Research

Work that isn’t a product yet.

Open problems we’re chipping at, written up in plain terms — including the parts that didn’t work.

Where, and what.
Under peer review
Machine learning & physicsUpdated August 2026
A lattice physics calculation handed to a neural network as a graph with two kinds of node — one for a point of the grid, one for the value the field carries there — a split borrowed from how the continuum theory is written down. Its sharpest finding is a measured failure in the standard single-node approach, not a win for this one.
Read →