The paper is about a property of graphs called "Schurian," which relates two different ways of detecting the symmetry of a graph. One way comes from algebra: you can run an algorithm called Weisfeiler-Leman (WL) coloring that groups pairs of vertices together based on shared structural patterns. The other way comes from the actual symmetries of the graph, meaning the ways you can permute vertices while leaving the graph unchanged. A graph is Schurian when these two approaches agree perfectly. Researchers have conjectured that every polyhedral graph, meaning every graph that could be the skeleton of a convex 3D shape (technically: planar and 3-connected), should be Schurian. This would also mean the WL algorithm only needs two rounds to fully distinguish such graphs.
The authors computationally checked this conjecture for every polyhedral graph with up to 16 vertices, which amounts to over 413 billion graphs. Their pipeline works in stages: first a fast filter eliminates most graphs quickly, then for the remaining cases a more expensive comparison directly checks whether the WL partition matches the partition coming from the graph's symmetry group. The computation was carefully validated using independent cross-checks, control graphs with known answers, and published counts to verify that no graphs were missed or double-counted. For sizes up to 13 vertices, the result was already indirectly known through a different mathematical route, but from 14 vertices onward, no prior method covered it.
The result is a strong piece of computational evidence supporting the conjecture, but it is not a mathematical proof. An interesting contrast emerges at 16 vertices: if you remove the planarity requirement, non-Schurian graphs do exist at that size (the Shrikhande graph is a well-known example), so the polyhedral restriction is doing real work. A companion paper looks at graphs on the torus, where non-Schurian examples actually do appear at 16 vertices, making clear that the planar polyhedral setting is genuinely special.