Skip to main content

Inspect, validate, and export

A successful kernel call is not enough to prove that a CAD artifact is correct. A dependable workflow records geometric evidence, checks domain invariants, writes the required formats, and verifies the files that were created.

Inspect a shape​

Core Shape properties are calculated in the native session:

evidence = {
    "kind": part.kind,
    "volume_mm3": part.volume,
    "area_mm2": part.area,
    "center_of_mass_mm": part.center_of_mass,
    "bbox_mm": part.bbox,
    "topology": part.topology,
}

Use part.describe() for a JSON-safe aggregate or part.describe(detail="mesh") when a mesh is explicitly needed. Avoid the mesh detail mode for routine checks because it produces substantially more data.

Validate geometry and intent​

Shape.validate() performs inexpensive backend-neutral checks and returns an OperationReport:

report = part.validate()
if not report.ok:
    raise RuntimeError(report.to_dict())

Then check the expectations specific to the part:

expected_bbox = (0, 0, 0, 80, 50, 8)

if part.topology.get("solids") != 1:
    raise RuntimeError("delivery must contain exactly one solid")

if part.volume <= 0:
    raise RuntimeError("delivery solid must have positive volume")

for actual, expected in zip(part.bbox, expected_bbox):
    if abs(actual - expected) > 0.1:
        raise RuntimeError(f"unexpected bounds: {part.bbox}")

Kernel validity and design intent are separate claims; check both.

Preflight agent-authored operations​

Model.preflight() catches common mistakes before invoking geometry:

check = model.preflight("fillet", part, 2.0, edges=[0, 4, 99])
if not check.ok:
    for diagnostic in check.diagnostics:
        print(diagnostic.code, diagnostic.message, diagnostic.hint)

Current checks include session mismatches, invalid edge indices, and non-positive shell thickness. Model.apply() combines preflight, execution, and structured failure reporting for agent loops.

Export artifacts​

from pathlib import Path


output = Path("artifacts/mounting_plate")
output.mkdir(parents=True, exist_ok=True)

part.export_step(str(output / "mounting_plate.step"))
part.export_stl(str(output / "mounting_plate.stl"), binary=True)
part.export_preview_glb(str(output / "mounting_plate.glb"), deflection=0.25)

Mesh deflection controls preview tessellation, not the underlying exact BREP. A smaller value usually produces more triangles.

Verify delivery files​

At minimum, confirm that each required file exists and is non-empty:

for path in output.iterdir():
    if path.stat().st_size == 0:
        raise RuntimeError(f"empty artifact: {path}")

For higher-assurance delivery:

  1. Re-import the STEP file through Model.import_step().
  2. Compare volume, bounds, topology, or BREP inspection results with the source shape.
  3. Render orthographic views and inspect them visually.
  4. Store validation evidence in a JSON report beside the artifact.
  5. Record hashes when artifacts move between systems.

The public cadflow.inspection and cadflow.inspect.brep surfaces provide detailed STEP/BREP inspection and comparison workflows. They are useful when exact delivery parity matters more than a simple round trip.

artifacts/mounting_plate/
├── mounting_plate.step
├── mounting_plate.stl
├── mounting_plate.glb
├── mounting_plate.png
└── report.json

Keep the model script and report in version control when reproducibility matters. Large generated geometry files can use release assets or external artifact storage.