In the discipline of information management, no term is mistaken for a product more often than the common data environment. I am Dr Sepehr Abrishami, I lead the MSc programme in building information modelling at the University of Portsmouth, and this season of Reimagining AEC reads the ISO standards one clause at a time. This episode is about the common data environment as ISO 19650 Part 1 defines it and ISO 19650 Part 2 specifies it, a process first and a platform second.
The scene I will open with is one I have constructed, on a project that is itself hypothetical. Two days before an information exchange on a school extension, the coordinator for the lead designer opens the facade contractor's model, which sits in the shared state, and sees brackets along one slab edge set out from the wrong grid line. The client's platform lets her open the model and move them. Ten minutes, and the exchange stays on time. Or she sends the model back, lets the facade contractor's own team correct it and clear it again, and perhaps misses the client's decision date. The platform's brochure says it is ISO 19650 compliant, and it will let her do either.
The school is Northgate Academy, an existing secondary school, and the project there is separate from the Northgate Quarter regeneration area heard elsewhere in this season. Northgate Borough Council is procuring a two-storey teaching block, a refurbishment of part of the existing block and external works. Kestrel Design Partnership is the lead appointed party and also produces the architecture, so the coordinator works for Kestrel. Vector Facade Systems produced the facade model, and Vector was appointed by Kestrel rather than by the council. The project is at Stage 4 of the Plan of Work of the Royal Institute of British Architects, Technical Design, and the exchange falls at the end of that stage.
The coordinator's choice opens the question for this episode. What does ISO 19650 require of a common data environment as a process, as distinct from the platform someone bought? By the end you will be able to answer the coordinator in one sentence, with a clause behind it. You will also be able to say what any container may be relied on for, take its identifier apart field by field, write a specification a supplier can be tested against, and audit an environment you did not choose.
There are six steps. First the definition and where the environment sits in the delivery cycle. Then the states and the two transitions. Then status, revision and classification, the three attributes every container carries. Then the container identifier. Then writing the specification, and last, auditing it, with two questions that look as if they belong here and do not. The Academy stays with us all the way.
The first step is the definition, and it repays a slow reading, because it is a single sentence. Before it, two more names, because the Academy's delivery team also includes Calder Structural Engineers and Halliwell Building Services, the appointed parties for the structure and for the building services. ISO 19650 Part 1 defines the common data environment at clause three point three point fifteen, and the definition has three parts. It is an agreed source of information. It covers every information container on the project or asset. And collecting, managing and passing on those containers happens through a managed process. Agreed means somebody wrote it down and everybody accepted it. Every container means the cost plan and the programme as well as the models. And managed means somebody decides who may move which container, at what point, and on what evidence. Most projects have the first two parts. The third is the one people skip.
A note under the definition separates two things that conversation runs together, a workflow and a solution. The workflow is the process. It says who may do what, to which container, in which state, what must be true before a container moves, and what is written down when it does. The solution is the technology that supports the workflow, where the containers sit and which buttons people press. The standard specifies the workflow. It does not specify the solution. That is why a standard published in 2018 still works on platforms that did not exist when it was drafted. So a system cannot make a project compliant. It can make a compliant process easier to run, and it can make a non-compliant process faster.
You will need to cite where each piece lives. Clause eleven of Part 1 says a solution and a workflow should be implemented, and accepts more than one route and more than one technology. Clause twelve of Part 1 sets out the states and the transitions. Clause five point one point seven of Part 2 makes establishing the project's environment a task of the appointing party, and Part 3 carries the equivalent for the operational phase at clause five point one point nine. One correction is worth carrying. Many templates cite clause eleven for the container states. They are one clause out, and the error spreads because people copy each other's reference lists.
Part 1 also allows for more than one system. It notes that on large or dispersed projects the information model does not all sit in one place, and working container by container is what lets one workflow span several platforms. The UK guidance on the subject, Guidance Part C of 2020, warns against the misconceptions that the environment is more about technology than workflow, and that single technology solutions dominate. What several systems demand is discipline at the joins. When a container crosses from one to another, its identifier, its attributes and the record of who moved it must all survive, or you have two environments and a gap. Nor does the standard mandate a folder structure, which surprises experienced people, because four folders labelled work in progress, shared, published and archive feel like the requirement. They are one way of expressing state, and state is not a location. A system that holds state as metadata and never moves the file meets the requirement. A system that moves files between folders and records nothing about who moved them is a tidy file store, not a common data environment.
