Two systems, two state models. TopoExport holds site data: geometry derived from geographic sources, positioned on the earth, changing only when the earth's records change. GenAnywhere holds design state: scenes authored to express a proposal, changing every time someone has a better idea. The integration problem between them is not "which tool wins." It is a data-handling problem with a specific shape: mutable design state consuming immutable reference geometry through a transient import boundary.

This post treats it that way.

TL;DR

Dimension TopoExport GenAnywhere
State Immutable site reference Mutable design proposal
Source Geographic datasets Authored scenes
Layers Buildings, roads, contour lines Composition, massing, presentation
Outputs PDF, SVG, OBJ, STL, DXF, IFC Composed scenes and presentations
Coverage France, United States, Japan Location-independent
Export cap 50 km² per export N/A
Failure mode Over-exporting; unit/origin mismatch Treating context as editable design

The rule that prevents most incidents: reference geometry is read-only inside the design environment. Enforce it socially first, procedurally second.

The format matrix

TopoExport's output list is not arbitrary. Each format maps to a downstream consumer, and choosing the wrong one costs either fidelity or time:

  • PDF — presentation and markup. Coordinates are gone; this is for humans, not for the pipeline.
  • SVG — vector linework that survives into layout tools. Good for plan diagrams, poor for anything dimensional.
  • DXF — the CAD interchange default. Survives into AutoCAD-family tools with layers intact, which matters when someone downstream needs to switch off the road layer or restyle contours.
  • OBJ — the visualization workhorse. Geometry plus UVs, broadly readable, no scene semantics. This is the format most teams want when the destination is a rendering or authoring environment.
  • STL — triangle soup for fabrication and quick previews. No layers, no materials, no units metadata worth trusting.
  • IFC — BIM coordination. Heavy, semantically rich, and almost always more than a design iteration needs.

The practical default for a design handoff is OBJ for geometry, DXF when a 2D base is required, and IFC only when the receiving model actually participates in BIM coordination. Exporting IFC "to be safe" produces the archive trap described below.

Where the pipeline breaks: units, origin, scale

Three failures account for nearly every bad handoff.

1. Unit assumption mismatch. A GIS export may carry meters while the receiving environment interprets units as feet. The model imports, looks plausible, and is wrong by a factor of 3.28. Nothing errors. The fix is a calibration step: after import, measure a known dimension — a standard road width, a building face visible in the source data — and compare against the recorded value before any design work begins.

2. Projection origin offset. Geographic data lives in a projected coordinate system. Some importers honor that origin; others drop geometry at the local origin. The result is a model that appears correct but sits hundreds of kilometers from where the renderer's sun angle, shadow study, or terrain reference expects it. Record the export location and verify the first import against it.

3. Scale-by-bounding-box. A common automatic normalization fits imported geometry into a viewport-sized box. It silently rescales everything, including contour intervals, and invalidates any dimensional comparison downstream. If an importer offers "fit to view" on import, disable it.

A minimal acceptance check, run once per project:

1. Import reference geometry 2. Measure one known distance (road width, parcel edge) 3. Compare to source-data value; record pass/fail 4. Verify orientation against a known bearing (north arrow or street grid) 5. Freeze the reference layer; revoke edit access 

If step 2 or 4 fails, stop. Every downstream minute is wasted until the boundary is clean.

The state model, stated plainly

Once the reference layer is frozen, the project has exactly one mutable thing: the design state. GenAnywhere is where that state lives — the scene, the massing, the sequence, the presentation framing, and every revision that follows. Its correctness criterion is not geometric fidelity to the source data; it is whether the proposal communicates and survives review.

That separation has a useful engineering property: design revisions never invalidate context, and context updates never invalidate design. A revised massing does not require a re-export. A refreshed context export does not require rebuilding the proposal, provided the reference layer swap is done deliberately and the units check is repeated.

The teams that struggle are the ones that keep the boundary fuzzy. They edit the imported context to make a view work, then lose track of which geometry is fact and which is wish. Three weeks later, a reviewer asks why the neighboring building is 20% wider than the survey, and nobody can answer.

Revision semantics

Design state wants versioning; site data does not. A workable convention:

  • Context versions are dated exports. v1 is captured at project start and normally never changes. A v2 exists only if the source data materially changed or the export area was wrong.
  • Design versions are proposal snapshots — v1 through vN — named by intent, not by date. "Entry sequence test," "reduced massing," "committee option B."
  • The pairing is explicit. Every design snapshot records which context version it was authored against. This is the equivalent of a lockfile, and it costs one line in the project log.

With that convention, "which version did we show the client" becomes a lookup instead of an argument.

Choosing an export area as a design decision

The 50 km² cap is a permission, not a recommendation. Export area should be derived from the design question:

  • Massing study — parcel plus setbacks plus immediate neighbors. Buildings and roads layers only; contours optional.
  • Grading-sensitive scheme — add contours, and verify the interval is fine enough to read the slope that matters.
  • Context/presentation study — widen the area for skyline and approach views, but keep the dense geometry out of the working scene; stage it as a separate background object.

The guiding principle is the same one that governs the rest of the pipeline: import at the resolution the question needs. Dense context is not more correct; it is just slower.

When the proposal becomes the deliverable

The final handoff of most site projects is not the context model. It is the proposal — the scenes, the comparisons, the version that got approved. That artifact is design state, authored in GenAnywhere, and it should never be entangled with the reference geometry beyond the deliberate freeze boundary.

TopoExport's contribution is that the ground under the proposal is real: georeferenced, correctly scaled, and available in the formats the rest of the stack already reads. GenAnywhere's contribution is everything above the ground. A pipeline that keeps them distinct — immutable reference, mutable design, one verified handoff — is boring in exactly the way good pipelines are.