Checking bridges, roads and ducts against an IDS: 348 of 348 results match IfcTester

IDS Model Check against IfcTester on buildingSMART's twelve IFC 4.3 and HVAC sample models: 348 of 348 results identical, plus an IFC2X3 gap.

Until today, the claim that IDS Model Check "gives the same answers as IfcTester" rested on five building models: walls, slabs, doors, a Revit house. IFC 4.3 is mostly about infrastructure, though, and infrastructure models don't look like buildings. A bridge is an IfcBridge built from IfcBridgeParts (piers, abutments, decks), a road is an IfcRoad of IfcRoadParts, and a track's sleepers and rails sit in an IfcRailwayPart of type TRACK. Building services add another wrinkle: the type of a duct segment often lives only on its type object.

What was run

buildingSMART publishes a "Simple-Scene" set of sample models under CC BY 4.0: ten in IFC 4.3 (architecture, HVAC, landscaping, structure, bridge, electrical, plumbing, rail, road and the river landscape around them), plus the HVAC model in IFC4 and IFC2X3. I wrote a new starter IDS for them, infrastructure-and-mep.ids.xml, with 18 specifications. A few examples:

  • bridge parts must be a named kind (pier, deck, abutment...), so NOTDEFINED fails;
  • beams, columns and footings must sit somewhere under an IfcBridge, through any chain of containment and aggregation;
  • bridge beams must be 1 m to 20 m long, where the model writes 4000 in millimetres;
  • track elements and rails must sit in a TRACK railway part;
  • duct segments must be RIGIDSEGMENT, a value written only on IfcDuctSegmentType.

Then I ran each model against this IDS and the two older starters, once through this checker and once through IfcOpenShell's IfcTester 0.8.0, and compared each specification's verdict, how many elements it applied to, and exactly which elements failed.

Result

348 of 348 specification results were identical, element by element. That covered 70 track elements, 32 bridge members reached through two levels of the spatial tree, and the millimetre-to-metre conversion: both tools pass the three beams that give a 4000 mm length and fail the five that give none. Two tools agreeing can still both be wrong, so I checked the counts against the raw data with IfcOpenShell as well. The bridge really has 2 NOTDEFINED parts, the road really has 2 parking bays, and the plumbing model has no classification at all, which is why all 26 of its services fail the "classified" check.

The only differences were on the IFC2X3 copy of the HVAC model. IFC2X3 has no air-terminal or duct-segment occurrence classes, so the model writes generic IfcFlowTerminals and an IfcFlowSegment, typed by IfcAirTerminalType and IfcDuctSegmentType. This checker maps those to IFCAIRTERMINAL and IFCDUCTSEGMENT, as buildingSMART's IFC2X3 mapping table says to. IfcTester finds nothing to apply the specifications to, and reports them as passing. For someone checking an older MEP export, "passed, nothing applicable" looks clean when it isn't: three elements there have no classification.

One more thing came up while writing the IDS. My first draft used an empty <classification/> to mean "any system". IfcTester refused to load the whole file, because ids.xsd requires a system. This checker marked only that specification invalid, ran the other 17, and suggested writing a .* pattern instead. The sample now does that.

Limits

These are buildingSMART's own demo models: small, tidy, and all from one exporter. They widen the classes covered, not the exporters. The next step is a real IFC 4.3 export from an infrastructure tool, and a Revit MEP model with systems and ports. The full output and the steps to reproduce it are in eval/infra-mep-crosscheck-2026-11-17.md. A check result reads the model against the IDS. It is not a design review or professional advice.

All writing · More in Tools and projects · All tools and research