IDS Model Check checks IFC building models against buildingSMART IDS files. We had compared it with IfcOpenShell's IfcTester before, but only with IDS files we wrote ourselves. This time we used the ones other people publish: 65 IDS files from public GitHub repositories (national requirement sets, research projects, tool samples), run against four real exports from the opensourceBIM/IFC-files collection (three HHS Office models and OpeningDesign's Schultz Residence, all IFC2X3).
What matched
IfcTester 0.9.0 could not open 12 of the 65 files: eight fail its schema validation, two use an old or misspelled namespace, and two are not well-formed XML. On the other 53, the two checkers produced 1,736 specification results per tool, and 1,661 were identical: the same verdict, the same number of applicable elements, the same failing elements by id.
All 75 differences come from three causes:
- 50: what
IFCWALLmeans in IFC2X3. Revit's IFC2X3 exports write most walls asIfcWallStandardCase. The IDS user manual says entity facets have "no automatic inheritance", soIFCWALLdoes not take them. IfcTester also matches any occurrence whose type object is anIfcWallType, so on the Schultz Residence it counts 105 walls where we count 4. - 16: a prohibited specification with requirements. Four public files forbid building element proxies this way. buildingSMART's own test suite says such an IDS is invalid; IfcTester runs it anyway.
- 9:
Trueon a boolean with no data type. XSD writes booleans astrueorfalse. IfcTester acceptsTrue; we don't, and our audit already warns that checkers disagree.
In the first two cases we follow the standard's text and test suite. But that doesn't make our reports useful.
The gap both checkers share
The HHS models hold 112 IfcWallStandardCase walls with no type object. There IfcTester's type rule doesn't apply, so the two tools agree: on 51 specification results per model, both report the same thing, and neither has looked at most of the walls. A requirement that walls carry a fire rating can pass on a model where almost none of them do. Comparing two checkers can't catch a mistake they both make. We only found this one because the Schultz model made them disagree.
So the report now says what an entity facet left out:
[FAIL] Wall Fire Rating Check (optional)
applies to: entity "IFCWALL" -> 4 element(s)
warning: not checked: 101 element(s) of subclasses the entity facet does not name:
101 IFCWALLSTANDARDCASE (a subclass of IFCWALL). IDS entity names do not include
subclasses; add these class names to the entity facet to check them
The verdict stays the one the standard gives. Across the 260 model and IDS pairs, 165 of 2,028 specification results now carry this warning, and 76 of them are a PASS that left elements unchecked. The fix in the IDS is one line: list IFCWALLSTANDARDCASE next to IFCWALL.
The invalid prohibited specifications got the same treatment. They still read INVALID, but they now add "85 element(s) in this model match its applicability and would fail it once the requirements are removed". Before, an invalid result with "0 applicable" looked like a clean model.
Check it yourself
The scripts are in the repository: eval/corpus-crosscheck.mjs, eval/corpus_crosscheck_ifctester.py and eval/corpus_crosscheck_compare.py, with the full write-up in eval/corpus-crosscheck-2026-11-17.md. The checker runs in your browser and your model never leaves your machine. It is free, open source (MIT), and not a substitute for a BIM coordinator's review.