Sequence by what expires, not what matters.

The ninety days aren't free

The first thing to be honest about is that no part of this period is clear.

A monthly report is due in your third week and every month after it. A progress figure is wanted on Fridays. Somebody needs a look-ahead. All of that continues while you are working out whether the numbers underneath it mean anything, and none of it can be suspended while you find out.

So the system-building happens in the margins of the output-producing, which sets a hard limit on how much of it can happen at all. A first ninety days that assumes otherwise produces a good plan and no progress against it.

Which is worth planning for rather than resenting. Two or three hours a week is what is actually available, and a list built for two or three hours a week gets worked. A list built for a clear fortnight gets abandoned in week two and never reopened.

Judged before anything is finished

The second honesty is about timing of a different kind.

Your first report goes out before you have fixed anything, built from the arrangement you inherited, and people form a view of you from it. That view is formed in the first fortnight, and it is formed on the quality of a document produced by a system you have just discovered is unreliable.

Which argues for something specific rather than for despair. Say what the report rests on. A short note on the front — these figures come from these sources, this one is unverified, this one is an estimate — converts a document you can't vouch for into a document you can. It also does something else: it is the first evidence anybody has that somebody is now looking at provenance.

What expires

Here is the list that decides the order, and every item on it appeared earlier in this track.

The site as it was on the day of access, from week 19 — obtainable on one day and used, in week 35, three years later. The seven answers from week 38, obtainable while the person holding them is still employed. A contemporaneous note, from week 23, obtainable on the afternoon something happens and never afterwards. The date something was first known, which stops being knowable within about a week.

Against those, set the things that keep. A coding structure can be designed in month one or month two. A report format can be improved whenever. A register can be opened late, provided week 12's test says a gap in it will be visible.

The perishable items are individually small and none of them is urgent. That is exactly why they lose to anything that is urgent and permanent.

And losing them is silent. Nothing reports that a window closed. The site record that was never taken produces no error in month one; it produces an argument in month thirty-six, conducted by people who will never know a photograph would have ended it.

Three horizons, and what waits

Sorted that way, a first ninety days has a shape.

The first fortnight is for asking. The seven answers, the provenance of every number you publish, who holds what, which systems you can't reach. It needs no access, no licence and nobody's permission, which matters because you may have none of those yet.

Weeks three to six are for joining. Trace one work item end to end and mark where the identifiers stop matching. Find the joins a person is making in their head. This is the work that produces the identifier map, and it can only be done once the first horizon has told you what connects to what.

Weeks seven to twelve are for attaching. Every output you have fixed gets tied to a weekly cycle, because week 14 established that nothing survives without one. This is where a set of improvements becomes a system.

And one thing waits deliberately: the report. It is what everybody is asking for and it is the last link. Rebuilt first, it is a better view of the same broken chain, and you rebuild it again in month five.

Where this leaves the track

Thirty-nine weeks began with one drawing moving through six pairs of hands and losing something at each. They end here, and the thing they have been describing throughout is the same one.

A project is a sequence of decisions taken by people who won't be present for the consequences, recorded — or not — in documents read by people who were not present for the decisions. Every failure this track catalogued is one instance of that: the assumption that lost its label, the estimate that lost its accuracy class, the lead time that lost what it ran from, the duration that lost its source, the commitment filed where designers had no reason to look.

None of them was a failure of technique. Each was a fact arriving somewhere without the thing that made it mean something — and the correction, in every case this track examined, was one line written by somebody who had ten seconds and no obligation to write it.

That is a cheerful finding rather than a bleak one. It means the thing standing between a project and a considerably better record of itself isn't a system, a tool or a budget. It is a habit, available to anybody, at any level, starting on a Tuesday.

System design

Row five is the last entry in this dictionary and the only one addressed to a person who hasn't arrived yet. It is also the only one that makes any of the rest compound: without it every arrival starts where you started, and the reverse engineering gets done again from nothing.

Record	Produced by	Required quality	Verified against	Feeds
Expiry list and keeps list	You, in week one	Two columns on one page, so the perishable items stay visible	What is no longer obtainable a month later	The order of everything else
Note on the front of the first report	You, before anything is fixed	Names the source of each figure and flags the unverified ones	The sources it claims	Whether the document can be relied on
The seven answers, dated	You, in the first fortnight	Obtained while the people holding them are still employed	Handing them to somebody new	Everything in weeks three onwards
Identifier map	You, from one traced item	One work item end to end, with every identifier it carries	The documents themselves	Which joins a person is making
Handover note	You, whenever you leave	What you found, what you built, and what is still missing	The next person’s first fortnight	Whether any of this compounds

Row one is the instrument this week exists for. Two columns, written in a week when you know least, deciding the order of everything you do in the twelve that follow.

Practical insight

Make two lists in your first week and keep them on the same page.

The first is what expires: what you can only find out while a particular person is still here, what you can only record while a site is in a particular state, what stops being knowable in days. The second is what keeps.

Then work the first list even on the days the second one is screaming at you. Nothing on the perishable list will ever be urgent, nobody will ask you for any of it, and it is the only work in your ninety days that has a deadline set by something other than a person.

And accept the last thing plainly. You won't finish. Ninety days produces a partial system on any project worth the name, and the honest measure of the period isn't whether it is done. It is whether the next person to arrive inherits something better than you did — which is the same test week 37 applied to every record on this track, turned around and pointed at you.

Key takeaways

• Week 38 gave an order of dependency; this is a calendar, and they are different problems.
• Some things can only be obtained inside a closing window and the rest can be obtained at any time.
• Sorting the first ninety days by importance loses the perishable items; sorting by expiry keeps both.
• No part of the period is clear: reports, progress figures and look-aheads continue throughout and can't be suspended.
• So system-building happens in the margins of output-producing, which limits how much of it is possible.
• Your first report goes out before anything is fixed, and a view of you is formed from it in the first fortnight.
• A short note saying what the figures rest on converts a document you can't vouch for into one you can.
• What expires: the site on the day of access, the seven answers, the contemporaneous note, the date something was first known.
• What keeps: the coding structure, the report format, and any register whose gap would be visible.
• The perishable items are small and never urgent, which is why they lose to things that are urgent and permanent.
• First fortnight to ask, weeks three to six to join, weeks seven to twelve to attach each output to a cycle.
• The report waits deliberately, because rebuilt first it is a better view of the same broken chain.
• You won't finish, and the measure is whether the next person inherits something better than you did.

Records born here. The expiry list and the keeps list, on one page · the note on the front of the first report saying what it rests on · the seven answers, dated · the identifier map from one traced work item · the cycle each fixed output is attached to · the handover note for whoever arrives next.

What is coming next

Nothing, on this track. Eight modules and two hundred weeks end here.

What they amount to is one claim, made from eight directions. The failures in these thirty-nine weeks were not failures of technique. They happened because a project moves facts between people who never meet, and a fact that travels without the thing that makes it mean something arrives as a number nobody can use. Everything on this site is a way of writing that thing down while it still costs nothing.