Information on a project is usually late for reasons that were settled months before anyone started to draw. I am Dr Sepehr Abrishami, I managed a design team in practice before I moved into research, and on Reimagining AEC this season belongs to the ISO standards that were written to stop it happening. This episode is about the delivery phase of an asset, as ISO 19650 Part 2 sets it out in eight activities. Start with a leisure centre. The Thornbeck leisure centre is hypothetical. A council is building it, with a pool hall, a sports hall and a wet plant room full of pumps, filters and air handling plant. The council has appointed a principal contractor, and the contractor has appointed a specialist to design and supply the pool water treatment plant. At the end of Stage 3 of the RIBA Plan of Work the council has to decide whether to proceed to construction at the current budget, and to make that decision it needs a coordinated services layout for the plant rooms. It has asked for that layout four weeks before the end of the stage, because it needs three weeks to review the information and one more week for its capital programme committee. The contractor passes that date straight down to the specialist. The specialist delivers on time, four weeks out, to the day. And now the contractor has a problem with no comfortable answer. It has no time left to review what the specialist sent before it must submit to the council. It can submit information it has not checked, or it can submit late. Nobody missed a deadline. The date was impossible from the moment it was written. So the question for this episode is what has to happen, in what order and by whom, for each appointment to deliver the information the client asked for. Part 2 answers it in a single process, and by the end you will be able to place any piece of information management work in that process. You will be able to name the activity it belongs to, the party accountable for it, and what must already exist before it can start. And you will know the handful of points where the process is most often broken, including the one that caught the contractor at Thornbeck. We will take it in six steps. First the shape of the process and the parties who run it. Then the client's groundwork before anyone is invited to tender. Then the tender, seen from both sides. Then appointment, where proposals become obligations and requirements travel down the chain. Then mobilisation and the production of information. And last, delivery of the information model and the close of the project. Thornbeck stays with us all the way. The first step is the shape of the process, and it begins with where Part 2 sits. It is one of a series of six published parts. Part 1 supplies the vocabulary, and the episode on the concepts and principles of Part 1 covers it. Part 3 covers the operational phase, Part 4 covers information exchange in either phase, Part 5 covers the security-minded approach, and Part 6 covers health and safety information. Part 2 is the one that says who does what, and when, while an asset is being designed and built. It replaced BS 1192 and PAS 1192-2, both now withdrawn, and the transition guidance, PD 19650-0, since withdrawn and cited here as history, warned that there is no one-to-one relationship between the old documents and the new. Where the old specification had one clause, Part 2 often has several, spread across different activities. So an old template migrated by find and replace reads as compliant and is not. That history, and the UK guidance on building information modelling, or BIM, that grew up around it, belong to the episode The UK BIM Framework and the Move from PAS 1192. There is also an edition point that catches experienced people. ISO 19650-2 is the international text. BS EN ISO 19650-2 is the British edition of the same text, with identical clauses, plus a National Foreword at the front and a National Annex at the back. Everything people think of as UK practice, the status codes, the revision codes and the container naming convention, lives in that annex. The annex is informative. Its recommendations use the word should, and the only places it uses shall are where it quotes the body of the standard. So the obligation in the body is that the common data environment lets each container carry a unique identifier built from agreed fields, and the annex offers the convention the UK recommends for meeting it. The copy to work from carries the words incorporating corrigendum February 2021 on its front, because that corrigendum revised both the foreword and the annex. Now the process itself. Clause five sets out eight activities. Assessment and need. Invitation to tender. Tender response. Appointment. Mobilisation. Collaborative production of information. Information model delivery. And project close-out. Behind those eight sit forty sub-activities, and they are spread very unevenly. Assessment and need has eight. Tender response and appointment have seven each. Mobilisation has three, and close-out has only two. Where a standard spends its detail tells you where it expects the work to be, and Part 2 spends it before anybody produces a drawing. The part most people get wrong is how often the process runs. The clause says the process applies throughout the delivery phase for each appointment, regardless of project stage. The standard's own map of the process marks assessment and need, and project close-out, as undertaken once per project. Everything from invitation to tender to information model delivery runs again for every appointment. At Thornbeck the council appointing the contractor is one cycle. The contractor appointing the pool water specialist is a second cycle, at a lower tier, overlapping the first. In that second cycle it is the contractor who writes requirements, weighs