Opening the file is not the same as receiving the building
The construction industry's shared file standard was supposed to let a building model leave one team's software and arrive intact in another's. The standard is real. The handoff between teams often isn't.
Escherbridge
The usual handoff on a building project goes like this. One team exports a model. Another team imports it. The file opens, and for a moment it looks as if the building has traveled. Then a beam is missing, a wall has the wrong type, or a property that lived on an object in the first tool is sitting in a comment in the second. What arrived was a file, not the building.
This is not the same problem as a model trapped in someone else's cloud. The file did leave. It arrived. What failed is the handshake: the assumption that a shared format means a shared meaning.
Industry Foundation Classes, or IFC, is the construction industry's shared file standard for describing a building. It is a public schema for describing buildings so that architectural, structural, and other models can be exchanged without each team living in the same proprietary tool. In 2018, Huahui Lai and Xueyuan Deng published a paper in the Journal of Civil Engineering and Management that tested bidirectional, IFC-based exchange among architectural, structural, and MEP models. The abstract is not optimistic. The fundamental causes, they wrote, are that "software tools cannot well interpret several objects belonging to other disciplines due to the difference in domain knowledge," and that "software tools have diverse methods to represent the same geometry, properties and relations, leading to inconsistent model data."
Differences among IFC-compatible tools are easy to treat as a vendor quarrel. It is more ordinary than that. A standard can be complete on paper and still underdetermined in practice. Geometry can be described more than one way. A property can be attached to an object, or to a type, or not attached at all. Two tools can both "support IFC" and still disagree about what a wall is.
The result is a file that validates and a building that does not survive the trip. That is not an edge case. It is what happens when a schema is asked to do the work of an agreement.
A standard is a shared vocabulary. A handshake is two parties confirming they used that vocabulary to mean the same thing. Exporting an IFC file is the first half. The second half is a check that the objects, properties, and relationships that left the first tool are the ones that arrived in the second, under identifiers that still point at the same elements. Without that check, "we exchanged IFC" is a status report, not a fact about the building.
This is why a successful import is a weak test. Software will often open a file it does not fully understand. It will drop what it cannot map, coerce what it can, and present a model that looks complete enough to discuss. The missing beam then shows up later, sometimes in the field. The format did not fail in the sense that the file was corrupt. The process failed in the sense that nobody wrote down, in advance, what "the same building" would mean after the transfer.
PrimusNeo includes a spatial runtime (AMP) and a layer for organizational memory (Fractal Engine). I will not claim that either has solved IFC interchange. The public AMP SDK, in its README, is a toolkit for 3D, geospatial, and media runtimes. I have not found a public, documented IFC round-trip in the repositories I looked at. Fractal Engine, still in alpha, can store models and hash a file and keep its history. Neither is an IFC translator. What they can offer, if they are useful here at all, is a record of what was sent, what arrived, and whether anyone accepted it. A hash of the file is a start. A list of objects that survived the trip, and a list of the ones that did not, would be better.
Interchange standards are worth having. The alternative is emailing native files around a project and hoping the right plugin is installed, which is not an improvement. A standard, by itself, does not close the loop. The loop closes when the receiving tool can show, in a form the sending team can check, that the objects they care about arrived with their geometry, properties, and identifiers intact — or that they did not, in which case the handover is incomplete and should be treated as such. That is a duller story than a new format, and it is the one the 2018 tests were already telling.