Then timing. Clause six of Part 1, the information delivery cycle, is about when information is needed, by whom and for which decision. Clause twelve is about how, meaning what state a container is in and who may move it. A project with clause six right and clause twelve wrong knows what it needs and cannot trust what it has. The reverse gives you well governed containers that arrive at the wrong time. The fourth principle of clause six names the common data environment as the place where information is shared and coordinated, using open standards where possible and clearly defined operating procedures, so the environment is the mechanism the cycle assumes. Clause six also puts two checks at every stage boundary. The first asks whether what was handed over met the requirement. The second runs the other way, when the team starting the next stage asks whether what it received is fit to build on. It matters most where the appointed party changes, because a newcomer reads only what is written, and after a delay, because a container authorised eighteen months ago can still carry a published status while describing a site that has changed. The procedures for both checks are to be agreed before any exchange. And Part 2 recommends that the environment is in place before the invitation to tender goes out, so tender information can be shared securely.
So the coordinator already knows two things. The brochure proves nothing, because compliance belongs to a process the council has to write down, and the platform's permission to edit is a setting, not an answer. That resolves the first step. The standard specifies an agreed source, every container and a managed process, and the platform is only how that process is carried out on this appointment.
The second step is the workflow itself, the states and the two transitions. Start with a point of precision that the familiar four boxes lose. Clause twelve point one says the current revision of a container sits in one of three states only, which are work in progress, shared and published. It then adds the archive and gives it two jobs, a journal of every transaction and the development history behind each container. Three states hold what is current, and the archive holds the record of what happened. Treat it as a journal and it is what you interrogate when a dispute arrives, so it has to be written as events happen.
Work in progress, at clause twelve point two, is where a task team develops its own material, and no other task team should be able to see it or reach it. The clause adds that this matters most where the solution is a shared system such as a web portal, which is where almost everybody is, because the default permissions of most products are more open than the clause allows. If Calder's half-finished structural model is visible to the council because permissions only work by folder, the environment is not compliant, and somebody will look at that model and form a view.
The move into shared is the check, review and approve transition at clause twelve point three. It compares the container with the delivery plan, to ask whether it was due and is what the plan promised, and with the production methods and procedures the project agreed, to ask whether it is named and structured correctly. The originating task team makes this transition, so responsibility for the content stays with the organisation that produced it. Part 2 splits it into a quality assurance check at clause five point six point three and a review and approval for sharing at clause five point six point four, and the checker should not be the person who produced the information. The gate sits inside the task team, and it is not self-marking.
The shared state, at clause twelve point four, is wider than people assume. Anyone appointed who needs to look should be able to look, including other delivery teams, and only security narrows that. Then the rule that matters for our coordinator. A shared container can be read and cannot be altered. Where a change is needed, it returns to work in progress so its author can amend and resubmit it, which preserves the principle that whoever produced a container stays responsible for what is in it. The shared state has a second use, which the standard calls the client shared state, holding a container the delivery team has finished with while it waits for authorisation or acceptance. It is a use of the shared state, not a fourth state, and the transition guidance PD 19650-0 of 2019, which BSI has since withdrawn and which I cite here as history, confirmed that the concept survived from the earlier British documents.
The move into published is the review and authorise transition at clause twelve point five, and it works differently. It compares all the containers at an information exchange against the information requirements, for coordination, completeness and accuracy. Coordination is a relationship between containers, and completeness is a question about what is missing from a set, so an exchange is an event, not a queue. If containers trickle into published one at a time, nobody ever tests coordination across the set. A container that fails goes all the way back to work in progress, not held in shared with a comment attached, because it is about to be changed, only its author may change it, and only work in progress permits change. In this model visibility and editability are opposites.
