An IDS file (buildingSMART's Information Delivery Specification) tells a BIM model what it must contain: every door needs a fire rating, every wall says whether it is load-bearing. Before any tool can check a model against it, the IDS file itself has to be valid XML against the IDS schema, ids.xsd. Tools that validate first, IfcTester among them, refuse a file that breaks it, whole, before reading a single rule.
IDS Model Check already audited what an IDS means: IFC class names that do not exist, property names one letter off the standard (AkousticRating), data types the template does not give. It did not check the file's structure. Its reader is lenient, so a file another tool refuses could still read as VALID on our page. That gap was in the README's Limitations until today.
What we found
We ran the 65 IDS files we found published in public GitHub repositories through IfcTester's schema validation. 12 break ids.xsd:
- IDS 0.9 habits.
minOccurs/maxOccurson<specification>(IDS 1.0 moved them to<applicability>), andifcVersion="IFC4X3"where 1.0 saysIFC4X3_ADD2. - The schema's address as the namespace.
xmlns=".../IDS/1.0/ids.xsd"instead ofxmlns="http://standards.buildingsmart.org/IDS"; three files. - Generated files. An
<author>of "AI Assistant" (the schema wants an e-mail address), and<n>written for<name>. - Plain XML faults. A bare
&in a title; an element left unclosed.
None of these change what the author meant, and our checker read most of them correctly. But a file the next tool in a project's chain refuses is a broken handover, and the author would rather hear it from the first tool.
What changed
The audit (the "Audit the IDS only" button, or node cli.mjs --audit) now checks element order, allowed and required attributes, the fixed values of cardinality, relation, ifcVersion and dataType, value content (<simpleValue> or <xs:restriction>, XSD 1.0 facets only), the info block and the root namespace, and names the line and what to write instead:
[ERROR] Doors need width of at least 1.05m / line 8: ifcVersion="IFC2X3 IFC4 IFC4X3" on <specification>: "IFC4X3" is not one of IFC2X3, IFC4, IFC4X3_ADD2 (IDS 1.0 names IFC 4.3 "IFC4X3_ADD2") (ids.xsd)
When a model is checked, such a specification still runs, with a warning, so no earlier verdict changes.
How we know it agrees
We compared the valid-or-invalid verdict with IfcTester's own ids.xsd validation on every file: the 145 public and IDS-Audit-tool files, then 2,935 copies of the valid ones with one random fault each (siblings swapped, an element dropped or renamed, an attribute misplaced or misspelt, a wrong fixed value, stray text). The first two mutated sets found three gaps, which we fixed; the last set of 1,129 was run once, after the rules were final, and agreed on all of them. Overall: 3,080 of 3,080 the same. The method, the tables and the commands are in eval/ids-structure-2026-11-18.md.
The checker is free, runs in the browser with nothing uploaded, and is MIT licensed. If you write IDS files for a project, drop one on the page and press Audit the IDS only.