A model leaves one organisation late on a Friday and is relied on by another first thing on Monday. I am Dr Sepehr Abrishami, and on Reimagining AEC this season works through the ISO standards one at a time. Blockchain, a technology built to record who passed what to whom, is one branch of the framework I call Reimagining AEC. Together. Part 4 depends on exactly that kind of record, and needs no blockchain to keep it. This episode is about information exchange as ISO 19650 Part 4 sets it out, meaning who decides that information may move from one party to another, and what has to be true before anyone relies on it.
Here is the scene, and it is hypothetical. A transport authority is delivering a project called the Harbourside Interchange, and at the end of a stage its principal contractor puts forward a submission of forty information containers. One of them is the asset data schedule, a spreadsheet the operator will live with for decades. It opens cleanly. Every date that remains is a valid date, and every duty value is a number. And it is wrong. Two dates have swapped their day and month, in opposite directions. One date has gone. Every duty value has been converted from one system of measurement to another, with no column to say so. One accented character in a location name has been replaced. No software raised an error. The authority's reviewer finds all this at the last gate, while the next stage waits. Accepting thirty-nine containers and rejecting one feels risky. Rejecting all forty feels absurd. And underneath sits the argument that will outlast the fix, about whose fault the dates are.
So the question for this episode is what decides whether information may move from one party to another, and what the party receiving it may check. Part 4 answers that precisely, in a standard short enough to read in an evening. By the end you will be able to name the actor behind every decision in an exchange, place a container on its route to publication, run both gates against the full criteria, and say what a record must carry so that the question of who checked this has an answer.
We will get there in six steps. First the three actors. Then the route a container travels, including what happens when the answer is no. Then the provider's decision, with a little known route that skips the shared state. Then the receiver's decision, and why it so often has two halves. Then the criteria, all seven clauses of them. And last, the record, and what can be handed to a machine. The Harbourside schedule comes with us all the way.
The first step is about who does what. When an exchange goes wrong, the information itself is usually recoverable within a day. What costs the fortnight is working out who should have caught it. Part 4 names three actors in clause three point two. The information provider provides information in a container. The information receiver receives it. The information reviewer reviews the information and its container. None of the three definitions names a job title, an organisation or a contract. Each says what an actor does in relation to a container, and nothing else. So the same person can be a provider in the morning and a reviewer in the afternoon, and the roles are allocated exchange by exchange, not appointment by appointment.
The provider definition carries a note that catches experienced people out. It counts the authors of requirements as providers. A client issuing exchange information requirements is the provider in that transaction, so a badly formed requirement is an exchange failure in the same sense as a badly formed model.
The receiver note says a receiver can be the appointing party or a lead appointed party, responsible for authorising and accepting information into the published state. Two verbs, and two separate acts, which the fourth step takes apart. For now, hold on to this. The receiver holds authority over the published state, and an exchange can have more than one receiver in sequence.
The reviewer definition repays slow reading. Its first note puts the task team leader at the first gate, before anything leaves work in progress, which is the review nobody writes down and everybody assumes happened. The second puts the delivery or operational team, including the lead appointed party, at the gate into the published state. And the third says that an information reviewer can be an artificial intelligence agent or an automated rules-based process. So automated validation on upload, often treated as a nicety beside the real review, can be the review, provided it applies the right criteria. The note changes who may hold the function and nothing about what the function requires. A machine can review. The decision to share or to publish stays with the provider or the receiver, and with a person who can be named on a record.
What a reviewer must do is set at clause six point three, the hinge of the standard, tucked between the two decisions. The reviewer either supports the exchange because the criteria are met, or raises the pertinent issues for resolution. There is no third option in which a concern is noted privately and the container waved through. And every criterion applies to every container, with one exception that belongs to the third step.
At Harbourside, follow the reviewer function through one exchange. It starts inside the building services task team, moves out to the principal contractor's delivery team, and ends with the transport authority. No appointment changed. Meanwhile each task team is the provider of its own container and a reviewer of everybody else's. So the first step resolves simply. Every decision in an exchange belongs to an actor defined by function, and after a failure the question is not who was senior, but which function each person was holding at that moment.
