Half past four on a Thursday, and the coordinator is re-exporting for the third time. The services model came through with the ductwork but without the insulation, so the clash run reported a clear corridor that anybody standing in it can see is not clear at all. Nobody drew anything wrong. A property set was mapped one way in one office and another way in the other, and it took two days and a phone call to establish that much.
That afternoon has a price, and somebody pays it whether or not the project ever admits it. IFC is a tax you pay for an open conversation. Native files are a tax you pay for fidelity. Jobs that treat one of those as free tend to pay both, late, in clash meetings and claims.
The useful exercise is naming what each format costs, naming what it buys, and writing the rule down once — so nobody is renegotiating it every Friday afternoon.
The tax is the conversion, not the file extension
A native file is the file the authoring tool writes for itself: the parametric objects, the constraints, the families or assemblies, the views, the worksets, everything that makes a model editable by the people who built it. An IFC file is a published snapshot of a subset of that model, written to a schema other tools can read without owning the same licence.
The tax sits in the work between those two states. Export mappings. Property-set decisions. Geometry that tessellates instead of remaining a sweep. Identifiers that change between drops. A clash that exists in one file and vanishes in the other. Hours spent proving to a room that the missing fire damper was an export setting rather than a design decision.
That work is real. So is the work of not converting: the contractor who cannot open the native file, the facilities team who inherit a format they cannot query, the archive that dies quietly when the licence lapses. Refusing IFC does not abolish the tax; it just sends the invoice to a different desk.
What IFC costs
Pay this cost with your eyes open. Writing “IFC coordination” into the BEP and then acting surprised is the expensive route.
Geometry arrives thinner than it left. Curved work, nested families, voids, fabrication-level detail. A coordinator who clashes two IFC drops as though they were the authoring models will lose a morning to false positives and a longer one to false negatives — the pipe that looks clear because the insulation never came across.
Properties are a contract, not a courtesy. If the exchange information requirements never named the property sets, the classification and the units, the IFC turns up as a handsome solid with a few leftover attributes hanging off it. “We issued IFC” and “we issued the attributes you can price” are different sentences.
Identity is fragile. Objects split, merge or pick up new identifiers on the next export, and a comment pinned to last Tuesday’s express id is pinned to air. Hold type, location and a stable name instead, and treat a broken identifier as something to review rather than something to shrug at.
The schema has versions and flavours. IFC2x3 and IFC4 do not carry the same things, and model-view definitions strip different subsets. A coordination view is a decision about what you are allowed not to send. Two teams exporting different views and then arguing about completeness are arguing about a setting.
Somebody has to own the export. This part of the tax is plain labour: mapping tables, trial exports, a named person who can explain why a type vanished. If that person is the BIM intern on a Thursday, the open conversation will be both late and mute.
What IFC buys
Not moral superiority. A conversation the other party can enter without buying your stack.
A shared room. Architecture, structure, services, the contractor, the specialist fabricator and the owner’s reviewer, all looking at the same published snapshot in tools they already have. On a job with forty authoring licences and two hundred firms, that is worth real money.
A record that outlives the seat. Native files age with the version of the tool that wrote them. An IFC in the archive is still a model somebody can open in ten years without the original subscription, which is how you avoid paying a resurrection fee at every novation and every facilities handover.
A federation you can name. A coordination model built from published drops has a date, a revision and a purpose — an information container with an identity. A folder of native files on a shared drive is a set of private conversations that happen to sit next to each other.
Independence at the interfaces. You do not need every party on one platform to run a clash, raise a comment, or take a federated model into a meeting. You need an agreed schema, an agreed view and an agreed clock. That is what openBIM actually buys. It does not buy parametric editability, and it does not buy lunch.
What native files cost
Native is not free either. The invoice is simply less visible, because it lands on the design team’s own machines.
A closed room. Anyone without the licence, the version and the linked library cannot check the claim. The supply chain then works from plots, or from a viewer screenshot, or from last month’s IFC that nobody re-exported, and the unofficial set grows in that gap.