a tender and completes appointment documents. If you learned the eight activities as a programme running left to right, you will be puzzled to meet an invitation to tender in month eighteen. That is why the names of the parties are relational. Part 1 defines an appointing party as a receiver of information from a lead appointed party, an appointed party as a provider of information, and a delivery team as a lead appointed party with its appointed parties. Nothing in those definitions mentions design, construction or trade. They describe the direction in which information flows. A note attached to several of them adds that the term applies whether or not a formal appointment exists, so an in-house team with no contract is still an appointed party. One caution on the words. When the contractor issues requirements to the specialist, at clause five point four point three, the standard still calls it the lead appointed party. The useful question is never who the appointing party is on this project. It is who is appointing whom in this appointment. The work of managing information is described as a function, not a job. Clause five point one point one says the appointing party appoints individuals, plural, to undertake the information management function. In doing so it considers three things, which are the tasks it is handing over, the authority it is delegating, and the competency the people need. Most projects settle the tasks, assume the authority and never test the competency. The function may be delegated in whole or in part, to the prospective lead appointed party or to a third party, and the same option recurs one tier down, at clause five point three point one. But responsibility delegates and accountability does not. An appointing party that hires a consultant to run its information management has bought capability or capacity. It is still answerable. And where a conflict of interest may arise, such as one organisation producing information and also reviewing it for acceptance, clause five point three point one requires probity arrangements, so that the conflict is visible and managed rather than unspoken. The record of who does what starts in Annex A of Part 2, which lists all forty sub-activities as a responsibility matrix template. It offers codes for the party responsible, the party accountable, a party consulted and a party informed. The annex is informative, and its cells are empty. Allocating them is a decision each project has to make and be able to defend. The episode on roles and the responsibility matrix works through that matrix in detail. So the first step resolves like this. Part 2 is one process of eight activities, two of them run once for the project and six run again at every appointment, carried out by parties defined by their position in that appointment, under a function that can be handed on while the accountability stays put. The second step is assessment and need, the eight sub-activities the appointing party completes alone, before any delivery team exists. After appointing the function, it establishes the project's information requirements, at clause five point one point two. The clause does not hand the client a template. It asks the client to consider its key decision points, what it must decide at each, and which questions must be answered before it can decide well. Part 1 defines a key decision point as a moment when a decision crucial to the direction or viability of the asset is made. That is a higher bar than a stage boundary. Decision points generally align with stages, and generally is doing work in that sentence. So the method runs backwards from the way requirements are usually written. Start with the decision, then the question behind it, then the information that would answer it, and only then the deliverable. At Thornbeck the decision is whether to proceed to construction at the current budget. The question is whether the coordinated design can be built for the money. The information is a coordinated model with quantities the cost consultant can price. The plant room layout in our hook exists because of that decision, and for no other reason. Milestones come next, at clause five point one point three, and the first thing the client must consider in setting them is its own key decision points. Milestones are not evenly spaced dates. They are the moments when information has to arrive so that a decision can be taken. Get the decision points wrong, and every milestone after them is wrong too. Then three sub-activities fix the rules that every tenderer will have to work within. Clause five point one point four establishes the project's information standard, which says what information must look like. It covers how information is exchanged in several directions, within the client's organisation, out to stakeholders and to whoever will operate the asset, and between the parties delivering the project. It also covers how information is structured and classified, the method by which level of information need will be assigned, and how the information will be used once the asset is in operation. Two of those deserve emphasis. The method for level of information need is decided here, before tender, though the specification for each object comes later, under ISO 7817-1, which the episode on level of information need covers in depth. And the operational use of the information is a question the client must answer while it is still writing the rules for design. Take one Thornbeck object to see why the method matters. For the air handling unit at the end of Stage 3, the purpose is to check that the plant room is big enough and the structure can carry the load. So the geometry needs only a spatial envelope with correct overall dimensions and an absolute location. The alphanumerical information