SuperContinuity – Open-Source G4 Surface Continuity (Alpha 0.1)

Hi everyone,

I’d like to introduce SuperContinuity G4 Analysis, a free, open-source Rhino 8 plug-in released under the MIT license.

The long-term goal is to develop a reliable tool for analyzing NURBS surface continuity up to G4. And I really mean long-term project. The AI is making automatically the project 24/7 and G3 level beta maybe ready 2027 spring. Furthermore the G4 level may take even years to be really finished. Let’s see how good mathematics head AI has.

This is still an early experimental alpha, but the first implementation of sampled geometric G4 analysis is working.

Some progress so far:

  • G0–G4 sampled continuity diagnostics with viewport colors, numerical results and CSV export.

  • Fourth-order geometric surface-derivative comparison, rather than simply comparing parametric C4 derivatives.

  • 180/180 tests passed in the Rhino 8 validation suite.

  • Four mathematically defined NURBS reference seams tested in Rhino 8, including a genuinely G4-continuous surface pair, a G3 pair and a G2 pair. All 546 sampled stations produced the expected classifications using the documented test tolerances.

Important: This is not a mathematical proof of G4 continuity over an entire seam. Results are based on finite sampling and tolerances. Incomplete knot coverage, trimmed surfaces and more complex NURBS configurations still require further work. The current alpha does not always clearly distinguish incomplete coverage from passing sampled results.

Nevertheless, I wanted to share the progress rather than wait until every problem has been solved.

GitHub – source code and Alpha 0.1 download:

The source code, Rhino 8 plug-in and mathematical validation documentation are included.

Feedback, critical evaluation and challenging NURBS test cases would be greatly appreciated!

Tommi P. Laiho

MSc of Econ, MA.
Turku, Finland

SuperContinuity G4 Analysis Alpha 0.1.1 is the next step in an openly AI-assisted development project. The implementation, fixes, tests, and mathematical documentation were developed with AI writing the code, guided by human requirements, modeling experience, and practical testing in Rhino.

The results are promising. The current version passed 180 Rhino 8 regression tests, and four analytic NURBS references produced the expected G4, G3, G2, and G4 classifications. This update also improves how uncertainty is reported: incomplete sampling coverage must not appear as a confirmed green G4 result, even when every measured point passes.

This remains an alpha. Sampled geometric G4 analysis is not mathematical certification of an entire seam, and the documentation explains that distinction. AI authorship is part of the project’s story; the code, reference geometry, measured results, and limitations are available for others to inspect and challenge.

The aim is to build a useful, understandable tool—and to see how far transparent collaboration between human judgment and AI development can take it.

Source code, reference models, and Alpha 0.1.1 release

SuperContinuity_G4_Mathematics.pdf (1.3 MB)

G4 is about matching the change of the change of curvature at a given point. Under which circumstances does this solve any problem? Just curious :winking_face_with_tongue:

For example in car ceilings sometimes there’s need for even G4 continuity. Also next spring 2027 G3 level beta release maybe :thinking: possible. That’s much more possible need. But this is also scientific experiment. I want to see what AI can really do. And so on. Just curious :face_with_monocle:, you know.

Well in my experience its about getting the best continuity under the situation possible. This means in practice, you aim for G2 and then try to get close to G3. I’m just addressing the G4 pointlessness, because its a type of exaggeration made by some people in the industry. What’s funny is, that A.) if you increase the amount control-points and B.) if you move them closer to the boundary, you can cheat to a better continuity. While your surface model is getting worse overall. So actually a higher continuity is easier to achieve, the worse your surface model gets…

If you detect this form of cheating, you could develop a score of how good the actual transition is.

Anyway, nice project.