The delivery lands at twenty to five on a Friday. A link, a covering note with the word “final” in the filename, and a polite request to confirm receipt so the invoice can go in. The file opens. The geometry looks like the building. Somebody says it looks all right, and the week ends.
That is hospitality, not acceptance. Accepting a delivery means agreeing that this identity — these units, this classification, these views, these clash closures, this issued set — can carry the next decision. From then on it is the thing everyone else measures against.
So the coordinator taking the delivery needs a routine that fits inside an hour and produces a written answer. Five gates, run in order, before the status changes, before the model is federated into next week’s coordination, and before you tell site it is current. Record fails as container names and revisions. A dashboard reporting “uploaded” says nothing about it.
Before the first gate
Name what you are testing against, from the information standard or the BEP: units, classification system, view-naming convention, clash-close rules, and what “issued” means on this job. If any of the five is missing from the project documents, write that down before you open a single model — it is already a finding, and not the deliverer’s fault alone.
Do not start with a fly-through. The pretty model is the last thing you look at.
Units come first, because a millimetre–metre mismatch poisons every measurement taken after it. Classification next: a misclassified object will not survive the next schedule or COBie extract. Named views after that, because a model you cannot navigate costs ten minutes of every meeting it appears in. Clashes fourth — a green federated view with an unclosed hard clash inside it is an argument on a delay fuse. The issued set last, because a model can pass all four technical gates and still be the wrong object if it is not the identity on the transmittal.
And refuse on the same day. “We’ll pick it up in coordination” is how a failed gate quietly becomes next month’s baseline.
1. Units
Open the project units before you open a 3D view. Confirm length, area, volume and angle against the BEP — typically millimetres for building geometry on a UK or Gulf delivery, not a silent mix of metres in one linked file and millimetres in another.
Then prove it on something you already know. A grid spacing, a slab thickness, a door leaf, a pipe diameter. Read the dimension in the model, then read it on the last issued drawing of the same identity. If they disagree, the model is not ready. If they only agree after you have quietly changed the display units, you have applied a viewer preference to somebody else’s problem.
Check the linked and exported identities the same way. A native file in millimetres that exports IFC in metres, or a consultant link that arrived in feet because the authoring template was never reset, will federate as a building sitting in the car park. The receiving party owns that check. The author owns the fix.
Write down the unit setting and the object you measured. “Units look fine” is not a record, and in six weeks you will not remember which door you opened.
2. Classification
Classification earns its place this high because it is what lets an object be found, scheduled and handed on without somebody standing there pointing at it. A filled-in parameter does none of that.
Sample rather than admire. Take twenty objects across the disciplines in the delivery — a wall type, a slab, a door, a light fitting, a cable tray, a plant item — and for each one:
- Is there a classification value on the agreed system (Uniclass 2015, the project mapping, whatever the BEP named)?
- Does that value belong to that object, or is it a leftover from a template, a copied family, or last job’s code?
- Can you schedule those twenty without an “undefined” row?
- Does the object carry the same code in the native file and in any export you were asked to accept — IFC, COBie stub, data drop?
A delivery that classifies the architecture and leaves the MEP as “Generic Model” is not half finished; it will fail the first quantity or asset extract anyone runs against it. Do not accept a promise that classification will be completed for handover. If this issue is for coordination or construction, the objects in scope need their codes now.
Write the twenty down, and write the misses. A percentage of “classified” from a plug-in is not this gate.
3. Named views
A named view is a shared place to stand. Without the agreed set, every conversation for the next fortnight opens with five minutes of orbiting.
Open the view list against the BEP or the MIDP, not against what the author thought might be helpful. You want the views the meeting will genuinely use: the agreed floor plans, sections on the grid lines that matter for this package, a 3D starting view that opens on the right building and the right zone, and whatever discipline-isolation views the clash process named.
For each required view:
- Does it exist, with the agreed name rather than a close synonym?
- Does it open on the right crop, the right phase, and the right workset or discipline visibility?
- Is it placed on a sheet, if the BEP said the issue includes sheets — and does that sheet number match the drawing register?
- Can a second person, sitting down cold, find it inside a minute?
Unnamed {3D} views, “Copy of Copy of Level 02” and a default camera aimed at the survey origin all tell you the same thing: nobody prepared this file for a recipient.