Published, at clause twelve point six, holds information cleared for use, such as building something new or running an existing asset. At the end of a project the information model holds only published and archived containers, so nothing that never crossed the authorisation line is part of the deliverable. Published is not a claim that the design is finished or a warranty that it is correct. It is a statement about permitted reliance. That is what authorisation does. It draws a line between what the next stage may build on and what can still change, and the person authorising needs the standing to accept the consequences.
The archive, at clause twelve point seven, logs every container that has passed through sharing or publication and keeps each one's history. An entry for a once published container stands for information somebody may already have built on, which makes the archive a legal record rather than a housekeeping convenience. Clause twelve point one adds that project containers not carried forward stay available read-only, for any later dispute and for learning from the project, for a period set in the exchange information requirements. At close-out Part 2 separates archiving the project information model, at clause five point eight point one, from capturing lessons learned, at clause five point eight point two, and only what asset management needs crosses into the asset information model.
The owners follow from the clauses. The originating task team owns the move into shared. Part 2 splits the move into published into authorisation by the lead appointed party at clause five point seven point two and acceptance by the appointing party at clause five point seven point four. So on the Academy project three parties touch one facade container in turn. Vector's own task team lead approves it for sharing, Kestrel authorises it, and the council accepts it.
The whole workflow reduces to four questions you can ask of any state on any platform. Who can see it, remembering that seeing a container exists and opening it can be separate permissions. Who can change it, the content or only the metadata. What it may be used for, which is the question status answers. And what is recorded about how it got there. Answer those for every state and you have specified the workflow, and the rest is configuration. The answers make a pattern. Exactly one state permits change, and it is the one nobody outside the task team can see.
Then the difference between a permission and a precondition. The permission is who may press the button, and every platform handles that as a role setting. The precondition is what must be true before pressing it is legitimate, such as the quality check done, the review carried out by someone other than the author and the coordination issues closed. Most environments enforce the permission and record nothing about the precondition, so an audit can prove who moved a container and cannot prove whether they should have. Clause five point one point seven requires the user and the date at every transition. Add where the evidence for the precondition lives, which the clause does not demand, and the journal becomes something you can defend. Leave the workflow unwritten and four failures follow. Other teams reach work in progress because permissions were set by folder. A helpful coordinator edits in the shared state. Containers are published one by one and the coordination test never runs. And the archive is a folder of old files, so a question about what the structural team had issued in March takes three days of email. None of these is a technology failure. Each is a workflow nobody specified, so a default filled the gap.
So the coordinator has her answer. She sends the facade model back. It cannot be edited in shared, and editing it would move responsibility for Vector's content to Kestrel without anyone agreeing to that. If the error went into the exchange, the model would fail review and return to work in progress anyway, so the slow route is the only route. The Academy's container register shows a related defect, a container in shared whose move was recorded against Kestrel rather than its own task team lead. That may be legitimate, if the container has entered the client shared use on its way to authorisation, or it may be the wrong party making the move, which is a governance defect rather than a typing error. And a container in shared with no authoriser recorded at all is simply a defect under clause five point one point seven. That resolves the second step. The states decide who may see and change a container, the transitions decide who answers for it, and the record of every move is what proves the process ran. Part 4 calls the two transitions decision A and decision B, and the episode Information Exchange in ISO 19650 Part 4 takes them further.
The third step is the three attributes. Clause five point one point seven of Part 2 requires every container to carry status, revision and classification. It gives no codes, because codes are a national matter. The UK codes live in the National Annex, which exists only in the British edition, BS EN ISO 19650-2, as corrected in February 2021. A copy with only ISO 19650-2 on the cover has no code list in it, so when someone says their codes conform to ISO 19650, ask which edition they are reading.