Version lock. A file that only opens in last year’s release becomes a liability the day a package is novated. So does the one that only opens if three central models and a stacked library path resolve. Fidelity that cannot travel is fidelity you cannot defend.
A read-only seat is not a federation. Handing the contractor a viewer does not hand them a model they can federate, query or archive. If the commercial model assumes they will coordinate in native, price the seats; if it does not, do not bury the assumption in a BEP clause.
You convert anyway, later. Handover, COBie, the asset register, the twin, the next consultant — something downstream will want a schema that is not your authoring format. Deferring IFC until practical completion compounds the tax, on data nobody validated.
What native files buy
Pay for this deliberately, in the places it earns its keep.
Editability. Parameters, constraints, connections, and the views the authoring team actually works in. Coordination that requires a design change has to go back to the native file, because an IFC comment that never returns to the authoring model is a sticky note on a photograph.
Fidelity for the people who can use it. Fabrication, reinforcement, a tightly constrained plant room — the native model holds what IFC will thin. Where two disciplines share a tool and a live link, the cheapest coordination is often the native one, and making them export to each other every night is a tax you chose rather than one the standards imposed.
Somewhere the change can happen. The issued IFC is a snapshot; the native file is where the object still has a future. Keep that distinction loud. Site relies on the issued container. The designer still edits the native.
| IFC | Native | |
|---|---|---|
| What it costs | Export mappings, thinner geometry, fragile identifiers, schema/version decisions, a named export owner | A closed room without the licence and version; you convert anyway, later |
| What it buys | A shared room others can enter without your stack; a record that outlives the seat; a nameable federation | Editability; fidelity for fabrication and tightly constrained work; somewhere the design can still change |
| Rule of thumb | Issue for coordination in IFC, with schema, MVD, origin, units and property sets named | Keep native where a change will actually be made next week |
Budget both, in writing
The usual failure is a BEP that says “IFC for coordination” sitting alongside a practice of sharing native files in a chat thread. The reverse happens too — a BEP demanding native from every trade, then genuine puzzlement when half the chain issues PDF.
Three rules, short enough to survive contact with site.
- Author in native. Issue for coordination in IFC. Name the schema, the model-view definition, the origin, the units and the property sets, and name the person who owns the export. A drop without those five is a file rather than an issue.
- Keep native where a change will be made. Same-tool design teams, fabrication models, anything you will edit next week. Do not flatten a live authoring model into IFC and then carry on designing in the snapshot.
- Never clash a format against a hope. If the coordination set is IFC, clash IFC against IFC and treat native-only findings as a separate review. Where two packages share native and have agreed to, clash native — and still issue IFC so everyone else can see the same decision.
Then run the test on Monday. Take last week’s coordination drop and, for each model, write two lines: what went out, with its format, its schema or version and its purpose; and what was held back, whether native only or not issued at all. If a decision in that meeting depended on something held back, the conversation was not open; it was private, with an audience. If a decision depended on geometry or a property the IFC does not carry, what you had was a picture.
Both failures are ordinary. Both get expensive when they stay unnamed, and the naming is a great deal cheaper before the pour.
Who owns what
buildingSMART owns IFC and its schema versions. Authoring-tool vendors own their native formats. ISO 19650, CDM, the Building Safety Act, NEC, FIDIC, JCT, COBie and MOMRA each have their own owners, and each of them assumes somebody has already decided which container carries which information. None of them will settle your export mapping.
That sentence belongs to the appointing party, written into the BEP before the first coordination drop: schema, model-view definition, origin, units, property sets, the named exporter, and whether native seats are in the contract or IFC is the coordination set. A Friday afternoon spent once is the cheap version. The expensive version is a trade contractor buying a licence they were never paid for, or a clash run against a native file half the room cannot open. Put the three rules above in the appointment, then run last week’s drop through the two-line test. If a decision depended on something held back, stop accepting that meeting as open.
Read next: Level of information need: specifying enough, not everything, Monthly Register — June 2025, Clash detection found twelve thousand issues. Nobody fixed any of them..