The second step is the route. Clause four sets out the exchange process on about a page, and one sentence in it organises the whole document. Processes are detailed in clause five, decisions in clause six and criteria in clause seven. Clause four also ties every exchange to the delivery clauses of Part 2 and the operational clauses of Part 3, so Part 4 invents no new process. It details steps your project already runs.
The route has four process steps, best held as a loop rather than a line. Generate work in progress. Use the shared state. Use the published state. Develop change actions. The fourth is not a stage after the third. It is the return path, reachable from either decision, and it always leads back to the first, because work in progress is the only place a container can be changed. Between the first and second steps sits decision A, the provider's. Between the second and third sits decision B, the receiver's. Two decisions, one set of criteria.
Part 4 does not define the states or their codes. Those come from clause twelve point one of Part 1 and the UK National Annex to Part 2, and the episode The Common Data Environment: Process Before Platform takes them in depth. One summary will do. Work in progress is the only state in which a container changes. Shared is where it is read against everything else, and clause five point two places the review there, with the rest of the shared state, including the client's reference information, as context. Published is where it is relied on, for key decisions, for completing a stage or trigger event, and for the stages that follow. That last use gives the published state its weight, because what is published is what the next team inherits.
Now the requirement in clause five that almost no project meets. Clause five point one point two uses shall. Before requirements are finalised, and before deliverables are exchanged, the methods and procedures you intend to use are tested to establish that information flows through them. Read the timing. Not before the first real exchange. Before the requirements are finalised, while you can still change what you are asking for. A test that meets the clause settles whether the route can pass the criteria it will later be judged against, exchanges something deliberately and looks at what arrives, and is written down, because a result held in one person's memory has to be run again when that person leaves. The deliberate exchange matters most, because a note to clause seven point four says that degradation shows only after a container has been exchanged.
Harbourside shows the cost. Every fault in the asset data schedule is a fault of the route, and one deliberate test exchange during mobilisation would have found them for about half a day's work. The replaced character also teaches something about test design. An encoding fault shows only if the test data contains a character that can break, so a tidy sample of plain text proves very little. Harbourside has one fault of a different kind, a building services duct and the structural frame designed into the same space, and no route test could find that, because testing a route says nothing about where two teams choose to design.
Clause five point one point three governs implementation, and its first requirement is easy to skip. Every exchange runs under the appropriate security arrangements. Not the sensitive ones. All of them, and the episode Security-Minded Information Management in ISO 19650 Part 5 covers how those are decided. The same clause asks for information shared coherently, early and often. Coherent sets the floor, and early and often set the frequency. Share as soon as a container makes sense on its own terms, and again whenever it changes.
Then the return path, at clause five point four, in three steps, identify, allocate and implement. Identification is limited to issues and risks relevant to the purpose of the exchange, which is the test between a reviewer who raises every imperfection and one who raises nothing because the deadline is close. At allocation, under clause five point four point three, the reviewer documents the issues and routes them to the lead appointed party, or wherever the appointment says, and somebody determines whether each one goes straight to a single party or to a collaborative review. The standard leaves that rule to you, and a simple one works. Allocate directly where one party can fix the issue inside its own container with no consequence elsewhere. Refer to collaborative review where the cheapest fix may sit in someone else's container.
The duct and the frame are that second case. The container stays in the shared state while the issue is identified, allocated and reviewed, because none of that moves a container anywhere. The review belongs with both teams, because the cheapest fix might be to move the duct rather than the frame. Only when the provider takes the container back into its own work in progress does it move.
A note at clause five point four point four carries the rule that settles the most arguments. Only the information provider may modify and revise a container in its own work in progress. Not the reviewer who found the fault, not the receiver who could fix it in five minutes, and not anyone else while the provider is on holiday. The rule makes the revision history mean something, and it protects liability, because a change by anyone else moves part of the provider's design liability without anybody agreeing. The corrected container then returns to clause five point one point three and faces decision A again on the full criteria, because a fix made under pressure is exactly the kind that breaks something else. So the second step resolves. The route is a loop with one way back, through the team that made the container, and the cheapest moment to find a fault in the route is a deliberate test before the requirements are fixed.
The third step is decision A, the gate most projects never write down. Clause six point one sets it out in four parts. The provider takes it, on its own information. It applies the criteria of clause seven, the whole set, the same set decision B applies later. It has two outcomes, approval for sharing or approval withheld. And the route out of a withheld approval is already specified. There is no third outcome in which a container goes forward with a reservation attached. Part 1 names the same transition at clause twelve point three as check, review and approve, and Part 2 places it at clause five point six point three, so a project running Part 2 already has this gate in its programme. Part 3 has the operational equivalent.