Status, at clause N A point four point two of the National Annex, answers what a container is being offered for, and to whom. S zero is work in progress, inside its task team. The shared codes widen the audience as they climb. S one is for others in the delivery team to coordinate against, geometrically or not, so a schedule can be S one. S two is for other task teams to read and refer to, which is reference rather than coordination. S three is for review and comment within the delivery team. S four goes up to the lead appointed party for review and authorisation, and S five goes to the appointing party for review and acceptance. The annex also ties each shared code to the clause of Part 2 that produces it. S one, S two and S three come from the information model review at clause five point six point five, S four from submission for authorisation at clause five point seven point one, and S five from submission for acceptance at clause five point seven point three. So you never guess a code from a description. You name the step in the process that is happening, and the step gives you the code. If you cannot name the step, you are not ready to assign one.
The published codes work differently. An A code records that the party asked to authorise or accept the container came back with no comments. It is a signature, and it does not say why the container was issued. So the meanings of A one, A two and onward must be defined in the project information standard. A note in the annex ties those numbers to the work stages of the British briefing standard, BS 8536, in the two-part form current when the annex was written, and lets a project document another approach, which in UK practice usually means the 2020 Plan of Work stages or the client's own gateways. If the project standard is silent, the project is running on a default nobody chose.
Then the change that matters most in working life. The revised annex marks the B series as deprecated. B codes marked a container signed off with comments outstanding, a halfway position, and the workflow has no such position, because clause twelve point five sends a failing container back to work in progress. What you do about a B code depends on where you meet it. On a live project, treat the container as not authorised and send it back for a fresh decision. In a template, strip the codes before use and record the finding, and never renumber B one as A one, because that claims a sign-off nobody gave. In a legacy register from an earlier edition, leave the history intact and mark the codes as withdrawn, because rewriting a historical record to look compliant is worse than the original fault. The Academy's master information delivery plan template still carries the B series, so the question is a live one.
Three rules govern the code list as a whole. Extend it, never replace it, because a project may add its own codes in its information standard as additions, and a parallel set abandons the list. The S codes and the A codes differ in kind, the first shared and non-contractual, the second published and contractual, which is a change of legal character rather than a step up a scale. And Part 1 places the risk of using a container for anything beyond what its status permits on the person who used it. Not on the author. On the user.
Revision answers a different question, which is which one am I looking at. Status is about reliance and revision is about identity in time, and the two are independent. A container at its fifth iteration may still be unfit for anyone outside its task team. When someone says a drawing must be nearly there because it is at P zero six, they have read a revision as a status. Clause N A point four point three sets the convention. A preliminary revision is P followed by two digits, a contractual revision is C followed by two digits, and in work in progress a further two-digit suffix tracks iterations, so that two people holding P zero two can tell whether they hold the same file. Follow one of the Academy's containers. It starts as S zero at P zero one point zero one, which is the first value of every container. Three iterations later it is S zero at P zero one point zero four, and nobody outside Kestrel's team has seen any of them. Shared for coordination it becomes S one at P zero two, and the suffix drops away because it belongs to work in progress. Submitted to Kestrel for authorisation it is S four at P zero three, still preliminary. Authorised and accepted, it becomes A one at C zero one, and the next stage may build on it. Shared codes take preliminary revisions and published codes take contractual ones, so an S code with a C revision, or an A code with a P revision, means a step was skipped. The Academy's register holds both, and a filter finds them in minutes. Whether a revision restarts when status changes is a project decision, to be written down before the first exchange.
Classification is the third attribute, and the one most often left out. Clause N A point four point four points to Uniclass 2015, the UK implementation of the framework in ISO 12006-2, using its project management codes where possible, because a container is project information rather than a building component. It earns its place because the identifier is a good address and a poor description. A structural engineer's specification is identified as a specification from the structural discipline. Only classification says it is about the steel frame, and subject is how people search two years later.
How the metadata behaves matters too. The author sets it first, because only the author knows what they made. Approval and authorisation then amend it, so an environment that lets an author move their own container from S four to A one has a hole in it. The annex expects the environment to prompt for metadata on arrival and on revision. Where a project adds metadata, such as an issue date or a package reference, the annex points to BS EN 82045-2, the document management metadata standard, as the basis, so the project does not invent a private vocabulary that someone must map at handover. That resolves the third step. Status says what a container may be relied on for, revision says which one it is, classification says what it is about, and all three sit on every container, not only on the models.