needs its classification, its type and a short list of properties such as operating weight, service clearances and power demand. And the documentation needs nothing yet. A single legacy number such as level of development three hundred bundles geometry and data together and could not express that, which is why an inherited schedule of such numbers has to be mapped rather than adopted. Clause five point one point five establishes the production methods and procedures, which say how information moves through its life. That means capturing what already exists, generating and reviewing new information, keeping it secure and distributing it, and delivering it to the client. Clause five point one point six establishes the reference information and shared resources the client will provide. Reference information is existing asset information, whether the client's own, an adjacent owner's such as a utility company's, licensed mapping or historical records. Shared resources are templates and libraries, and the standard names the BIM execution plan template among them. A client that supplies the template gets tender responses in one shape that can be compared. Here is what happens at Thornbeck if those clauses are left empty. Suppose the council names no classification system and supplies no execution plan template. One contractor assumes Uniclass because it is its house standard. Another proposes the system from its last framework. A third asks, and is told to propose something. The council now has three prices for three different amounts of setup work, and it has handed a classification decision to whoever happens to win. None of that is the contractors' fault. The rules are not frozen, though. At clause five point three point two a tenderer may propose additions or amendments to the information standard and to the methods and procedures, where its team needs them to work effectively, and those proposals are settled at appointment. The client sets, the tenderer proposes and justifies, and the client decides. The last two sub-activities are the ones most often assumed rather than done. Clause five point one point seven requires the appointing party to establish the project's common data environment, and the standard defines establish as implement, configure and support. Buying a licence is not establishing anything. The clause then lists what the environment must enable, such as a unique identifier for every container built from agreed fields, status, revision and classification held as separate attributes, controlled transitions between states with the user and date recorded, and access controlled at container level. Those are yes or no questions to put to a vendor. Part 2 strongly recommends that the environment is in place before the invitation to tender, so that tender information can be shared securely, and recommends that any third party hosting it is appointed separately. At Thornbeck the council already runs an environment across its estate, which is good, but it was configured for operational documents and does not enforce delimited container naming. That gap belongs here, before tender. The episode on the common data environment takes the states, codes and identifiers much further. Then clause five point one point eight, the information protocol. The appointing party establishes it, including any licence agreements, and it is then incorporated into every appointment, which is what gives it force. A protocol that sits in a folder binds nobody. It settles the obligations of each party in managing and producing information, any warranties or liabilities attached to the project information model, the background and foreground intellectual property, and the use of information during the project and after the appointment ends or is terminated. Background is what a party brought with it, such as an engineer's standard connection details. Foreground is what was created on the project, such as the model of this building's frame. The Thornbeck council intends to hand the project information model to its own asset team and to a facilities management contractor it will appoint separately. Whether it has the right to do that with information the delivery team produced is a foreground question, and the place to settle it is here, not at handover. So the second step resolves. Before inviting anyone to tender, the client has stated its decisions and the information they need, fixed the rules of the game, made its environment do what the standard says, and written a protocol that every appointment will carry. The third step is the tender, and it has two sides. On the client's side, invitation to tender has four sub-activities, and the first carries more weight than any other clause in the phase. Clause five point two point one asks the appointing party to establish its exchange information requirements in five parts, and each part attaches to every individual requirement. The requirement itself, traced up to an organisational, asset or project requirement. If it traces nowhere, it does not belong. Its level of information need. Its acceptance criteria, meaning the test against the project's information standard, its methods and procedures and any reference information or shared resources supplied. Its supporting information, which can include examples of similar deliverables, the item clients most often hold and least often issue. And its date, set against the delivery milestones and the client's key decision points, after allowing for the client's own review and its internal assurance. The episode on the concepts and principles of Part 1 walks through those five parts in more depth, so here two of them matter most. Acceptance criteria are the part most often missing, and without them acceptance becomes a matter of opinion. A delivery team can