I've been working on an open-source simulation lab called CAELIX and wanted to share it here. It's a discrete lattice field framework for testing whether continuum-like signatures can emerge and be routed using strictly local update rules on a balanced-ternary substrate {-1, 0, +1}.

There are 11 real-time animations running directly off the simulation code on the website. (I locked these specific visualisations to 2D so they actually run smoothly in the browser on phones and modest hardware). The architecture enforces a strict separation. The discrete balanced-ternary microstates act as a topological boundary generator which couples via a deterministic load functional into continuous carrier dynamics. The numerics are often FDTD-like, using diffuse and telegraph surrogate equations, but the framework as a whole is much broader than a conventional solver.

A bit of background: I'm not a physicist. My expertise is in computing systems, databases and audio engineering. I go all the way back to writing Z80 machine code and making ZX Spectrum games as a kid. CAELIX actually started out as a machine learning experiment in adaptive noise cancellation before evolving into this.

I'm not making any grand metaphysical claims either. The 1/r vacuum profile it generates isn't claimed as novel. It's just the expected asymptotic behaviour of the 3D discrete Laplacian and I use it as a regression-safe baseline.

The part I think this crowd might find interesting is the topological fan-out. The field dynamics can route stable non-linear solitons (specifically generative Sine-Gordon kink walls) through geometric logic gates like a T-junction. The computation lives purely in the passive waveguide topology rather than active switching states. There is also a module hunting for stable oscillons but the gates themselves use the kink walls. I've mapped the failure modes extensively to separate genuine kinematic-dilation proxies from stencil dispersion and discrete mode-locking artefacts.

Repo: https://github.com/caelixuk/caelix (I've included pre-extracted .h5 sprite assets so you can run the collision and routing experiments immediately without needing to run the extraction sweeps first).

White paper: https://doi.org/10.5281/zenodo.18823977

I'd be really interested to hear if anyone working on unconventional compute, cellular automata or lattice field solvers has seen similar passive routing behaviour in other discrete substrates. Happy to answer any questions on the numerics or the routing grids.