The fourth step is the container identifier. The commonest complaint about naming is that you cannot tell what a file is from its name, and that complaint misunderstands the job. The identifier is an address, seven coded fields in a fixed order, separated by a delimiter and used whole as one string. A note in the National Annex says it deliberately has no description field, to keep identification apart from description, and description lives in the metadata.
The seven fields each answer a question, from widest to narrowest. Project asks which project. Originator asks which organisation answers for the information, which the annex settles by pointing to the detailed responsibility matrix rather than to whoever typed the name. Functional Breakdown asks which part of the works by function, such as a system or a work package. Spatial Breakdown asks the same by location, such as a level or a zone. Form asks what kind of thing it is. Discipline asks which technical specialism answers for it. And Number makes the identifier unique once the other six are fixed.
Here is the trap that catches experienced people. Templates still say volume or system, level or location, type and role, which is the vocabulary of earlier British practice. The revised annex calls the same positions Functional Breakdown, Spatial Breakdown, Form and Discipline. It is more than a renaming. In the old vocabulary, role was the specialism. In the current annex, Discipline is the specialism, in the sixth position. Map an old template onto the new names by matching words and you cross two fields over, and everything downstream is wrong.
Three structural rules hold the identifier together. The delimiter is fixed at clause N A point two point three as the hyphen-minus, so that software reads the identifier correctly, and not an underscore or whatever the platform inserts on upload. The characters are letters and digits only, in either case, so a project that insists on capitals is applying its own rule, and a finding should say which rule was breached. And the annex sets no lengths. It asks for fields as short as possible and leaves the values to the project information standard, agreed with each lead appointed party.
The code lists follow one principle. The annex standardises only what is the same on every project in the country, and leaves the rest to the project. Form and Discipline carry national lists, in single characters. Form is based on BS ISO 29845, the standard for document types, so D is a drawing, M a model, L a list such as a schedule, and T is text. A specification is therefore T, not the two-letter S P of older practice that some guidance still shows. Discipline has twenty codes, such as A for architecture and S for structural, and three of them do the awkward work. O covers a discipline that is not on the list, X a container that belongs to no discipline, and Z one produced by several together, such as a federated model. A client's own team puts its specialism in Discipline and the appointing party's code in Originator, so a cost plan from the council's in-house quantity surveyors would carry Q and the council's code. Discipline is about the work, and Originator is about the organisation.
Functional and Spatial Breakdown each carry two national codes, Z Z where several subdivisions apply and X X where none does, and the annex sets an order for the rest. First Uniclass 2015, then codes published for a sector, and only then codes the appointing party invents. Projects that jump straight to inventing pay for it in mappings at every handover. The functional breakdown can also be notional, and the annex's example is security classification, which lets the identifier carry the distinction that access control has to enforce. Project and Originator have no national lists. The project code should be recognisably distinct from any appointed party's job number, or it becomes a fossil when that firm is replaced. Number has a fixed length with leading zeros, and should not repeat information another field holds, because two fields carrying one fact will eventually disagree.
Now code one of the Academy's containers. The project is N G A E. The originator is K D P, for Kestrel Design Partnership. The functional breakdown is N B, the project's code for the new teaching block, and the spatial breakdown is G F, for the ground floor. The form is M, for a model, and the discipline is A, for architecture. The number is zero zero zero one, four digits because the project information standard fixes four. With a hyphen-minus between each field, the identifier reads N G A E, K D P, N B, G F, M, A, zero zero zero one. Four of the seven values came from the project information standard and two from the National Annex, which is normal, and it is why any audit starts with the project's own standard.
Then the provision that explains a decade of confusion. A note at clause N A point two point two asks the project information standard to set out how metadata may be appended to an identifier as extra fields, and in what order, where a container is taken out of the environment or exported from it. So appending status and revision is contemplated in one circumstance, and only where the project has defined it. Templates describing that exported form have been read as describing naming in general. Inside the environment the identifier is seven fields, and the metadata sits beside it. Put metadata in the name and three things go wrong. It goes stale at the next transition, because nothing forces a rename. It travels with copies into places where nobody knows it is stale. And it contradicts the record, and when the name and the environment disagree, no rule says which wins, so each side of a dispute picks the one that suits it.