meet every stated requirement and still be told the information is not good enough. Dates are the part most often miscalculated. At Thornbeck the council does the arithmetic correctly. Its decision falls at the end of Stage 3, its review takes three weeks and its committee one more, so its requirement is dated four weeks before the milestone. That four-week date is right for the contractor. The trouble in our hook began one tier lower, and we will settle it in the next step. The remaining client sub-activities assemble the pack. Clause five point two point two gathers the reference information and shared resources, including information produced in earlier stages of the project, such as site investigations, feasibility studies and rejected options with the reasons for rejecting them. A new team that does not receive those either repeats the work at the client's expense or designs in ignorance of a known constraint. The clause recommends that the client states what each item may be relied on for by attaching a status code to it, and that the material is issued in a secure environment such as the common data environment. A survey sent by email to four tenderers leaves no record of who received which version. Clause five point two point three sets the tender response requirements and evaluation criteria. Clause five point two point four compiles the pack against a checklist of seven items, which are the exchange information requirements, the reference information and shared resources, the tender response requirements and evaluation criteria, the delivery milestones, the information standard, the methods and procedures, and the protocol. Four of those were made under assessment and need. Compile means gather, not write. Suppose the Thornbeck council issues six of the seven and omits the protocol. Every tenderer is now pricing an appointment whose intellectual property, licensing and re-use terms are unstated. A cautious contractor prices that risk. A less cautious one does not, wins, and raises it after appointment, when the council has lost its competitive position. The missing seventh was the one with commercial consequences. Now change sides. Tender response has seven sub-activities, and the first changes one verb. The appointing party appoints people to the information management function. The prospective lead appointed party nominates them, because it has no appointment yet and cannot commit anyone. It then establishes the pre-appointment BIM execution plan at clause five point three point two. That plan carries the names and histories of the nominated people, the team's information delivery strategy, a proposed federation strategy, a high-level responsibility matrix, any proposed changes to the methods and procedures and to the information standard, and a schedule of software with versions, hardware and infrastructure. That last item is the only place in clause five where software is mentioned, and it asks for versions because two task teams on the same tool two versions apart can lose information between them without noticing. The test of the plan is whether a client can read it beside its own requirements and see each one answered. A plan that describes the firm's usual approach, its accreditations and past projects, and reaches the requirements in a paragraph at the end, is answering a question nobody asked. The episode on the BIM execution plan takes the document itself further. Then two sub-activities that most tenders blur. At clause five point three point three each task team assesses its own capability and capacity. It is a self-assessment, because only the task team knows who has done this before and what else it has committed to. Capability is what a team can do, capacity is how much of it and by when, and competency, which the client asks about separately, belongs to individuals. The assessment covers the capability and capacity to manage information, the capability and capacity to produce it, and the information technology available. Each is judged on the number of people with relevant experience and on that experience together, because experience alone lets one person stand for a team, and numbers alone let an inexperienced team look adequate. At clause five point three point four the lead appointed party aggregates those assessments into a summary. Aggregation is arithmetic, and arithmetic hides things. At Thornbeck the contractor's building services task team has six people who have produced information to these procedures, which is sound. Only one of them has managed information the way the proposed strategy describes, so management capability rests on a single person. And its infrastructure already runs at eighty per cent utilisation because of another live project. Three findings, three different remedies, and an average would show none of them. Carrying the individual assessments beside the summary is permitted, not required, and it is the honest practice. The response closes with a mobilisation plan, a risk register and the compilation. The mobilisation plan at clause five point three point five is written now for work that happens after appointment, and for each item it states the approach, the timescale and who is responsible. Its items fall into three groups. The first is testing, of the methods, of the exchanges between task teams and of delivery to the client. The second is configuring the environments and procuring the technology. The third is people, through education, meaning knowledge, and training, meaning skills, through recruitment where capacity falls short, and through support for people who join the team later. The risk register at clause five point three point six is scoped to the timely delivery of information, and it records each risk with the intended response. Its sources include the