The IDS editor already offered the values a model holds for a property (the last post). Today attribute rows and classification rows get the same treatment. Open the model you are writing for, add a requirement on Name, ObjectType, Tag or PredefinedType, and the row lists the values the model's objects hold, with Allow exactly these values to turn them into a one-of list. A classification row lists the codes the model uses in that system, and names the systems the model actually uses when the one you typed isn't among them.
What the codes turned up
Of the four public exports we test the editor on (three HHS Office models and the Schultz Residence, from opensourceBIM/IFC-files), only Schultz carries a classification: Uniformat, on 210 elements and 134 type objects. But 93 of its 134 classification references look like this:
IFCCLASSIFICATIONREFERENCE('http://www.csiorg.net/uniformat','','',#739);
The system is there and the code is an empty string. 143 of the 210 classified elements (68%) have no Uniformat code.
An IDS requirement written the obvious way, "these elements must be classified in Uniformat", passes all 210. A coordinator who writes it sees green and believes the model is coded. The checker is right: the system is present. The mistake is in the requirement, and the author can't see it without reading the file.
So the classification row now says so. On the windows: "The model classifies 25 of 25 in Uniformat. 11 of them have a Uniformat reference with no code: a requirement that names only the system passes them, and one that asks for a code fails them." Next to it, Require a code writes the value as a .+ pattern, which fails exactly the uncoded objects. Attributes get the same warning: 68 of Schultz's 146 building element proxies have Name and ObjectType written as empty strings, and IDS fails an empty attribute on any check of it, even an "if present" one.
Does an offer pass the model it came from?
We built a requirement from every offer and checked it against its own model with the same checker the site runs (eval/model-attr-codes.mjs):
- 243 of 298 attribute rows offered a one-of list (the rest hold more than 50 different values, mostly a Name or Tag per object). All 242 rows with no empty objects pass. The one with empties fails on exactly those 68 objects.
- Of Schultz's 12 class and system rows, the 2 that are fully coded pass and 8 fail on exactly the uncoded objects the row warns about. The other 2 (a duct fitting and its type) have no code at all, so they offer nothing.
The values come from the checker's own reader, so the editor offers exactly what a check will compare. Numeric attributes such as a door's OverallHeight get a one-of list but not a range yet. That needs the facet written with a numeric base, and it's next.
Everything runs in the browser. The model isn't uploaded. The write-up with every number is eval/model-attr-codes-2026-11-17.md.