A naming audit follows. Read the project information standard first. Check structure before values, meaning seven fields, the right order and the hyphen-minus between fields and nowhere else. Then check each value against its list, the number for length, any eighth element such as a status or a description, and finally duplicates. Write each finding in four parts. Quote the identifier exactly. Name the rule, and say whether it comes from the annex or the project standard, because a project can change its own standard and cannot change the annex. Give the corrected identifier. And state the consequence, because a wrong discipline code sounds trivial until you write down that the mechanical team will not see the container when they filter the register. The fourth step comes down to one sentence. The identifier is an address and nothing more, with the attributes kept beside it rather than inside it.
The fifth step puts the standard back together as a specification. Read clause five point one point seven of Part 2 as a procurement document. It assigns establishing, implementing, configuring and supporting the environment to the appointing party, which settles who owns it. It lists six behaviours the environment shall enable, not features. And it names no product, architecture or interface, because a requirement that names a product is a decision already made.
The six behaviours are a unique identifier for every container to an agreed convention, every field drawn from an agreed code list, the three attributes on every container, movement between states, the user's name and the date recorded at every transition, and access controlled container by container. Each becomes a requirement a supplier must demonstrate. The environment rejects a malformed identifier and says which rule it broke. Each field is restricted to a code list under change control, because a list any user can extend is not a code standard. All three attributes must be present before a container leaves work in progress. Movement happens only by controlled transition, and an unmet precondition blocks it. The transition record can be exported for a single container. And the access test is two users, one folder and different contents, which is the one most products fail.
The clause recommends a functional and non-functional requirements specification. The six behaviours are functional, tested by demonstration and not negotiable. The non-functional half covers how well, how fast, how safely and for how long, and it is negotiable, which means it is priced. Three of its items are left out most often and regretted most. Retention and export, because exporting the containers without the journal exports the evidence and leaves the proof behind. Support at the end of the appointment, when access quietly stops. And administration, meaning who controls permissions and how fast a change takes effect on the day someone leaves the project.
Timing carries as much weight as content. Where the appointing party hands the running of the environment to a third party, the clause recommends a separate appointment, made ahead of procuring anyone else, because a delivery team cannot plan or price against an environment that does not yet exist. And the environment should be in place before the invitation to tender, for security.
Every line of the specification should pass three tests. It names something a witness could see happen or fail. Its subject is the environment, because a line that says the team shall is a procedure, and belongs in the information production methods and procedures. And it names no product, interface element or job title, so it survives a change of platform. An untestable requirement in a contract is worse than none, because it suggests the point was covered. So the fifth step ends with an instruction. The council should have specified those six behaviours for the Academy before tender and appointed whoever runs the environment first, translating a clause that was already written as a specification.
The sixth step turns the same clause round and uses it to audit. A review is an informed opinion. An audit compares an environment against a stated instrument. Saying the environment is badly configured is a view. Saying it does not restrict field values to a code list under change control, contrary to clause five point one point seven and the project information standard, is a fact about documents, and with evidence of what you did and saw, it becomes a record. The three instruments carry different weight. The standard gives the six behaviours and the states, the National Annex gives the codes and the identifier structure, and the project information standard gives the project codes, the lengths and the meanings of the A codes. Most of what you check lives in the third.
Each requirement becomes a test. Attempt an upload with a malformed identifier. Ask who can edit the code lists, because the answer is often everybody. Report the proportion of containers missing an attribute, because a proportion makes a finding proportionate. Attempt a transition without its precondition, and record whether the environment prevented it or merely discouraged it, which is the heart of the audit. Export one container's journal. Have two users open the same folder. Severity is where reports lose credibility, because everything gets marked high. High means the environment cannot meet a requirement of clause five point one point seven however it is configured, and the finding goes to whoever chose the product. Medium means it can and the configuration does not, which someone can fix promptly. Low means the configuration is right and practice is inconsistent, which is a training finding. A compliance checklist sets the scope, but a tick records a conclusion, not an observation, so write the findings yourself.
