Hi guys,
sorry for this AI generated topic, but Claude found out about this, so I lets give him some credit (trimmed it down a bit, but left the AI dashes in.. 
)
Following up on @rklaschka’s thread #15779 (auto-closed without public resolution): we hit the same defect on the v3 Revit connector and have a precise reproduction plus a likely root cause. Our case has no toposurfaces — regular building model on the UK OSGB36 grid, shared coords set via Acquire Coordinates — so the bug is broader than the topo-tool theory.
The problem in plain terms
When we publish from Revit with Reference Point = Survey, the building comes out the right distance from the OSGB grid origin, but rotated to the wrong place.
The numbers
Survey Point in Revit, real OSGB coords (west London):
N +179 383.585 m, E +510 914.175 m
Same model, Survey-publish, an actual element near the SP:
start = (122 479.56, 527 504.09, 62.45) m
referencePointTransform rotation column-0 = [0.876358, −0.481660, 0] → angleToTrueNorth ≈ 28.794°.
| Distance from OSGB origin | Bearing atan2(N,E) |
|
|---|---|---|
| Real (Revit) | 541 490 m | 19.346° |
| Speckle | 541 537 m | 76.928° |
| Δ | only 46 m apart | −57.582° ≈ −2 × angleToTrueNorth |
So the bearing is off by exactly twice the project’s angle to true north. To check: rotating Speckle’s output by −2θ around the OSGB origin lands within 72 m of the real SP — and that 72 m is consistent with the actual in-model distance from the SP to that specific beam.
Where the bug likely lives
In ToSpeckleSettingsManager.cs (Survey branch in speckle-sharp-connectors):
translation = Transform.CreateTranslation(surveyPoint.Position);
rotation = Transform.CreateRotation(XYZ.BasisZ, angleToTrueNorth);
matrix = translation.Multiply(rotation); // T · R
Our guess: the connector both (a) rotates the geometry by R before publishing it, and (b) stamps R into the published matrix as well. So anyone applying the matrix downstream rotates a second time, ending up with R · R = R² total. One rotation too many.
That also matches @rklaschka’s observation that survey-publish models “don’t locate correctly… but do locate relative to each other”: a single rotation about the origin preserves every distance and angle inside the model — only its placement on Earth breaks.