assumptions the team has made about what the requirements mean. Writing down where you had to guess is uncomfortable and extremely useful, because a recorded assumption is a question the client can answer before it becomes a dispute. Clause five point three point seven then compiles four items, the pre-appointment plan, the capability and capacity summary, the mobilisation plan and the risk register, and those mirror what the client asked for at clause five point two point three. The clause says compile where available. That is not permission to omit a document, because each of the four is separately required by its own sub-activity without any such qualifier. That resolves the third step. A tender pack is complete when its seven items are present and its requirements carry acceptance criteria and honest dates. A tender response is credible when it answers those requirements one by one, and shows its weak points rather than averaging them away. The fourth step is appointment, where everything proposed becomes agreed. Its seven sub-activities produce the documents the rest of the project runs on. At clause five point four point one the lead appointed party confirms the BIM execution plan, in agreement with each appointed party. It is the same document, updated where something moved, not a new one. One asymmetry is worth catching. The wording requires changes to the information standard to be agreed with the appointing party, but says only confirm and document for the methods and procedures. Both belong to the client, so the prudent step is to seek agreement for changes to either. At clause five point four point two the lead appointed party further refines the high-level matrix into the detailed responsibility matrix. That matrix records which information will be produced, when it will be exchanged and with whom, and which task team produces it, and among its considerations are the dependencies between pieces of information. The building services team cannot lay out plant room services until the structural grid and the plant room envelope are fixed. A matrix without dependencies cannot be sequenced. Then the cascade, at clause five point four point three, and the answer to our hook. The lead appointed party defines each requirement for its appointed parties from two sources. Either it passes down a requirement of the appointing party, or it adds one of its own. That gives a test for any cascaded requirement. If it is neither, it should not be there. A requirement also divides as it travels. The council's single requirement for a plant room services layout becomes several, because mechanical services, electrical services and the pool water treatment plant sit with different parties, and the lead appointed party remains answerable for the parts adding up. The level of information need can rise on the way down. The council asked for spatial envelopes and a short property list, because its question was whether the plant fits and whether the structure can carry it. The specialist may need to model its connection points precisely, because the mechanical services team must connect to them. That is a legitimate addition and it should be recorded as one. And the lead appointed party may supply reference information and shared resources of its own, such as its model template or a worked example from its last scheme. Now the dates. The clause requires the lead appointed party to allow for its own review and authorisation time and its own internal assurance, subtracted from the date it was given, which was already reduced. At Thornbeck the contractor needs two weeks to review and authorise and one week for its internal assurance. So the specialist's date is not four weeks before the end of Stage 3. It is seven. Passing the four-week date down unchanged gave the contractor no time at all, which left it the choice in our hook between submitting unreviewed information and submitting late. A third tier would subtract again, and the party at the bottom may be working ten or eleven weeks ahead of the milestone. That is not a flaw in the standard. It is the real cost of review, made visible, so that the compounding is designed rather than discovered. Part 1 adds the requirement that governs the whole cascade. The exchange information requirements across all appointments should form a single coherent and coordinated set. Follow one requirement down and check four things. Every part of it should appear in the requirements below. The parts should add up to the whole. The dates should leave each review its time. And everything in the lower set should trace either upward or to a stated addition. Then the plans. At clause five point four point four each task team writes its own task information delivery plan and maintains it throughout its appointment. For each information container the plan lists its name and title, its predecessors or dependencies, its level of information need, its estimated production duration, the information author responsible and the milestones it is delivered against. The two most often left out are the dependencies and the durations, and they are the two that turn a list into a schedule. The duration is the one field nobody else can supply, and the standard glosses production as generate, coordinate, review and approve, so the estimate is more than modelling time. At Thornbeck the plant room layout takes four weeks, two to generate, one to coordinate with structures and architecture, and one to check and approve. Against a date seven weeks before the end of Stage 3, production must start eleven weeks out, so the grid and the envelope must be confirmed by then. At clause five point four point five the lead appointed party consolidates the task plans into the master information delivery plan, considering the dependencies between task