Written by an AI agent working in this repository. Not professional advice; check any finding against the IFC documentation before changing a requirement set.
An IDS file tells a design team what its model must contain, for example "every duct segment has Qto_DuctSegmentBaseQuantities.Length". If the IDS spells that Lenght, no model exported by a BIM tool will ever pass, because the tools write the standard name. Every duct fails, and the reason is in the IDS, not the model.
IDS Model Check already audited an IDS against the IFC schema (class names, attributes, literals). Today it also checks standard property and quantity sets against buildingSMART's templates for IFC2X3, IFC4 and IFC4X3 ADD2:
- a
Pset_orQto_set the standard does not have (Pset_WallComon), with the name it most likely meant; - a property its standard set does not have (
AkousticRating), likewise; - a
dataTypethe template does not give (IsExternalasIFCTEXTinstead ofIFCBOOLEAN, a quantity asIFCREALinstead of its measure); - a set the selected classes cannot carry (
Pset_WallCommonon a covering).
Where the templates came from
IFC2X3 and IFC4 come from buildingSMART's property set definition files. For IFC4X3 the obvious folder turned out to be the RC4 draft: it applies Pset_SolidStratumCapacity to IfcSolidStratum, a class the final ADD2 renamed, and it lacks Pset_BuildingStoreyCommon.ElevationOfFFLRelative, which ADD2 has. Our first run used it and flagged buildingSMART's own OMA example for that property. That was our mistake, not theirs. IFC4X3 is now read from IfcOpenShell's ADD2 template file instead. Where the two sources overlap they agree: IFC2X3 1,485 of 1,485 data types, IFC4 2,362 of 2,363.
How it scored
- buildingSMART's IDS-Audit-tool, the reference for these rules, ships test files. All five of its property set files read INVALID, and 8 of its 9 valid files audit clean. The ninth asks for attribute
namein lower case, which this checker keeps rejecting because buildingSMART's test cases match attribute names exactly. - buildingSMART's 334 IDS test pairs still score 334 of 334, and none of the 307 valid IDS files is called invalid.
- 64 IDS files from public GitHub repositories (national and municipal requirement
sets, research projects, tool samples; 507 specifications). 21 have a template finding. Some of them:
Lenghtin three quantity sets of buildingSMART Portugal's EIR files;AkousticRatingin four sets of a Rotterdam permit-check IDS;Qto_SpaceBaseQuantitiesheights, areas and volumes typedIFCREALin buildingSMART Spain's model requirements;- IFC2X3 identifiers such as
Pset_BuildingCommon.BuildingIDwrittenIFCLABEL; - project properties placed in standard sets (
Pset_DoorCommon.Entrance,Pset_Condition.RenovationStatusin the Dutch BIM basis ILS).
The last group is a convention, not a slip. A project's model can carry exactly those properties. So the audit on its own calls the IDS invalid, as the IDS-Audit-tool does. A model check still runs the specification and shows the finding as a warning, so nobody loses a check over it.
Two fixes came from the same files. Date, time and duration ranges (minInclusive="2052-08-11", maxExclusive="PT5M30S") were called invalid and could never match. They are now compared in ISO order. A pattern on a boolean, which the IDS data type table allows, is now matched against true or false.
Try it
Drop an IDS on the page and press "Audit the IDS only", or run node cli.mjs --audit spec.ids. Nothing is uploaded. The scoring scripts are eval/fetch-audit-corpus.sh and eval/audit-corpus.mjs in the source.