Published technical report · v0.11 · 22 October 2026
Screening STEP-to-web conversion errors
A mesh can look correct and still miss part of its source. This report documents sampled surface errors, an incomplete STEP import, and what independent comparisons did—and did not—establish.
What the fresh cohort shows
After correcting the import, twelve new files were selected before inspecting their geometry. None had been used in the earlier tests. The table compares the original small sample (old gate), samples around local geometric features (local), and random samples using the same number of distance queries (matched random). A separate evaluation then checks more points on the original exported mesh, called the baseline.
- Requested inputs
- 12
- Measurable sources
- 11
- Sampled baseline passes
- 10
- Observed baseline errors
- 1
The old gate, local probes and query-count-matched random sampling all detect the same violating baseline, x11. The empty upstream file x06 is retained as an invalid input, not replaced or counted as a pass. File identities are disjoint, but all sources come from the same community repository.
| Case | Old gate | Local | Matched random | Separate evaluation | Screen outcome |
|---|---|---|---|---|---|
| x01 | 0.003419 | 0.003419 | 0.003419 | 0.003419 | Sampled pass |
| x02 | 0.002020 | 0.002020 | 0.002020 | 0.002020 | Sampled pass |
| x03 | 0.004973 | 0.004973 | 0.004973 | 0.004973 | Sampled pass |
| x04 | 0.004041 | 0.004041 | 0.004041 | 0.004041 | Sampled pass |
| x05 | 0.001785 | 0.001787 | 0.001787 | 0.001787 | Sampled pass |
| x06 | — | — | — | — | Inconclusive |
| x07 | 0.001683 | 0.001864 | 0.001688 | 0.001755 | Sampled pass |
| x08 | 0.003720 | 0.003729 | 0.003729 | 0.003729 | Sampled pass |
| x09 | 0.016188 | 0.022011 | 0.020538 | 0.020675 | Sampled pass |
| x10 | 0.023594 | 0.031289 | 0.030865 | 0.030676 | Sampled pass |
| x11 | 0.064140 | 0.068194 | 0.066637 | 0.067650 | Observed error |
| x12 | 0.004365 | 0.004392 | 0.004391 | 0.004392 | Sampled pass |
A dash means no valid measurement. A sampled pass is not a continuous error bound or a topology certificate. The 0.05 mm budget is an illustrative experiment setting, not a manufacturing tolerance.
Previously exposed misses
All three are now detected and withheld, but their mesh geometry has not been repaired by the screen. These are development cases, not new generalization evidence.
| Case | Old gate (mm) | Local (mm) | Matched random (mm) |
|---|---|---|---|
| v11 | 0.021619 | 0.062413 | 0.062183 |
| v12 | 0.043503 | 0.087314 | 0.087111 |
| v15 | 0.021490 | 0.059406 | 0.057178 |
When two checks share the same incomplete source
In the initial twenty-file cohort, an independent Gmsh import exposed a disagreement on bearing w20. The earlier CadQuery path used .val(), retaining one of three roots. The amended loader combines all returned .vals() without changing their placements.
Retained partial export
One of three source roots represented. Against the complete source, its sampled maximum is 2.500777 mm.
Rejected by the complete-source screen.
Fresh complete export
All three roots represented. Its sampled maximum is 0.001554 mm.
Retained byte-for-byte after completed checks pass.
These are two exports of one exposed model. The final twelve-file cohort contains no valid multi-root source, so the planned assembly-omission ablation has zero eligible cases. Three separate analytic controls check root retention, placement and order; they do not establish generalization to unseen real assemblies.
How the evidence was checked
- Retain source bytes and record the imported roots, solids and placements.
- Compare a small gate, local triangle/edge/face probes and a random control matched by distance-query count.
- Save detector decisions before a separate 8,192-point-per-direction evaluation.
- Attempt Gmsh at fixed settings and cross-check retained distance witnesses through separate implementations.
