You will never be the one who reads it.

Four clocks, none of them yours

How long a record has to survive isn't a project decision, and there is no single answer to consult.

The contract sets one: the defects periods, their extensions, and the long-stop beyond them. The law sets another: a limitation period after which a claim can no longer be brought. An insurer sets a third, through whatever the policy requires be retained and for how long. And the operator sets a fourth, which is the longest of them, because an asset in service needs to know what is inside it for as long as it runs.

Four authorities, four periods, and the longest one governs each document. Nobody assembles that view, which is why records get destroyed on a retention schedule written for office paperwork rather than for a construction project.

And destruction is the failure that can't be reversed. Everything else in closure can be redone late and badly; a file that was shredded on a seven-year rule, for an asset with a forty-year life, is gone.

A signature at award decided the length

The legal clock is the one worth understanding, because its length was fixed before the delivery team existed.

Limitation periods are statutory and they differ by jurisdiction. In some systems the period is short and the parties aren't permitted to lengthen or shorten it by agreement at all. In the English tradition, executing a contract as a deed rather than as a simple contract secures a longer period — twelve years instead of six.

Read that as a lifecycle fact and it is remarkable. Whether your records matter for six years or twelve was settled by how a document was signed, in a room, at award, by people who never came to site. It is the last and purest example of what week 5 described: a decision taken early, for reasons unrelated to its largest downstream effect, and never restated to anybody who lives with it.

Two documents wearing one name

The as-built record is treated as a single deliverable and it serves two readers who need different things.

One is the operator. They need to know what is actually there: what was installed, where it runs, what it is made of, what was changed from the design. They will open it in a hurry, years from now, because something has failed.

The other is whoever has to reconstruct what happened. They need to know not the final state but the sequence: what was built when, in what order, against which revision of which drawing.

The first is a description of a thing. The second is a history of a process. A record built for one of them answers the other badly, and the second gets neglected because keeping it feels like preparing for a dispute — which nobody wants to say out loud in the month a project is being packed up.

The split also decides who should own each. An operator's record belongs with the asset and is maintained as the asset changes. A sequence record belongs with the contract file and is frozen, because its value depends on it not having been touched since.

The record that fails by design

The cost engineering treatment puts lessons learned somewhere specific: into a store of history that later work draws on, so that the next set of decisions about assets and about control starts from what actually happened rather than from what was assumed. The purpose points forward — other projects, other people, other years.

And the exercise fails by design rather than through anybody's carelessness. It is scheduled at the point where the team is smallest, the budget is closed, and the people with the knowledge have been assigned elsewhere. The reader is a stranger, so nothing can be assumed. And there is no feedback loop at all: the person who writes one never finds out whether it was read.

Every one of those is a property of when the task sits rather than of who is doing it, which means better intentions don't fix it. Only a different document does.

Which suggests writing a different document. A lesson expressed as a general principle is unusable by a stranger. A lesson expressed as a number — the output rate actually achieved, the real review turnaround, the true lead time from order to release — is usable by anybody, needs no context, and is the thing the next estimator is looking for.

An archive without an index is a warehouse

The last observation is about retrieval, and it decides whether any of the rest mattered.

A complete archive that can't be searched has the same practical value as no archive at all. The reader is a stranger, in a hurry, who doesn't know your folder structure, your abbreviations, or the name of the person who filed things.

So the index is the deliverable. What is here, what each thing is, which period it covers, which system or area it belongs to, and where it physically or digitally sits. Everything else in the archive is the material; the index is the only part that makes the material reachable by somebody who wasn't there.

Which is the same claim week 12 made about identifiers, arriving at the end of the project instead of the beginning. A thing that can't be found by rule has to be found by a person, and by this point there is no person.

And the index has to be written by somebody who still remembers what things are called. A stranger can't index an archive; they can only search one. Which puts the last useful act of the project in the hands of the person with the least time left on it.

System design

Row one is the only record in this whole dictionary whose required quality is about a person who has never seen the project. Every other entry could be verified by somebody who was there; this one can only be tested by handing it to somebody who wasn't and watching them look for something.

Record	Produced by	Required quality	Verified against	Feeds
The index	You, in the last week	Readable by somebody who has never seen your folders	Trying to find one thing with it	Whether the archive is reachable
Retention date per line	You, from the four clocks	The longest of contract, law, insurer and operator, per document	Each of the four separately	What survives and what is destroyed
Deed or simple contract	A phone call	Established, because it moves every retention date	The executed contract	Six years or twelve
Sequence record	The project, throughout	What was built when, against which revision	The as-built for the operator	Any later reconstruction
Numbers page	You, before leaving	Ten figures, no commentary, no recommendations	What the project actually did	The next estimator’s starting point

Row three is a phone call and it moves every date in row two. It is the cheapest item on this table by an order of magnitude, and the last person likely to ask the question is the person reading this in their final week.

Practical insight

Spend your last week on the index rather than on the archive.

One page listing what exists, what each set covers, and where it is. Then, against each line, the date it can be destroyed — which means going through the four clocks once and writing down the longest for each. Nobody else will do this, and once you leave nobody will be able to.

Then write your numbers page. Ten figures from the job with no commentary: the output rates you actually achieved, the review turnaround you actually got, the lead time from order to drawing release, the closure rate on the punch list. No lessons, no narrative, no recommendations. Just what the project actually did, which is what the next estimator would give a week to have.

And check one thing before you go: whether the contract was executed as a deed. It takes a phone call, it changes the retention dates on every line of your index, and you are the last person who will think to ask.

Key takeaways

• Every record this track has asked for is made by somebody who will never be the one to open it.
• The reader is a stranger, and the writing happens at the moment of least energy.
• Four authorities set retention periods — the contract, the law, the insurer and the operator — and the longest governs each document.
• Nobody assembles that view, so records get destroyed on a schedule written for office paperwork.
• Limitation periods are statutory and vary by jurisdiction, and some systems forbid the parties from altering them by agreement.
• In the English tradition, executing as a deed secures twelve years rather than six.
• So whether your records matter for six years or twelve was settled by how a document was signed at award.
• As-built serves two readers: an operator who needs the final state, and whoever must reconstruct the sequence.
• The second gets neglected because keeping it feels like preparing for a dispute.
• Lessons learned fail by design — smallest team, closed budget, stranger reader, and no feedback on whether anybody read them.
• A lesson as a principle is unusable; a lesson as a number needs no context and is what the next estimator wants.
• A complete archive that can't be searched is worth what no archive is worth, so the index is the deliverable.

Records born here. The index — what exists, what it covers, where it is · the retention date per line, taken as the longest of the four clocks · whether the contract was executed as a deed · the as-built set for the operator · the sequence record for reconstruction · the numbers page, ten figures with no commentary.

What is coming next

That is the life of one project, from a decision taken before it existed to a folder nobody will open for a decade.

Two weeks remain, and they run the other way. Instead of following a project through its phases, they ask what you would build if you arrived at one with nothing — no system, no baseline, no reliable data, and a report due on Friday. Which is the situation this last week describes.