How to Review STL Revisions Without Losing Intended Geometry

A revision review asks whether the changes are the ones you intended. Start in the original coordinate frame, confirm unit assumptions for both files, and use a known control before interpreting a different export. File bytes, triangle counts and bounding boxes answer different questions from surface geometry.

A reproducible control and a known change

Open the STL comparison tool and run the unchanged retriangulation control. Both models describe a 20 mm cube, but one face uses the other diagonal. Triangle connectivity changes while the surface stays the same. The bidirectional point-to-triangle samples should be effectively zero apart from floating-point noise. A nearest-vertex method could give misleading distances here.

Then run the widened-cube example: before is 20 × 20 × 20 mm, after is 24 × 20 × 20 mm in the same frame. The moved outside face is 4 mm away from the old face. Much of the original surface remains coincident; one average value therefore cannot describe every local change.

Choose the review frame deliberately

Situation	First review
Repair exported in the same coordinate frame	Original coordinates; confirm that meaningful holes and ribs survive
Export has a deliberate coordinate move	Keep the move visible; compare with the expected placement
Only placement should be ignored	Explicit centre translation; record the applied offset and inspect shape again

Centre translation is not automatic rigid registration. It can hide a purposeful placement change and does not rotate or rescale a model. Do not normalise both models to the same size to make an overlay look tidy.

Interpret a sampled result

The tool distributes finite samples over surface area and measures nearest points on target triangles in both directions. It displays the sample count, tolerance, sampled maximum and samples over tolerance. A value below tolerance is evidence about those points, not proof of geometric equality or a certified maximum error. Small features can fall between samples.

For a repair handoff, retain the original file, revised file, unit assumptions, selected alignment and report. Run batch topology checks on both, then verify the revised copy in your slicer. No Boolean difference, CAD tolerance certification or physical testing is implied.

Technical basis

The numerical controls are owned software fixtures generated from cube vertices and explicitly changed face indices. The implementation uses triangle queries from three-mesh-bvh; method details and tests are recorded with the application. These examples are not scanned physical parts.