Beyond the CDE: why the next standard is about data, not documents

Every large job has a document controller who could tell you, without looking anything up, which part of the system is lying. She knows the published folder holds a general arrangement that the steel subcontractor superseded by email on Friday. She knows the suitability letters in three of the filenames were typed by somebody guessing. She knows the register and the files disagree about revision C, and which one she would bet on. All of that lives in her head, because nothing in the environment is obliged to hold it for her.

Most of the last decade's argument about common data environments was an attempt to get that knowledge out of her head and into a tool. Could the thing you called a CDE actually hold a state, carry a suitability, produce one issued set — or was it a shared drive with permissions, folders dressed as states and uploads dressed as an issue? That was a fight worth having, and largely it has been won on paper. It was never the destination.

A well-run CDE can still be a warehouse. Folders do not answer questions, and neither does a container whose entire payload is a portable document. The people who come back to the record years later — a facilities lead hunting a setpoint, an adjudicator hunting an issue date, an accountable person hunting a residual risk — are not asking for a library. They want an answer, without a fortnight of reading.

The story we tell ourselves about the CDE

The reassuring version runs like this. The industry has already bought its system of record. The folders carry names that sound like the standard. The models federate. Somebody is “on BIM”. The plotter still sings.

ISO 19650-1 and ISO 19650-2 describe something considerably more demanding than that. A CDE is an agreed source and a managed process, not a drive with a login. An information container is a named, persistent set of information — a file, a model, a schedule, a subset of any of them — rather than a polite synonym for PDF. States are claims about maturity. Suitability is a statement of purpose. Information requirements are the reason the store exists at all.

What most projects implemented was the diagram: four folders, a status letter in a filename, a BEP that uses the vocabulary correctly. That costume is convincing enough to get through a bid review, and it comes apart on an ordinary Tuesday when two trades want the same answer.

When a revision of those clauses does arrive, it will be read as a rebrand of the same four folders unless the text refuses to be satisfied by a permissioned library of pages. Waiting for a new number to fix it is mostly a way of postponing the question.

The failure has a shape

You can watch the same failure at a dozen different dates, in different clothes.

The unofficial set taped up in the trailer exists because nobody trusts the official cupboard to say which sheet is live. The clash report nobody closed exists because a PDF count is a deliverable and a geometry change is not. The twin that froze at practical completion exists because federation is a coordination memory rather than an as-built. The claim that collapsed for want of a contemporaneous file exists because issue lived in a mailbox. Owners wait a year for data they already paid for because handover was treated as a pack rather than a transfer of facts.

Each of those is a document problem in process-problem clothing. In every case somebody did produce a file, did transmit a ZIP, did tick a green cell. What operations or adjudication or the following trade actually needed was sitting in that file as a picture of a fact rather than as a fact.

Which is why “we have a CDE” stopped being an interesting sentence some years ago. The interesting one is shorter: can you ask the record a question and get a single answer?

A container is only the wrapper

ISO 19650 was right to put the container at the centre of the workflow. Without a named, persistent object there is nothing to hold a state, take a revision, appear on a transmittal or go into an archive. The four states still matter, suitability still matters, and an issued set is still what separates a managed process from a drop-box.

The trouble starts when the wrapper gets mistaken for the payload.

A published drawing is a container. “Which revision went out for construction on this package, to which organisations, on which date?” is a fact about that container. If the fact exists only as ink in a title block, or as a row in a register that is free to disagree with the file, then whoever needs it has to open the picture and infer.

A commissioning certificate is a container. What matters inside it is a set of joins: this tag, this test, this as-left setpoint, this revision of the drawing the test was run against. If those joins exist only as pages in a PDF named after a package, the client's facilities team cannot query them. They can hunt.

An operation and maintenance manual is a container, and four hundred of them are a pile. What operations will actually use is a dataset keyed to the asset tags they already maintain.

A standard that only tightens how containers move between folders will give the industry tidier warehouses, which is not the same thing as a record anyone can interrogate.

The PDF won, and it is easy to see why

Portable document format looks like the drawing, the certificate, the manual. It plots. It takes a stamp. It survives email and a change of software. Courts and authorities understand a page, which is no small advantage. None of that is a vice.

The vice is treating the page as the information model.

A PDF is a container of ink. It can carry structured fields and attachments, and almost never does, because the appointment asked for a file rather than for fields. Searching inside the PDF is not a query; it is a faster way of reading a pile, and the full-text filter hits every brochure in the ZIP.

