A 3D model opening successfully does not prove that geometry, units, assembly structure, material information or manufacturing intent survived transfer. Exchange validation asks whether an export from the source system means the same thing after import into the target system for the intended use. This matters when a file moves among CAD, repair software, slicer, service provider and measurement system.

1. Define the purpose first

A visual mock-up, fitted enclosure, loaded machine part and controlled industrial component require different evidence. List critical dimensions, permitted geometric deviation, material and colour information, assembly structure, PMI, coordinate system and downstream operation. Choose the format from these requirements, not popularity.

2. STL: a simple surface with little context

STL describes a surface with triangles. The Library of Congress format description notes its lack of rich metadata and the possibility that tessellation creates gaps or open loops. Units are not standardised in the file, so the same coordinates interpreted as millimetres or inches produce radically different parts.

Record units, chordal and angular tolerances, and source-CAD version. Check that the mesh is closed and manifold, normals are consistent, and there are no self-intersections, duplicate triangles or zero-area faces. A coarse mesh changes surfaces and dimensions; an excessively fine mesh creates large files and unstable processing without necessarily improving printed accuracy.

3. 3MF: a manufacturing package, not automatic truth

The 3MF Core Specification defines units, mesh topology, objects, transforms and metadata. Extensions can carry materials, colour, beam lattices, production identifiers and protected content. The suite is also published as ISO/IEC 25422:2025.

More information does not mean every application supports it. Identify required and recommended extensions before import; the specification says a consumer must not process a model when it lacks a required extension. Test whether units, instances, transforms, material assignments and metadata survive both import and resave.

4. STEP: B-rep and PMI, with implementation differences

STEP belongs to ISO 10303 and may exchange solids, assemblies and, with AP242, machine-readable product and manufacturing information. NIST's STEP File Analyzer checks syntax, references, semantic and graphic PMI, and validation properties.

After a successful import, verify body count, assembly hierarchy, units, colours, names, geometric and dimensional tolerances, and validation properties from the source system. Volume, area and centre of mass are useful quick checks but insufficient: models with equal volume may differ at a critical hole or surface.

5. Golden models and round-trip tests

Create a representative reference set containing planes and cylinders, a small hole, thin wall, freeform surface, pattern, multiple bodies and, where needed, assembly, colour or PMI. Keep a versioned golden source model and a controlled export recipe.

Measure the same characteristics in the target system using absolute and relative deviations, not screenshots. Record alignment method for dense surface comparison: best-fit alignment can hide origin or coordinate errors, so production evaluation may require datum-based registration. An import-save-reimport round trip exposes information lost in a second generation.

6. Acceptance rules with measurement uncertainty

Set acceptance limits and decision rules before transfer. If a critical dimension has a ±0.10 mm tolerance, comparison uncertainty cannot be ignored. Near a limit, use a guard band or manual review. Digital-equivalence tolerance should usually be tighter than later print tolerance, otherwise exchange and manufacturing errors accumulate.

7. Versioning, automation and AI agents

Retain source-model ID, exporting application and version, format revision, settings, date and result. For automated or agent-generated models, a command log and reproducible script strengthen evidence but do not replace geometric and functional acceptance. A system update, geometry-kernel change or export-library change should trigger regression testing.

8. Practical acceptance sequence

1. Define requirements and critical characteristics. 2. Select format and export settings. 3. Check syntax, topology and units. 4. Compare source and target geometry. 5. Verify metadata, material, assembly and PMI. 6. Run the actual slicer or CAM downstream. 7. Produce and measure a representative part. 8. Record result, limitations and approver.

Verified fact and editorial conclusion

STL is a mesh format with little self-description, 3MF defines explicit units and an extensible manufacturing package, and STEP can carry B-rep, assembly and PMI. None of those facts proves that a receiving application interpreted a particular file correctly.

The editorial conclusion is that the best format is the one whose required content demonstrably survives the entire local toolchain. Data exchange is not an administrative export click; it is a manufacturing process that needs validation.