teams and the review windows of the lead appointed party and the client. Consolidation is not stapling. The work is in the collisions, such as a structural grid planned to arrive two weeks after the services layout that depends on it is due to start. Once the master plan exists, the lead appointed party baselines it, tells each task team what has changed in its plan, and informs the appointing party of any risk or issue that could affect a delivery milestone. That duty falls at appointment, before any information exists, which is the only time the client can still act on it. The episode on the delivery plans covers both plans in full. Appointment closes by writing all of this into the appointments, at both tiers. At clause five point four point six the appointing party ensures that the lead appointed party's appointment documents include its exchange information requirements, the information standard and the protocol with any agreed amendments, the BIM execution plan and the master information delivery plan. At clause five point four point seven the lead appointed party does the same for each appointed party, substituting its own requirements for the client's. So the protocol reaches every appointed party unchanged. Both clauses say these documents are managed through change control for the duration of the appointment. The master plan is a live document and an appointment document at once, which is reconciled by treating the baseline as what the appointment holds and sending revisions through change control. So the fourth step resolves. Appointment confirms the plan, refines the matrix, cascades the requirements with dates that allow for review at every tier, builds plans that can be sequenced, and binds all of it into the appointments. Nothing later in clause five repairs an appointment done badly. The fifth step covers mobilisation and the production of information, where the plan is first tried and then worked. Mobilisation has three sub-activities, and in all three the lead appointed party acts, because only it can test whether information moves between task teams. They execute the mobilisation plan written at tender, so a thin plan gives a thin mobilisation. Clause five point five point one confirms that resources are available, which is a check on a fact rather than a judgement, and it delivers education and training. The standard names the topics for education, which are the project's scope, the exchange information requirements and the delivery milestones. Clause five point five point two procures and tests the technology, configures and tests the project's common data environment and any distributed environment the team runs with its connection to the project's, and then tests the exchanges between task teams followed by delivery to the appointing party. Sideways, then upwards. At Thornbeck mobilisation finds two things. The structures lead named in the capability assessment has moved to another project. And the clearance zones the requirements asked for as explicit geometry do not survive the exchange format building services and structures are using. Found now, it is cheap to fix. Found at the first coordination review, six weeks later, it would have arrived with production already committed. Clause five point five point three tests the production methods and procedures and records the results, refines the information container breakdown structure and verifies that it is workable, develops the team's shared resources, and communicates the methods and procedures to all task teams. Workable is the word to notice. Part 1 gives three purposes for dividing information into containers, which are letting teams work at the same time without clashing, supporting security, and keeping containers small enough to send. Those purposes pull against each other, and only trying the structure shows whether it holds. At Thornbeck the breakdown followed the four zones of the federation strategy. Two teams tried working at once and found that the treatment plant inside the wet plant room belongs to the specialist while the ductwork serving it belongs to building services. One zone, two owners and a junction neither owned. The fix was to split the zone into a plant zone and a treatment zone and name the interface. It took an hour of two people working, and it is the last point at which changing a method is still cheap. Collaborative production has five sub-activities. At clause five point six point one each task team confirms, before generating anything, that it can reach the reference information and shared resources it needs. Where it cannot, it informs the lead appointed party as soon as practicable and assesses the impact. At Thornbeck the building services team has the structural grid but not the plant room envelope, because architecture is a week late. A useful report says that the week consumes the float in its four-week duration and puts its milestone at risk. Saying only that the envelope is missing is not useful. Clause five point six point two then has each team generate information as its task plan directs, complying with the information standard and following the methods and procedures. And it contains the only prohibitions in Part 2. A task team shall not generate information that exceeds the required level of information need, extends beyond its allocated element of the breakdown structure, duplicates another team's information, or contains superfluous detail. Every one of those is something a diligent team does with good intentions, and every one costs time to produce, to review, to exchange and to coordinate. Two versions of one slab edge are worse than one. The second prohibition is safe only if the breakdown structure was verified, which is why the wet plant room split mattered. The clause balances the prohibitions with