The reason the first decision goes to the party that made the work is cost, not trust. At this gate the container is still in the provider's own work in progress, so a fault found here costs an edit. Found later, it costs a documented issue, an allocation, a revision and a second trip through both gates.
Clause six point one also asks the provider to consider the effectiveness of two processes. Quality assurance, by the provider, to secure compliance. Quality control, by the provider and the receivers, to check for it. Assurance is preventive and acts before the container exists. Control is detective and acts on the container in front of you. A provider with no assurance can still pass by checking hard enough, and it pays for that at every exchange. In practice, a fault that recurs across three exchanges is telling you about the assurance, not the container, and a check that has never failed in two years is either watching something that never goes wrong or not watching.
Clause six point three rules out the light first review most projects run, and something more damaging. A provider who knows about a fault, shares anyway and plans to correct it next revision has taken the wrong outcome, and everyone downstream is coordinating against something the provider knows is wrong. If the programme cannot bear the delay, the honest route is to share with the fault recorded against the container as an open issue.
At Harbourside, decision A runs on three containers. The architecture container is withheld, because two rooms were renamed with no revision note. The building services container is withheld, because it had been exported against a different version of the agreed schema, and one wrong setting is enough when conformance is a criterion rather than a preference. The third, from the team designing the external works, passes correctly and still carries a fault, because the provider cannot detect degradation that happens only on somebody else's import. That is not a failure of decision A. It is why the mobilisation test exists.
The standard gives no form for this decision, so it helps to know what a record must carry to answer the question asked six months later, which is who decided this was ready, and on what basis. It names an individual, because a decision taken by a team is a decision nobody took. It carries a result against every criterion, and a blank criterion is itself evidence of a gap. And it records the consideration of effectiveness.
Now the provision almost nobody uses, the exchange that never reaches the shared state. The key that explains the process in clause four, and a note printed with it, provide for an exchange to omit the shared state, and both send you to clause six point four. Turn there and you find a short clause headed Exceptions, about leaving containers out of the review process, which never uses the word omission. The two readings join through one idea. A container that will never be revised needs neither the state that exists for coordination and review, nor the review that exists to catch what would be revised.
The clause is a conditional permission with a duty attached. A container that will not be revised may be left out of the review, provided the receivers still check its quality, its naming and its associated metadata. It names three case types for reviewers to consider. Certificates, because a certificate records something that happened. Monitoring signals, once their filters and triggers have been established, tested and first activated, because a reading is superseded rather than corrected. And photographs, because a photograph is evidence of a day. Consider is the operative word. The standard asks reviewers to consider exempting these cases. It does not exempt them. A project that takes up the permission writes it into its information production methods and procedures. And a feed nobody has configured is not exempt. It is unmanaged.
The reason is volume. A stage exchange on a large project carries hundreds of certificates and thousands of photographs. Route them all through review and the queue gets cleared dishonestly, with a stamp applied at speed, until the same stamp lands on a container that needed someone to look at it. The provision keeps the review meaningful by spending it where there is something to find, and the residual check sits very close to the criterion at clause seven point one, so it is a proportionate review rather than an escape from review.
The test is revision, not file type. A drawing issued in a fixed document format still represents a design that will change, so the full review applies. A photograph qualifies, and the same photograph with proposed changes drawn over it does not. Ask whether a later version of this container will exist and supersede it. At Harbourside, the asset data schedule feels like a fixed record, but it is derived from models that are still moving and is reissued at every exchange, so the full review applies. Exempt it because spreadsheets look simple, and the worst fault in the case reaches the published state unexamined. So the third step resolves. Decision A is a named decision by the provider against every criterion, and the only shortcut is for containers that will never be revised, with the residual check still owed.
The fourth step is decision B, where most of the confusion in daily practice gathers. Clause six point two mirrors clause six point one almost exactly. The actor changes from provider to receiver, the object from your own work to somebody else's, and the destination from shared to published. The criteria do not change. And the second outcome has two verbs, authorise and accept. A note to the clause explains why. The decision can be carried out in two steps, authorisation by the lead appointed party and then acceptance by the appointing party.
