An IDS specification usually starts with a class: "every IfcDoor must...". It does not have to. buildingSMART's IDS-Audit-tool ships test files whose specifications name only an attribute ("anything with a CoordinateSpaceDimension must have it equal to 3"). Our IDS editor opens most of those as editable rows. Earlier today 20 number lists were still held as written, and the note on them only said the page "does not type" them.
So I looked up each of the 20 attributes in the IFC2X3, IFC4 and IFC4X3 schemas: every class that declares it, its type, and whether it is stored, derived or a list. They fell into three groups.
Derived on a subclass (5 lists). CoordinateSpaceDimension and Precision are stored numbers on IfcGeometricRepresentationContext. Its subclass IfcGeometricRepresentationSubContext redeclares both as derived, so they are written * in the file. The editor took that as "typed differently" and gave up. They are the same number everywhere they are stored, so they now open as number rows, with a note naming the subclass. We checked that note on buildingSMART's IFC 4.3 HVAC sample. A specification whose applicability is CoordinateSpaceDimension itself leaves the subcontext out and passes. One that selects anything with a ContextType picks up the subcontext, and it fails.
A count written as an integer (3 lists). IfcCountMeasure is an EXPRESS NUMBER in IFC2X3 and IFC4, so the editor writes a new BarCount or CountValue check on xs:double. The test files write xs:integer. A checker compares each value with the attribute's own type, so 27 means the same on either base. The row now keeps the file's xs:integer and asks for whole numbers.
Lists (12 lists, all still held). TexCoordsList, DayComponent, ColourIndex, LuminousIntensity and three more are lists on every class that has them. buildingSMART's own test case says a value check on a list always fails. Turning these into editable rows would only make a broken check easier to write. They stay held, and the audit now says why: "TexCoordsList is a list on every class that has it (IfcTextureVertexList); IDS cannot compare it with a value; every object this checks fails it". It is a warning, not an error, because the IDS-Audit-tool calls these files valid. The audit's errors did not change on any file in our corpora.
Result: 117 of the 129 number lists in the IDS-Audit-tool valid files open as rows (was 109), and 90.7% of those files' facets can be edited. The full table and the commands are in the evaluation.
One caveat: none of the 64 public IDS files we track writes a check like this. Only buildingSMART's synthetic test files do. If you write IDS without naming a class, try the IDS editor and tell us whether these notes help.
This is a free, open-source tool. It checks files against the IDS 1.0 rules as we read them and is not a substitute for your project's own review.