built in week one, when building it is a pleasure and nothing has gone wrong. Then it is not opened again, because opening it means finding out the date moved and telling somebody, which is the half of the work with another person in it. The file on the drive stays true and the project quietly stops matching it.
The test is a fact rather than a feeling: the modified date on the plan file is the status. A weekly project whose plan was last saved 11 days ago has run on memory for a week and a half, and from the inside that is indistinguishable from running well. Selin saves hers on Tuesdays, and the lines she edits most often are the two she least wants to write.
What to open in 2026
A board is what most projects actually live in. Jira, Asana and Trello all have a free tier that covers a small team, and all three replaced the status spreadsheet mailed round before a Friday meeting. None does the arithmetic above: a board holds what is being worked on and what is stuck, has no opinion about which of those decides the date, and read as a schedule it is the ordinary way a slip goes unnoticed for a fortnight.
Microsoft Project computes the chain, the float and the effect of a change, and it replaced the chart drawn by hand and re-drawn every time a date moved. It is a paid subscription with no free tier, as of 2026. For a move the size of Selin's it is more machinery than the job needs: a sheet with a duration column and a starts-after column does the same arithmetic on five tasks, and doing it by hand once is how the chain stops being a picture and becomes a number that can be argued with.
The one that changed since 2023 is a model that drafts. Pointed at the board's own change log it writes the weekly status note, replacing an hour spent restating what the board already says, and every major assistant covers that much on a free tier. What it cannot do is the judgement the note exists for, which is whether the date is still real: it will write that a project is on track out of a change log whose only movement was a task with a fortnight of float, because nothing in the log says which task that was.
Seven clusters, and the one you are in decides where to start
Short cycles are a different answer to the same problem. Working in short cycles, and what that requires is best read as a way of shrinking the chain until a wrong estimate is cheap, rather than as an escape from having one: a two week sprint has a critical path inside it and it is 10 days long.
What the plan actually lives in is the cluster to read after a first plan rather than before it, since the tool question is easy once there is a chain to put in it. The ones on the job listings answers a different question again, which is what a hiring manager scans for.
The Tuesday sentence is a writing problem before it is a scheduling one, and being understood, in every form the job takes is where that sentence gets built. Once the people on the chain report to you rather than lend you a fortnight, the subject changes into getting work done through other people, which is not a promotion from this one. All of it sits under running work, getting paid, and the software it happens in, and the sheet holding a five task chain is the one the programs most work actually happens in teaches.
Selin should have asked one thing in week zero and did not: whether the racking is on the chain because it must be, or because nobody asked the supplier for shelves that arrive assembled. Re-sequencing is the only move that shortens a date without anybody working faster, and it stops being available the day the first date is said out loud. No cluster below takes that on.