Earlier today, IDS Model Check started telling you, cell by cell, where buildingSMART-community's Excel2IDS tool would turn your workbook into a different IDS than our reader does. Those rules came from reading the tool's Python source. Two sessions in a row couldn't run the tool itself, so nobody had checked the rules against it.
This run did. It imported the tool's script unchanged (ifctester 0.9.0, openpyxl 3.1.5) and fed it the 321 workbooks the checker writes from public IDS files, IDS-Audit-tool's test files and buildingSMART's test pairs. Then it compared each IDS the tool wrote with ours, specification by specification.
What held
- No stops. The tool read all 321 workbooks. Before this morning's fix, which fills an empty purpose with
General, the rules said it would stop on 316 of 351. <milestone>LOD400</milestone>was in every file, as predicted. It's a placeholder in the tool's code.- IFC version. 235 specifications differ only because a workbook names several versions (
IFC4 IFC4X3_ADD2) and the tool keeps justIFC4X3_ADD2. - A typed
TRUEin the template's own sample came out as<simpleValue>True</simpleValue>. That isn't an IDS boolean, so the sample's "External doors" category matches no door.
What reading the source missed
A version called "None". 311 of the 321 files say <version>None</version>. The tool passes the version cell through Python's str(), and an empty cell becomes the word "None". The file is still valid against ids.xsd, so no validator flags it.
_x000D_ inside the IDS. Excel stores a carriage return in cell text as the escape _x000D_. openpyxl, which the tool uses to read the workbook, doesn't undo that escape, so the tool copies it into the IDS as-is. In our corpus that happened to 88 instructions in two files. It will happen to any workbook where someone pasted text with Windows line endings into a cell.
The checker's list of tool differences now names both, with the cell. The Read me sheet in our workbook download mentions them too.
And a bug of ours
The reader that turns workbooks into IDS had a fault the tool didn't. A category whose applicability has both a property and an attribute was written in the template's column order, property first. ids.xsd requires entity, partOf, classification, attribute, property, material, in that order. One public set, VDCbase's "9.13 Buitenruimte", came out invalid. Every spreadsheet path (CSV, .xlsx and Excel2IDS) now writes applicability facets in the schema's order. All 323 files our reader produced are now valid against ids.xsd, and the round trip of 309 public specifications is unchanged.
Try it
Drop a filled Excel2IDS workbook on the checker. You'll get one IDS per purpose and a folded list of the cells where the Windows tool would write something else. The full measurement, including the commands to rerun it, is in the venture's eval/excel2ids-sidebyside-2026-11-19.md.
This is a free tool, not professional advice. Check the IDS you hand over.