Where the delivery is IFC or another exchange format, apply the same test to the published views or the viewpoint file you were told would travel with it. A model that only makes sense inside the author’s own session is not portable.
4. Clashes
Do not accept a clash count. Accept a close-out.
Ask for the clash test this issue claimed to have run — the same selection sets, the same tolerance, the hard and soft split the BEP named — and then ask for the exceptions. A report listing twelve thousand intersections with no owners against them is a renderer with a large number on it.
Four facts, for this delivery:
- Which tests were run, against which other models and which revisions.
- Which clashes remain open, grouped so a person can act — by zone, by system, by owner — rather than dumped flat.
- Which clashes this issue claims to have closed, each with a dated action: remodelled, accepted as-is with a named approver, or deferred to a named later issue. “Will coordinate” is not a close.
- Whether any remaining hard clash sits in the area this model is authorised to instruct.
If the clashes live “in Navis” or “in the coordination model”, ask for the identity of that file and the revision it was run against. A clash result that cannot be pointed at the models you are being asked to accept is not evidence about them.
You are not re-running the whole federated project on a Monday afternoon. You are doing something narrower: refusing a file that arrived with a smile and no close-out. The longer argument about why counts without closures fail a job is in the twelve-thousand-clash note. Today’s question is only whether this issue may enter the receiving environment.
5. The issued set
The issued set is the identity. Everything above is a property of it.
Reconstruct four facts from the system of record rather than from the filename:
- Container name and revision, matching the naming convention and the register.
- Transmittal number, date, named recipients, and purpose — coordination, construction, information, not “FYI”.
- Suitability or status code from the agreed list, matching the title block or model property and the CDE state.
- Completeness of the set: native file, exchange format, views or sheets, clash close-out, and any linked files the model is useless without. A model that needs the architect’s link from last Thursday’s email is not a set.
Then run the test document control has applied to drawings for decades. Is there exactly one current revision of this model identity, and has the previous one been marked superseded? Two live federated models of the same building carrying different Friday dates is how a clash gets closed in one file and reopens in the other.
If the transmittal lists five files and four arrived, you do not have a delivery. If five arrived and the fifth is a different revision from the cover sheet, you do not have a delivery. Upload is not issue. Issue is a numbered act with a recipient and a purpose.
The receiving score sheet
Print it. One row per gate. The receiving party fills it in; the author does not edit it afterwards.
| Gate | What you sample | Pass | Fail — write the identity |
|---|---|---|---|
| Units | Project units plus one known object vs the last issued drawing | Setting matches the BEP; measured object agrees | |
| Classification | Twenty objects across the disciplines in scope | Agreed system, correct code, schedulable, export matches | |
| Named views | The view list the BEP or MIDP required for this issue | Named, correct crop/phase/visibility, findable cold | |
| Clashes | The tests this issue claimed, plus remaining hard clashes in the instructed area | Tests identified; opens grouped; closes dated; no unowned hard clash in scope | |
| Issued set | Transmittal vs what actually arrived | Name, revision, number, recipients, purpose, status, complete set, previous superseded |
A pass is boring. A fail is a container name. Do not average the five rows into a percentage: one failed gate is a refused delivery, and it carries forward until a new issue closes it. A gate that resets because somebody “uploaded a fixed file” without a new transmittal is how a receiving environment fills with ghosts.
Saying no on a Friday afternoon
The hard part is not technical. It is turning down a delivery from a team you will be sitting opposite on Monday, while someone more senior asks whether this really has to hold up the programme.
What makes that survivable is a sentence with an identity in it. Not “the model isn’t great”, which invites a negotiation about taste, but “the MEP objects in this container carry no Uniclass values, and the transmittal lists a revision that did not arrive”. A finding tied to a container and a revision is hard to argue with. A general misgiving is not.
ISO 19650, CDM, the Building Safety Act, NEC, FIDIC, JCT and COBie each have their own owners and their own texts, and the project’s BEP sits on top of the lot. None of them will come down to the receiving bay and check your units. That belongs to whoever signs for the delivery — on most jobs a coordinator with an hour, a printed sheet and a stamp they would rather not use.
Read next: Running a coordination meeting that ends in decisions, Agentic BIM: what happens when the model coordinates itself, Clash detection found twelve thousand issues. Nobody fixed any of them..