Kwame's contract says 37 hours and he assesses planning applications for a district council in Nottingham. In the week he counted properly, the assessments got 13 hours. Not because the week went badly. That was an ordinary October week with nothing wrong in it, and the other 24 hours had all been spent before Monday.
Recurring meetings took 9 hours, booked into the calendar in a previous job by a person who has left. The shared inbox took 6 hours, read in fragments across every hour of every day. Site visits and getting to them took 5 hours across two mornings, and neither morning was negotiable. Interruptions and the reload after them took 4 hours. What is left over is the number that matters: the number you plan against is what is left after the week is already spent. Kwame had been planning against 37 and wondering why Friday was always a surprise.
The arithmetic changes what a plan is. Four assessments at three hours each does not fit in 13 hours once the third one turns out to need a phone call, so the honest plan is three assessments and a named gap. Planning against the contract produces a week that fails by Wednesday and a person who concludes they are slow. The same person planning against the residue finishes what they said they would, which is a different job with the same hours in it.
An interruption costs the reload, and the reload is not billed to anybody
Kwame timed this for a fortnight because he did not believe it. A colleague's question at twenty past eleven, genuinely a two minute question, cost 2 minutes of answer and another 19 before the paragraph he was in made sense again. An interruption costs the reload, not the minutes it took. Six of those in a day is most of an afternoon, and not one of them appears anywhere as time spent.
The measurement that follows is the useful part, because hours are the wrong unit for this. Count how many times you started a thing, not how many hours you spent on it. An assessment started once and finished takes about three hours. The same assessment started five times takes about five hours and is worse, because each restart re-reads the same four pages to find the sentence that was being argued with. The number of starts is countable, it is honest, and unlike a feeling of busyness it moves when something is actually fixed.
Where the hard thing goes, and the meetings after it
The other half of the same arithmetic is which hours the residue is made of, because 13 hours of Kwame's best attention and 13 hours of his worst are not the same 13 hours. His good stretch is the first two hours of the morning, which is also when the calendar is easiest for other people to book. Put the hard thing where your best hours already are, and the meetings in the rest. That is one conversation with a manager and one recurring block, and it is the highest paid hour of admin available to anybody in this topic.
The task list underneath it has one rule doing most of the work. A task with no verb in it is not a task. Vodnik site is a subject and it will sit on a list for five weeks; write the access paragraph for the Vodnik report can be started in the four minutes before a meeting, which is where a surprising amount of finished work actually comes from.
The stall is a shelf of systems and an unfinished week
This is where people stop, and it is disguised as effort, which is why it can last for years. The method gets researched. A new app is set up on a Sunday and the tasks are migrated into it, tagged and coloured and dated, and setting it up is genuinely satisfying because it is a finite job with a clear finish, which is exactly what none of the actual work is. Three weeks later the app has 140 items and is avoided, and the search starts again.
What makes it a stall rather than a phase is that it never touches the count. Nothing about the app changed the 9 hours of inherited meetings or the 6 hours of shared inbox, because both of those are held in place by other people, and every rung above the first one costs somebody else something. A system's only load-bearing part is a fixed hour when the list is re-decided: everything is looked at, most of it is deleted, and the next week is planned against the residue rather than the contract. An hour on Friday in a notebook beats any tool with a Sunday setup and no Friday.
The order
Five rungs, and no app moves anybody up one
Each rung is a state that can be checked rather than a thing to do, and each is only reachable from the one under it.
- 05The list is re-decided whether you feel like it or notOne fixed hour that survives a bad week
- 04What you start, you finishCounted in starts, because hours hide the restarts
- 03Two of those hours are defendedBooked against other people, in writing, and kept
- 02You know how many hours the work actually getsThe residue, and it is nothing like the contract
- 01You know where last week wentCounted as it happened and written down, not remembered
Every rung above the first costs somebody else something, which is the whole reason the shopping is more popular than the climb.
The order, and the one the shop window suggests
Count first: where the hours go, and taking them back is the only cluster that has to come before the others, because everything after it is a decision about a number nobody has yet. Then doing one thing long enough for it to count, which is where the reload gets attacked. Then a method you follow when motivation is absent, which is worth having only once there is a real week to run it on, and making behaviour automatic, which is what keeps the Friday hour after the enthusiasm goes.
The order the shop window suggests starts at the system, because that is the part that can be bought and installed on a Sunday afternoon. What breaks is not vague: a method applied to an uncounted week schedules work into hours that were never available, so it fails in week two, and the conclusion drawn is that this method was the wrong one. The three clusters that pay best are the ones read last. Keeping what you learn so you can find it only matters once something is being finished worth keeping. The skill that makes every other one cheaper is the highest return on this page and almost nobody arrives looking for it. Reasoning better, on purpose is the one that decides whether the finished work was worth finishing, which no amount of the rest can rescue.
What to open in 2026
Toggl Track, or a spreadsheet with a start column and a stop column, for one honest week and then never again. It replaced the estimate of your own week made from memory, which is wrong in a specific direction: the parts that felt like work are remembered and the fragments are not, which is how 6 hours of inbox becomes about an hour when you describe your own week to somebody. Toggl Track has a free tier and a spreadsheet has always been free.
A list in one place, which is Todoist or Microsoft To Do or a text file on the desktop, and the choice between them matters far less than the number of places. Both apps have a free tier. All three replaced the list held in your head with three sticky notes as a backup, and the failure they fix is not disorganisation, it is the twenty seconds spent every hour re-remembering what was outstanding.
Calendar blocking, in Google Calendar or Outlook, both free with an account, is how the two morning hours stop being available to be booked. It replaced hoping the afternoon would be free. The mistake is a block with no name: an unnamed block gets moved by the person who booked over it, including you, whereas one that says which assessment it is for survives contact with a Tuesday.
Where this topic stops
This is the version of the work with nobody else in it, and it stops being enough the moment somebody is waiting. Once the deadline belongs to a group, the subject is delivering something by a date, with other people and the residue calculation happens across four calendars instead of one. The inbox half of the count is a writing problem before it is a filing problem, which is being understood, in every form the job takes: Kwame's 6 hours are mostly other people's unclear messages, and the ones he sends make somebody else's count worse. The retrieval methods in the method, which is the same whichever language you picked are the same ones the learning cluster teaches, tested at scale on a subject where forgetting is obvious. All of it sits under running work, getting paid, and the software it happens in.
The question none of the seven clusters below answers is the one Kwame's count actually raises. The 9 hours of recurring meetings were booked by somebody who no longer works there, and no method he can adopt alone will remove them; that is a conversation with a manager about a calendar, and it is worth more than every other hour on this page combined. Whether that is a productivity problem or a completely different one is the honest question, and the page that answers it would be filed somewhere else.