The cross-references settle what that means. For decision B as a single act, the clause cites clause five point seven point two of Part 2. For the two-step form, its note cites clause five point seven point four. Those are different steps of information model delivery, each preceded by its own submission, at clauses five point seven point one and five point seven point three. So the delivery team submits twice, to two parties, for two purposes, on any project where a lead appointed party sits between the task teams and the client, which is most of them. Part 3 runs the same pattern in operation, through its clauses five point six point six and five point seven point two, with a trigger event in place of a stage.
The two steps ask different questions. The lead appointed party asks whether the package is fit to leave its delivery team, which is about coordination across its own task teams. The appointing party asks whether this is what it asked for, and whether it can rely on it. Neither gets its own criteria, so the two-step form splits one review between two parties rather than doubling it.
Projects collapse the two for understandable reasons. In ordinary speech acceptance means any sign-off, so a lead appointed party that has authorised a package says it has been accepted, and everyone nods. And where the appointing party has delegated its information management function, both steps may be taken by people who sit together. They remain two steps, on behalf of two parties. Delegation moves the task. It does not merge the parties.
Two questions decide whether a package needs one step or two. The first is whether a lead appointed party sits between the task teams and the appointing party for this package. The second is whether the package answers the appointing party's own exchange information requirements. Two yeses mean two steps. If a sceptic needs convincing, open the status codes in the UK National Annex to Part 2. S four is for review and authorisation by a lead appointed party. S five is for review and acceptance by an appointing party. Two codes, for two reviews by two parties, each cross-referred to its own submission. That code set is already on the project, which the standard often is not.
The single step is real. Where an appointing party appoints directly, it is the receiver and decides once, and inventing a second signature there adds ceremony rather than control. But label the step, because one unlabelled signature is indistinguishable, eighteen months later, from a two-step package whose second signature never happened.
At Harbourside, the principal contractor sits between the task teams and the transport authority, and the package answers the authority's own requirements, so decision B has two steps. Now change one fact. If the authority had delegated its information management function to the principal contractor, both steps would be taken inside one organisation, and the record would have to be more careful, not less, showing which party each signatory was answering for. A gate review form with one signature box can record one step, and cannot show which step it recorded.
Clause six point two carries a second requirement. The reviewers consider the effectiveness of the quality management processes, the provider's assurance and control before sharing and their own control after it. At decision B you judge the chain as well as the container, and the evidence tells its halves apart. If the same fault keeps arriving, the provider's assurance is weak. If faults arrive that any coordination review would have caught, the review in the shared state is not happening. Those need different answers.
Now the dilemma from the opening. Clause six point three means forty applications of the criteria, one to each container, not one to the package. And Part 2, at clauses five point seven point two and five point seven point four, recommends that the appointing party accepts or rejects the whole information model rather than part of it, because partial acceptance can lead to coordination issues. An information model is a set of containers coordinated against each other. Accept thirty-nine, and you have accepted a set coordinated against the one you just rejected. Where a project does accept in part, the honest departure names the accepted containers by identifier rather than accepting with exceptions, says what they can and cannot yet be used for, and holds the published status until the outstanding container arrives. A rejection here also weighs more than one at decision A, because others have coordinated against the container, so the documented issue is how they learn that what they relied on is being withdrawn. So the fourth step resolves. Decision B usually has two halves held by two parties, and the record has to show which half each signature is, and for whom.
The fifth step is the criteria, which both gates have been applying all along. The standard's own key shows them fed by sources such as the project's information standard and its exchange information requirements. The criteria say what to check, and the sources say what to check against. With no information standard there is nothing to test conformance against, so a reviewer who says a criterion cannot be applied is often right, and the fix is a document upstream that a named party owes.
Then the count. UK practice often summarises the review as the Five Cs, which are conformance, continuity, communication, consistency and completeness. It is a good shorthand from guidance rather than from the standard, and accurate about the five it names. But clause seven holds seven criteria clauses. Clause seven point seven is informative. And clause seven point one, the common data environment criterion, is mandatory and left out. A checklist built from the shorthand applies five of seven, which is not what clause six point three asks for.