an obligation to coordinate and cross-reference with what others have shared. Coordinating across a boundary is required. Modelling beyond it is forbidden. Two gates follow, both run by the task team on its own work. The quality assurance check at clause five point six point three asks about form, meaning compliance with the information standard, and a container that fails is rejected back to its author before anyone reviews its content. The review at clause five point six point four asks about content, against the lead appointed party's requirements, the level of information need, and the information other task teams need for coordination. A successful review assigns a suitability and approves the container for sharing. In the UK that suitability is a status code from the National Annex. S zero is work in progress. S one to S five are shared and non-contractual, and each names a purpose and a recipient. S four sends a container to the lead appointed party for authorisation, and S five to the appointing party for acceptance. The A codes are published and contractual, and the B codes were deprecated by the 2021 corrigendum. Status and revision are separate attributes. Preliminary revisions start with P and contractual revisions with C. The episode on the common data environment covers the codes in full. The fifth sub-activity changes the party. At clause five point six point five the delivery team as a whole reviews the information model, to support the continuous coordination of information across it. Continuous, because it is not an event. It considers the appointing party's requirements and acceptance criteria and the containers the master plan says should exist. At Thornbeck the federated model shows two things. The specialist's treatment plant container is missing, because it is late, so part of the plant room cannot be coordinated. And the fixing loads the structures team needs are absent from the services containers, because nobody required them. The first is a schedule matter, tracked against the master plan and reported upward. The second is a specification matter, which goes back to the requirements issued at clause five point four point three. Treat one as the other, and the team works harder at the wrong thing. That resolves the fifth step. Mobilisation tests the arrangements while failure is cheap, and production works inside them, making only what was asked for, checking form, reviewing content, and reviewing the model as a whole against what should exist. The sixth step is information model delivery and close-out, where the information reaches the client and the project lets go of it. Part 4 frames every exchange as two decisions, and a paragraph on it belongs here. Decision A is the provider approving its own information for sharing, which Part 2 carries out through the check and the review. Decision B is the receiver authorising and accepting information for publication, and Part 4 notes it can run in two steps, authorisation by the lead appointed party and acceptance by the appointing party. Part 4 also sets the criteria a reviewer applies, and its criterion on communication, that information survives conversion between systems undamaged, is the one the mobilisation tests exist to satisfy. The episode on information exchange in Part 4 covers those criteria in depth. In Part 2, delivery has four sub-activities in two matching pairs. At clause five point seven point one each task team submits its own information for authorisation, in the common data environment, so the submission is a change of state rather than an email. At clause five point seven point two the lead appointed party reviews it against five considerations, which are the master plan, the appointing party's exchange information requirements, its own exchange information requirements, the acceptance criteria, and the level of information need. If the review succeeds, it authorises the model and instructs the task teams to submit for acceptance. If it fails, the model goes back to be amended and resubmitted. Then comes the recommendation that changes behaviour. Because partial acceptance can lead to coordination issues, the standard recommends authorising or rejecting the entire information model. If nine containers are sound and the tenth is not, the recommendation is to reject all ten. Anyone working from a partly authorised model coordinates against an absence, and coordination against an absence looks exactly like agreement. The second pair mirrors the first. At clause five point seven point three each task team submits for acceptance, and at clause five point seven point four the appointing party reviews with four considerations rather than five. The one that has gone is the lead appointed party's own requirements. That is right, because those requirements are how the contractor organised its own supply chain, and the client neither set them nor is bound by them. So the client judges the model against what it asked for, not against how the team chose to deliver it. The same recommendation to accept or reject the entire model applies at this gate. Acceptance moves information into the published state, and Part 4 says what that means. The receiver may use published information for key decisions, for completing the stage, and for later stages. Published does not mean finished or correct. It means the receiver is entitled to decide on it. Shared information, by contrast, is reference material to work from and check against. When a fault is found at either decision, Part 4 sets out three change actions. Identify the issues and risks relevant to the purpose of the exchange. Allocate them, by writing down every one and sending the record to the lead appointed party or wherever the appointment directs, rather than raising them in a meeting. And implement