- Keep every invalid input, exceeded resource cap and inconclusive measurement in the accounting.
The final cohort includes 864 facewise CAD checks, 862 Trimesh checks and 432 Gmsh import/distance checks, all within the declared agreement thresholds.
What can you check with a fastener’s web model?
A web model helps you identify the head, shank and threaded regions, and understand where a labelled dimension starts and ends. Use it to explain which feature your enquiry concerns. A model that rotates smoothly or displays the expected overall length can still contain a local surface error or omit part of an assembly, as the report demonstrates.
Use the approved drawing, agreed standard and purchase documents to define dimensions, tolerances, thread requirements and acceptance criteria. The study’s 0.05 mm sampling budget is an experiment setting; it is not a manufacturing tolerance or a guarantee for every BYBOLT web model.
For a product example, open the Hex Bolts dimension guide. For an enquiry, send your drawing or specify the product type, material, size and quantity using the fastener enquiry guide. Any dimensions entered in a web guide still require review against your drawing and order requirements.
Read and cite the published report
Shiya Yuan · BYBOLT · 22 October 2026 · Version 0.11. This engineering technical report has not been peer reviewed. The numerical findings remain those of the frozen v0.8/v08b study.
Citation
Yuan, S. (2026). Screening STEP-to-Web Conversion Errors: An Engineering Technical Report on Local Sampling, Source Completeness and Independent Comparisons. Version 0.11. Zenodo. doi:10.5281/zenodo.23554447
The report PDF and corresponding LaTeX source are licensed under CC BY 4.0. The research code, CAD models and complete experiment archive are not included in this public release. File paths in the report refer to retained local evidence, not public downloads.
AI assistance supported implementation, analysis, literature screening and writing; responsibility rests with Shiya Yuan. BYBOLT has a commercial interest in fasteners and its website. No external dedicated funding was received and no additional competing interests are declared.
Prior work and references
Parameterized fasteners, geometric checks, face sampling and assembly validation have established precedents. The contribution here is the documented failure analysis and controlled evidence. Zhang 2011 was checked at abstract level; the Sap/Szabo 2025 full text was not obtained. No claim of a complete novelty review is made.
- FreeCAD community parts-library, pinned revision f5ca01cdcaa2f69b9295a6382719c2d90628780cRepository content is CC BY 3.0; individual part attribution remains necessary.
- Research on the Parametric Designing for Thread FastenersParameterized hex-bolt generation predates this report.
- Parametric modelling of bolted assembly and its components for industrial design and applicationDesign-table components and constrained bolted assemblies are established.
- cq_warehouse: parametric fastenersSoftware precedent for parameterized fasteners.
- A survey on 3D CAD model quality assurance and testing toolsGeometric quality alone does not establish broader modeling or semantic quality.
- Recommended Practices for Geometric and Assembly Validation Properties v4.6Geometric properties, surface sampling and assembly structure/placement checks predate this report.
- Representation and system-agnostic automated CAD interoperability testing based on model interchangeabilityPublisher describes representation-independent local/global comparison with user tolerances.
- An Automated Approach for the Discovery of InteroperabilityProperty-conditioned interchangeability and round-robin CAD testing are established.
- Robust tessellation of CAD models without self-intersectionsCritical regions and constrained triangulation address intersection-free mesh generation.
- Topology-First B-Rep MeshingTopology-preserving generation is a different objective; first-part evaluation is explicitly disclosed.
- Importing and Exporting FilesImplementation context for exchange formats.
- glTF 2.0 SpecificationMetre units and Y-up convention.
- trimesh.proximityAll-triangle point proximity is an independent mesh-distance implementation check.
- minimize_scalarHistorical supporting scalar minimization reference.
- Gmsh Reference Manual 4.15.2Implementation/settings reference and requested software-paper attribution.
- Gmsh: A 3-D finite element mesh generator with built-in pre- and post-processing facilitiesSoftware scholarly attribution.