Clause seven point one asks three things, each with shall. That naming and metadata are correctly applied, so the container can be found and relied on. That metadata covers both what the workflow needs and what describes the contents, where most environments capture only the first. And that security and handling requirements have not been breached, with particular attention to who can see the container. Where the environment spans more than one system, as when a delivery team works in one and the client receives into another, its second note asks whether naming and metadata survive the move and the archive is kept across both. Open the container in the receiving system and ask whether its identifier, status, revision and classification are still there and still correct.
Conformance, at clause seven point two, checks the container against the exchange schema and data format chosen for the project or asset. Not the newest, and not the software's default, but the one the information standard names. That is why Harbourside's building services container failed.
Continuity, at clause seven point three, is a comparison, and its range widens from the other containers in the same exchange out to previous phases, so an operational container can be compared with what was delivered years before. It includes previous status changes, which only a status history can show. A shift in geospatial position or in calendar assumptions does the most damage, because the container still looks correct on its own. Naming is where continuity fails most often, and the naming it covers reaches processes and method statements as well as physical things. The note gives the purpose, which is that comparisons and progress tracking work so that only genuine change is detected. Harbourside's architects renamed their rooms correctly. What was missing was the record that lets everyone else tell a rename from a change. That cost about two minutes at decision A. Left alone, every later comparison would report differences that are not differences.
Communication, at clause seven point four, is the schedule's criterion. It asks whether information has survived translation or conversion without degrading or being lost, and whether it can be or has been stored without degradation, so it reaches forward to whether the receiving system will hold it correctly. Its note, that degradation shows only after exchange, is the most practical sentence in clause seven. The provider cannot see the fault at decision A, because the container has not been anywhere yet. That also settles the argument from the opening about whose fault the dates are. Neither team could have seen them at its own gate, and the place to find them was the test of the route.
The three sources of degradation share one property. A date reinterpreted between day-first and month-first order is still a valid date. A converted value is still a number. A replaced character is still a character. Nothing raises an error, so the standard asks you to compare rather than inspect, and names three ways. Visual inspection finds almost nothing at scale, because nobody reads four hundred rows. Reports scale, and find altered counts and totals. A checksum is the strongest signal and the least specific, proving something changed without saying what. Use the checksum to know, the report to find, and your eyes to judge whether it matters.
Now read the schedule again. One date failed loudly, because twenty-nine is not a valid month. Every other row succeeded quietly and is wrong. The converted duty values cost the most, because they will sit in the asset information model for decades and one day size a replacement. And one fault touches three criteria. Communication, plainly. The common data environment criterion, because the altered location name means the metadata describing the contents is no longer accurate, even though the format passes conformance. And completeness, because a lost date is a required value now absent. Mark it once and you have recorded a third of what happened.
Consistency, at clause seven point five, asks whether a container hangs together, inside itself and against its neighbours. Duplication and conflict in spatial, physical and process content is what most people call a clash, and because it covers process, two method statements assuming different sequences for the same work conflict as well. Contradiction between attributes is quieter and often dearer, as when a fire rating says one thing in the model and another in the schedule. Against the federation strategy, the criterion looks for gaps and for incorrect allocation, content that is correct but sits in the wrong container. Accuracy is quantitative, for positions and values, and thematic, for whether a thing is correctly identified, so a pump classified as a fan is inaccurate even if every number is right. And the reviewer considers the status code, because the code claims what a container is fit for and the criteria say whether it is.
Completeness, at clause seven point six, measures against the specification the exchange information requirements set, and two words carry it. Presence is whether the information is there. Adequacy is whether it is good enough for the purpose. Forty files where forty were expected are present, not complete, if half the required properties are blank. The reviewer also considers three widening requirements, this exchange, the phase, and the long-term management of the asset with its societal requirements, through standard naming, classification and open standards. Without an agreed classification or an open format, that third requirement cannot be met, and the finding belongs to the information standard, not the provider. The clause asks for a formalised procedure, which may use structured rules run repeatedly by machine. And it adds something quiet and valuable. Share the checking procedures with the providers, through the information production methods and procedures. A provider who knows what will be checked can pass at decision A. One who does not is guessing, and every guess costs a change action.
One detail here has aged. The notes to this clause, and a note at clause six point three, cite a European standard on level of information need, BS EN 17412-1, which ISO 7817-1, published in 2024, has since superseded and withdrawn. The reference has aged, and the standard is not wrong. The criterion is unchanged, and the specification you test against is now written under ISO 7817-1, which the episode Level of Information Need in ISO 7817-1 covers.