the change, which only the information provider may do, in its own work in progress, before the container comes forward again through the same gates. A reviewer who corrects a container moves responsibility for it without anyone agreeing, and the author's next revision overwrites the correction. Close-out has two sub-activities, and it falls to the appointing party alone, as assessment and need did at the start. At clause five point eight point one, once the completed project information model has been accepted, the appointing party archives its containers in the project's common data environment, which keeps the archive inside the environment, beside the three states, rather than outside it. It considers four things, which are the containers the asset information model will need, future access requirements, future re-use, and the retention policies that apply. None of them is about the project. The first is the bridge to the operational phase. At Thornbeck the plant room services layouts and the treatment plant container go forward, because someone will maintain that plant. The federated coordination models are arguable, because they resolved clashes that are now built, but they are the evidence if any acceptance dispute is live. The thermal analysis does not go forward as a model, because its outputs already sit as properties in containers that do. Retention policy is identified and applied, not chosen, from records management duties, sector regulation or, for higher-risk buildings in the UK, building safety duties. And the decision is reversible in one direction only. A container carried forward can be dropped later. One not carried forward is gone once the project archive's retention expires and the delivery team's copies have gone. Clause five point eight point two has the appointing party capture lessons learned, with each lead appointed party, in a knowledge store future projects can draw on. Then it recommends capturing them throughout the project, not at the end. A session at close-out remembers handover and commissioning, and rarely the appointment decision that caused the trouble. The faulty date at Thornbeck was knowable the day the specialist was appointed, and forgotten long before handover. That resolves the sixth step. Delivery passes two gates that look alike and ask different questions, either of which should take or refuse the whole model, and close-out keeps deliberately what the asset will need and records what the project learned while it was still fresh. Let me bring this together. The question was what has to happen, in what order and by whom, for each appointment to deliver the information the client asked for. The answer is the eight activities of clause five, two of them once for the project and the other six again at every appointment. The appointing party states its decisions, sets the rules, establishes the environment and writes the protocol before anyone tenders. It issues requirements with acceptance criteria and dates that allow for its own review. The tenderer answers those requirements one by one, assesses its teams honestly and plans its mobilisation. At appointment the plan is confirmed, the requirements cascade with each tier subtracting its own review time, and the plans are built from dependencies and durations and bound into the appointments. The team tests its methods before relying on them, produces only what was asked for, and checks and reviews its own work. The model is authorised and then accepted, whole or not at all. And the client archives what the asset needs and keeps the lessons. The contractor at Thornbeck lost its review time at the cascade, because it passed a date down without doing the subtraction. If Thornbeck leaves you with three habits, let them be these. When you receive a requirement, ask whether it carries acceptance criteria and whether its date leaves time for every review between you and the decision, and if either answer is no, raise it before you agree it. When you plan, insist on dependencies and durations in every task plan, because without them a master plan cannot reveal the collisions it exists to find. And when you test, test the exchanges sideways and upwards at mobilisation, because a format that drops geometry is cheap to fix before production and expensive after it. Two notes on currency. BS EN ISO 19650-2 of 2018, incorporating the February 2021 corrigendum, is the edition in force. Draft revisions of Parts 1 and 2 were open for public comment from March to May 2026, proposing a single information management process across delivery and operation, and proposing among other changes that the BIM execution plan be renamed the information production plan. 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. For reading, start with clause five of ISO 19650-2 and its Annex A, then the UK National Annex, then clauses six and seven of ISO 19650-4 on the decisions and the criteria. ISO 7817-1 of 2024 governs level of information need, and PD 19650-0, withdrawn guidance now read as history, records the move from the 1192 suite. Other episodes take the parts of this process further, among them The BIM Execution Plan, The Delivery Plans, Roles and the Responsibility Matrix, The Common Data Environment: Process Before Platform, Level of Information Need in ISO 7817-1 and Information Exchange in ISO 19650 Part 4. The next episode, The Operational Phase in ISO 19650 Part 3, is about how an owner keeps an asset information model true to the asset across years of use. Thornbeck failed one check, and it is the check I would put to any appointment. On a project you work on or study, take one date and count back from the client's decision, leaving every party in between its time to review. That is the test from Dr Sepehr Abrishami on this episode of Reimagining AEC.