The model is the deliverable

The most honest sentence in this industry sits in the boilerplate of nearly every BIM execution plan and model transmittal, in one wording or another: the model is provided for coordination purposes only; in case of discrepancy, the drawings govern. We produce a database of the building, and then contractually demote it beneath its own printouts.

Here's the strange part: the argument over what the deliverable should be is already settled - in writing, in the model's favour. In May 2011, the UK government put it in its Construction Strategy: "Government will require fully collaborative 3D BIM (with all project and asset information, documentation and data being electronic) as a minimum by 2016." ISO 19650, the standard this industry now claims to follow, is built around the information model as the thing that gets delivered and accepted at every exchange. In its own ontology, a drawing is just one "information container" among the many that make up the information model - a sibling of the schedules, reports and geometry, not their master. The mandates are fifteen years old; the standard wrote the hierarchy down. And yet the industry went the other way: it reneged on a commitment it had already signed. Why?

It wasn't an accident - the breach is written into the paperwork. The AIA's own guidance describes standard practice bluntly: design teams use models internally to produce their contractual 2D drawings, and shared models arrive "accompanied by a disclaimer stating that the model cannot be relied upon." The UK notarized the demotion: the CIC BIM Protocol - the official instrument for putting BIM into appointments - states that the project team member "does not warrant, expressly or impliedly, the integrity of any electronic data delivered." NBS's own analysis called that clause what it is: a reversion to marking models "for information purposes only." The industry didn't fail to write the model into its contracts. It wrote the model out of them, carefully, in every jurisdiction.

None of this is new - the pathology is older than the file formats. From 1963 to 1966, the Tavistock Institute ran the Building Industry Communications Research Project, the UK's first serious study of construction communication. Its researchers went in assuming an orderly, plannable sequence of stages; the evidence made them drop it. The real project ran on what the study called the adaptive functions - variations, post-contract drawings, redesign, error and crisis handling, design decided on site. The 1966 report concluded that the informal systems "seem to produce more realistic phasing of decisions and more realistic flexibility in the face of… the inevitable uncertainties in the construction process" (Dainty et al., 2006).

Sixty years later, the split hasn't closed; it has been re-implemented in newer formats. The contractor sequences from the model, the fabricator details from it, the clash reports that keep the site out of trouble come from it. Where the formal channels can't carry the data, teams route around them - the CSVs, plugins and scripts that appear in no appointment document; I spent a chapter of my PhD documenting these informal channels, leaner than the official ones because nobody designed them by committee (Stefanescu, 2020). Everyone relies on the model in fact; the drawings carry the project in law. Even the lawyers who wrote the disclaimers concede reliance is implicit, whatever the transmittal says.

We ended up with the worst of both worlds: a deliverable nobody builds from (PDFs), and a database nobody warrants (the model). Every downstream consumer - cost consultant, facilities team, auditor - is pushed back onto derivatives, re-deriving quantities from sheets generated from data they're not allowed to trust. That's the machinery I described in the last issue, running at contract scale: we render the database onto paper, then make the paper the truth.

In fairness, practice reneged for reasons - and the deepest one is trust. A drawing is a frozen, bounded, signable statement: you can stamp it, insure it, argue about it in a dispute. Nobody wants to warrant a living database with ten thousand unexamined parameters in it. But let's not pretend the drawing's halo is earned. Back when I was a practicing architect, I once lost half a day convinced I was going mad - the dimensions on one side of a building refused to add up, off by twelve centimetres - until I realised someone had hand-edited the figures on the drawing to make it look "constructible." The frozen, signable, governing document did exactly nothing to protect us.

Trust - the kind sociologists insist must be continuously produced, not declared - doesn't come from warranties. It comes from verifiability. And the drawing won on access, not philosophy: for most people who must accept a model, it has always been a locked box - you can't sign what you can't open, read, or check. Even diligent ISO 19650 projects show the symptom: "acceptance" of the information model means, in practice, checking the containers - file names, status codes, CDE metadata - because checking the information inside was never feasible at scale. You can be fully compliant and never once look at the data.

That constraint has quietly disappeared. A model can now be opened in a browser by anyone, every property readable without an authoring seat - and, more to the point for acceptance, checkable: completeness, classification coverage, naming compliance, evaluated on the data itself in seconds, on every version. That is trust being continuously produced, mechanically - the exact thing the disclaimer gave up on. And making the model govern doesn't mean couriering a nine-gigabyte file with a signature on it: the deliverable is the information - addressable, queryable objects you can open, filter and verify - not a bigger monolith. The disclaimer is now protecting a workflow, not a party.

This is where asset owners come in, because deliverables are demanded, not volunteered. No design firm can unilaterally make the model govern - the appointment does that. The most sophisticated clients already work this way: design-build and integrated teams have been model-first for years - when designer and builder share risk, the model is the basis of the work, and drawings become what they always were: views generated from the record. Even the most conservative corner of the industry is building the bridge: the AIA's newest BIM exhibit (E201-2022) exists precisely so a frozen model version can be enumerated as a contract document - the legal system reinventing immutability. A content-addressed model version is more frozen than any drawing ever was: paper can be redlined; a hash cannot. The machinery is arriving; the missing piece is demand. If you write the EIR, stop accepting sheet counts as evidence of information. Demand the information.

I'll admit this one is personal. I spent my PhD arguing that every exchange of design data is a communicative contract - a promise between people about meaning, sequence and reliance - and that our systems should make those contracts explicit and traceable instead of burying them in file transmittals. Fifteen years of standards agree in principle. The industry's actual contracts still refuse to countersign. That's the gap this argument lives in - and it's closable now.