Clause seven point seven is informative and carries no shall, so no container fails for want of it. Its note says it helps at the end of a stage or trigger event, when what matters is whether the asset does what it was meant to do, and its items reach performance criteria, capacity requirements and health and safety, which the episode Health and Safety Information in ISO 19650 Part 6 takes up. So the fifth step resolves. Seven clauses, six of them mandatory, applied to every container, each tested against a document somebody owes.
The sixth step is the record, because in a dispute a review nobody can evidence did not happen. A record that says approved cannot show whether anybody checked. One that says seven criteria applied, six passed, one failed, corrected and applied again answers it completely. So a review record answers three things. Who decided, and for which party. Where, meaning which gate, which step and which container. And with what result, criterion by criterion, ending in the outcome and its route. A fault found at decision A goes back quietly into the provider's own work in progress. A fault found at decision B has to be announced, through a coordination issue, a message or a register entry, whichever the project already uses.
Construction operations building information exchange, known as COBie, comes up in every conversation about Part 4. BS 1192-4, the code of practice for COBie, was withdrawn in September 2022, the same month the British edition of Part 4 was published, and the records disagree about what followed. The BSI catalogue, BSI being the British Standards Institution, names no successor. Other records, including the National Foreword to Part 4, name Part 4. They do not reconcile, so quote both. What matters to a reviewer is simpler. The main text of Part 4 carries no COBie content, and its UK National Annex addresses COBie against three of the criteria. COBie is one format among several, and the annex says what to check. Annex A, which lists open schemas and formats, is informative, and the episode Information Delivery Manuals in ISO 29481 takes open formats further.
Then the machine reviewer. Forty containers against seven criteria clauses is two hundred and eighty judgements for one stage submission. That is why the standard points to machine checking in three places, the reviewer definition, a note at clause six point three and the completeness criterion. Most of the criteria reduce to a rule or a comparison, whether naming and metadata are present, whether the schema matches, whether a checksum or a report shows change. Two judgements resist, because someone has to know what the information is for. Adequacy, under completeness, and the appropriate accuracy test, under consistency. Automation removes the volume and leaves the judgements to people. And a reviewer who checked a sample has done something defensible, provided the record says it was a sample.
One look forward, 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. None of them is in force, and no publication date has been announced. Part 4 is not among them, and the 2022 edition is the one to cite. So the sixth step resolves. A decision is only as good as the record that shows who took it, for whom, and against what.
Let me pull the threads together. The question was what decides whether information may move from one party to another, and what the receiver may check. The answer is two named decisions against one set of criteria. The provider decides whether its own work may be shared, at clause six point one. The receiver decides whether it may be published, at clause six point two, usually in two steps, authorisation by the lead appointed party and acceptance by the appointing party. Both apply the criteria of clause seven to every container, except containers that will never be revised under clause six point four. And the receiver may check everything clause seven names, including the naming, metadata and visibility check at clause seven point one that the shorthand leaves out. At Harbourside, the schedule should never have reached the last gate. One test exchange during mobilisation, carrying one date, one value with its measure and one accented character, would have exposed the route before the requirements were fixed.
Three practices are worth adopting at once. Test the exchange route before the exchange information requirements are finalised, with test data designed to break, and write down what the test found. Build your review checklist from clause seven rather than from the shorthand, and publish it to the providers whose work it will judge. And wherever a lead appointed party sits between the task teams and the client, give every decision B record two labelled signature blocks, each naming the party the signatory acted for.
For reading, start with clause four of ISO 19650-4, then clause three point two for the definitions, clause six for the decisions and the exception, and clause seven last and slowly. Read clauses five point seven point one to five point seven point four of ISO 19650-2 with its UK National Annex, and ISO 7817-1 for the specification completeness is tested against. The episode The Delivery Phase in ISO 19650 Part 2 places these exchanges inside the eight activities of the delivery phase. The next episode, Classification and Structured Data, is about what it takes for information to be machine readable rather than merely digital.
Pick one exchange from your own practice or research, find the record of who let it move and against which criteria, and if that record does not exist, start there. I am Dr Sepehr Abrishami, and Reimagining AEC ends this episode on that record rather than on a summary.