An IDS file is a list of things an IFC model must contain: every wall has a Pset_WallCommon.Reference, every stair has NumberOfRiser. Each property requirement can also name a data type. The IDS editor fills it in from buildingSMART's property set templates. For Reference in IFC2X3, the template says IFCIDENTIFIER.
Revit-era exports write it as IFCLABEL.
We had already counted how often a real export disagrees with the template. On four public exports from opensourceBIM's IFC files (the HHS office's architect, construction and MEP models and the Schultz Residence), 21 of 122 standard properties use another type. Fourteen of the 21 are Reference as a label. Four are stair counts written as IFCINTEGER instead of IFCCOUNTMEASURE. One is a railing height written as a bare IFCREAL. Until now the editor pointed this out on the row ("a check with this dataType will fail them") and offered no fix.
What it costs
Today we wrote one specification for each of those 21 properties, exactly as the editor writes it, and ran each against the model it came from:
- 0 of 21 specifications pass. 966 of 966 applicable objects fail.
- Without the data type, 14 of 21 pass, and 175 objects fail.
So 791 objects failed on a technicality. The 175 that still fail are real gaps: 84 building-element proxies with no Pset_BuildingElementProxyCommon at all, 42 coverings without Finish, 33 spaces whose Reference is empty, and 16 failures on stairs and slabs missing the set. Those are the findings a coordinator wants. The other 791 are noise that hides them.
The obvious fix, writing the model's type (IFCLABEL), is worse. It contradicts the standard template, so the editor's audit flags it as an error, and so does buildingSMART's own IDS-Audit-tool.
Accept any data type
A property facet with no dataType checks the property whatever type it holds. If it asks only that the property be present, the audit has nothing to warn about. So rows like these now carry an Accept any data type button. It removes the dataType and keeps a text value if the row had one. The neighbouring Write dataType IFCIDENTIFIER button puts it back. That round trip works for all 21 rows.
We ran every file through IfcTester 0.8.0 as well. It gave the same verdict as this checker on every specification in each variant (21, 21 and 19): the template's type, no type, and no type with the model's own values as a one-of list.
The one that disagreed
The first run did not agree. One property, Pset_StairFlightCommon.NosingLength, is written as IFCPOSITIVELENGTHMEASURE(29.99999999999999) in a model measured in millimetres. The editor offers values in SI units, so it offered 0.03. With no data type, this checker converted the unit and passed 8 of 8 stair flights. IfcTester compared the raw number and failed all 8.
Neither tool is clearly wrong, because IDS 1.0 expects a data type whenever a value is given, and the editor's audit already warns about a value without one. So the editor no longer offers that combination. On a model type that is a real number, dropping the data type keeps only the presence check, and an opened row that already asks for a value gets a warning. Integers and labels compared the same way in both tools.
The full tables, the commands and the IfcTester versions that install here are in the project's evaluation note. The editor runs in the browser, and models and IDS files stay on your machine.
This is a free, open-source tool (MIT). Its checks are an aid, not a certification: confirm requirements with your project's information manager and the standard itself.