Structured data means a statement somebody else can retrieve without opening the picture: identity, state, suitability, recipient, date, tag, location, serial number, warranty start, the revision a test was run against. Those statements sit alongside a drawing or a certificate quite happily. What they cannot do is be replaced by one.

COBie tried a version of this for handover, and in my experience the trauma was never really the schema. It was applying the schema once, at the end, as a workbook exported from a model that had already frozen. Structure added at the end behaves like a document: a spreadsheet with all the queryability of a print.

A twin nobody can ask for the issued set is the same costume under better lighting; so is a federated model that cannot say which container was published. Geometry is not a substitute for a fact about issue.

What a next standard would have to name

ISO 19650 has owners. So do the national annexes, COBie, the Building Safety Act, CDM, NEC, FIDIC, JCT and the Gulf digital-handover mandates. Revisions will arrive with numbers and consultation drafts, written by committees whose job that is. The claim worth making from outside is narrower.

Any text that wants to be more than a reprint of the four folders would have to make a short list of objects mandatory — not optional metadata, not a vendor diagram in an annex.

It would have to treat state as an enforced attribute of a container rather than a directory name that anyone with write access can dress up. A container that can be work in progress and published at the same time, with no recorded gate between them, is not being managed.

It would have to treat suitability as a refusal. A coordination drawing that can be plotted for construction because the shop needed something by Friday tells you the letter in the filename was decoration all along.

It would have to treat the issued set as a first-class object: these identities, this purpose, these recipients, this moment, reconstructable a year later by somebody who never met the document controller who left. A version history is a stack of saves, not an issue.

It would have to treat identity as joinable. A tag, a container, a requirement, a test and a manual that cannot be asked about together are five documents, and both operations and the golden thread assume they are one record.

And it would have to treat exchange as the satisfaction of a requirement rather than an upload. ISO 19650-1 already says an information requirement specifies what, when, how and for whom. A standard that stops at “the file arrived” has not left the document age.

None of that needs a new logo. It needs the next text — revision, successor or national overlay — to stop being satisfiable by a permissioned library of PDFs. A hundred per cent complete transmittal of unqueryable pages moves the labour of turning pages into statements onto somebody downstream, and owners already pay for that twice.

The environment was the beginning

You still need an agreed source, a managed process and the four states, and a project that skips that work will not get to skip it by buying a database.

What the environment gave the industry was the first standard it could hear, and the container was the first object it could name. The questions that decide whether any of the filing was worth doing are easy to ask and often impossible to answer: which containers went out, to which organisations, at which state, on which date — and, after practical completion, which tag, which manual, which residual risk. If answering one of them takes three people and a fortnight, what you have is a library.

The 2030 job does not look like a dashboard. It looks like a Tuesday on which a site supervisor, a designer's technician and a client's facilities lead — three people who do not share an employer — all name the same current issued set, and the record agrees with all three. The print can still be in somebody's hand. The fact behind the print has to live somewhere as a fact.

PIM to AIM fails for the same reason. Delivery files a project information model, operations needs an asset information model, and the cut between them gets treated as a rename of a document store. A standard that only describes a better CDE will reprint that failure with a later date on it.

The objections, and what I would say to them

You will say ISO 19650 already allows structured containers — models, schedules, databases. It does. Practice used the allowance to store more PDFs, and to call a federated model a record. An allowance is not a requirement.

You will say the metadata columns already exist. Columns that nobody is required to complete, and nobody is prevented from contradicting, are decoration. Structure you can skip is a header row.

You will say the next revision will sort out the folders. A tighter folder is still a folder. If the revision restates work in progress, shared, published and archive and stops there, the industry will reprint the diagram and carry on emailing FINAL_v3.

You will say courts and authorities want a PDF. They often do, and they should get one. Produce the page from the record; do not let the page be the record.

You will say none of this matters unless somebody is selling you a database. Test it without buying anything: ask in the next coordination meeting whether a standard that can be passed with a warehouse of portable documents is a next standard, or the last one with a new cover.

For whoever keeps the register

None of us writes the next clause, and the revision will land when it lands. Appointing parties can still refuse to buy another library. Put into the exchange information requirements that a container's state has to be enforced, that a suitability letter has to be able to stop a plot, and that the issued set has to be queryable as a set rather than reconstructed from filenames. Then, before you appoint, sit the preferred bidder in a room and ask them to produce — from the record, not from a document controller's memory — the revision a commissioning test was run against and the sheet a gang actually held on a named day. That morning is cheaper than the fortnight you will otherwise spend after practical completion turning pages into statements, and it is the one demonstration a permissioned store of portable documents cannot pass.