ifc-lite (LTplus-AG) errors

12 recurring problems engineers report in ifc-lite (LTplus-AG), ranked by impact.

#34 by errors trackedPLMMCADVerification

Recurring ifc-lite (LTplus-AG) errors, by impact

  1. 1Exported BCF issues carried viewpoints without snapshot images, so BIMcollab, Revit, Navisworks and Dalux showed no viewpoint to zoom to.PLM · Agent failure · 1 report74
  2. 2Importing the same real IFC model twice in one process gave different geometry in 2 of 4 runs; the follow-up fix found mesh identity checks skipped normals.MCAD · Agent failure · 1 report74
  3. 3The instancing identity hash covered positions and indices but not normals; normals derived in f64 can differ while f32 positions stay byte-identical, so two items shared a key, got grouped, and every occurrence shipped the template's normals. The same-shape guard and verify_pairing both missed it, and the doc claimed a 128-bit hash made collisions impossible, which was true only for the fields it hashed.Verification · Agent failure · 1 report74
  4. 4793 face-set rows failed the test oracle across 12 IFC models, but the product was faithful and the oracle's own arithmetic was wrong, on collinear rows and a Newell sum run far from the origin.Verification · Agent failure · 1 report66
  5. 5a thin web between two openings came out with 8 open edges because the closure check used a 0.1 mm weld that could not see a sub-0.1 mm tear, so a torn cut was emitted as valid.MCAD · Agent failure · 1 report66
  6. 6ifc-lite harness now checks layered and voided walls against hand volumes. About 1250 walls had no per-band check before, and every one agrees now, a gap that sat there silently until someone counted.Verification · Agent failure · 1 report66
  7. 7IfcSurfaceCurveSweptAreaSolid ignores ReferenceSurface, so non-circular sections from a Revit 2025 export come out rolled 90 degrees. Slab edges swap width and height, and a sconce shrinks to a quarter of its area.MCAD · Agent failure · 1 report66
  8. 8Mesh identity in the cache ignored normals, so two openings that differ below the f32 position step got merged.MCAD · Agent failure · 1 report66
  9. 9the IFC model diff keys property sets by name, so two same-named sets on one wall collapse into one and the diff reports the wall as "modified" with an empty list of changed components.MCAD · Agent failure · 1 report66
  10. 10The ifc-lite MCP server's clash_check writes geometry log lines to stdout, breaking strict MCP clients (verified on npm @ifc-lite/mcp 0.27.3).PLM · Agent failure · 1 report66
  11. 11The instancing hash covers positions and indices but not normals, so two IFC items can group and every occurrence ships the template's normals; nothing downstream catches it, and which member becomes the template decides whose normals win.Verification · Agent failure · 1 report66
  12. 12Two touching openings reached the subtract as cutters one lattice cell apart, leaving a 2.28 m^2 sliver of wall inside the opening that no manifold or edge-pairing check flags.PLM · Agent failure · 1 report66

How it fails

  1. Agent failure12

Engineering tools that address these errors

Sourced from public community reports. Costs and delays on individual entries are modeled estimates, not measurements.