One criterion links exchange review to environment audit. Part 4 sets criteria for reviewing an exchange, and UK guidance calls five of them the Five Cs. The first criterion, at clause seven point one, is the common data environment itself, and it sits ahead of those five. It asks that containers carry the naming and metadata the environment requires, that both survive where the environment spans more than one solution, and that the legally significant archive is maintained. If the Academy's environment spans three systems and the journal exists in only one, it fails, and a review that ran only the Five Cs would never notice.
Two questions look as if they belong to the environment and do not. The first is security. The environment implements security decisions and does not make them. As information manager you decide how a permission is configured once a classification exists, whether it can be enforced container by container, and whether the audit trail shows who could see what, and when. You refer whether the information is sensitive, what classification it should carry, which controls are proportionate and what follows a breach. Those belong to the security-minded approach of Part 5, which begins with a sensitivity assessment and a triage, and the episode Security-Minded Information Management in ISO 19650 Part 5 takes them up. So when someone on the Academy project asks for wider permissions, the question is not convenience. It is whether the classification changed, and if it did not, the answer is no.
The second is the operational environment. Part 3 has its own task for establishing it, at clause five point one point nine, and it is not the project environment left switched on. It lives as long as the asset, through changes of owner, operator and system, so it will be migrated at least once, and it faces a harder test, because a trigger event years later asks someone to rely on a container after everyone who made it has moved on. The episode The Operational Phase in ISO 19650 Part 3 follows that environment. A forward look, labelled as such. Draft revisions of Parts 1 and 2 were open for public comment from March to May 2026, and a draft revision of Part 3 followed, open for comment until July 2026. Among other things, the drafts propose a single information management process across delivery and operation. None of them is in force, and no publication date has been announced, so an appointment today cites Parts 1 and 2 of 2018 and Part 3 of 2020. That resolves the sixth step. An audit tests behaviour against named documents, routes each finding by severity, and knows which questions to refer.
Let me bring the six steps together. The question was what ISO 19650 requires of a common data environment as a process, as distinct from the platform someone bought. It requires an agreed source covering every container, run through a managed process the appointing party establishes, before tender where possible. Every current container sits in one of three states, with an archive that journals every move. The originating task team clears containers into shared, and authorisation and acceptance of a whole set at an exchange clear them into published. Any container that needs changing goes back to its author. Status, revision and classification sit on every container as metadata, coded to the UK National Annex and the project information standard, beside a seven-field identifier that is an address and nothing more. And every move records who made it and when, so reliance can be proved. The platform is how all of that happens on one appointment, not what makes it compliant. So Kestrel's coordinator sends the facade model back, because the shared state forbids editing and responsibility stays with Vector.
In your own work, three habits follow from this. When you meet an environment, ask the four questions of each state, who can see it, who can change it, what it may be used for and what is recorded, before you ask anything about the software. When you assign a status code, name the step in the process first, let it give you the code, and check that the revision series matches. And when you specify or audit, write every line as a behaviour a witness could observe, cite the clause or the project standard behind it, and say who has to act.
For reading, start with clauses three point three point fifteen, eleven and twelve of ISO 19650-1, then clause five point one point seven of BS EN ISO 19650-2 with its UK National Annex, and clause seven point one of BS EN ISO 19650-4. The readable companions are Guidance Part C on the common data environment, part of the UK guidance now held in the IMI Framework, named after the Information Management Initiative, and PD 19650-0, withdrawn guidance now read as history, for the move from the earlier British 1192 documents. The episode Information Management in ISO 19650 Part 1: Concepts and Principles introduces the states, and The Delivery Phase in ISO 19650 Part 2 places each transition in the sequence of an appointment. The next episode, Level of Information Need in ISO 7817-1, is about why a single grade such as level of development 300 specifies nothing, and what replaces it.
Before you next accept a login to a project's environment, at work or for a piece of research, ask to see the written workflow it was configured from. If nobody can produce one, you will know which part of the definition is missing, and that is the argument this episode of Reimagining AEC has made, from me, Dr Sepehr Abrishami.