Symmetry groups classify. Given the symmetry, you know the universality class, you know the critical exponents, you know the scaling laws. This is the standard promise of the Landau paradigm: identify the symmetry that breaks and you've identified the physics.
These results break that promise, carefully.
Lee (2026, arXiv:2602.09090) studies a squeezed-photon model with Z₂ symmetry implemented in two ways: as a strong symmetry (commutes with every Lindbladian jump operator individually) and as a weak symmetry (commutes with the full Liouvillian but not with individual jump operators). The symmetry group is identical. The static fluctuations are identical — same Gaussian fixed point. But the dynamical critical exponents differ. The order parameter and the asymptotic decay rate scale differently under the two realizations. Same symmetry, different universality classes.
This is not a violation of the Landau classification. It's a refinement. In closed systems, there's only one way to have a symmetry — the Hamiltonian commutes with the generator. In open systems, there are at least two: strong (each dissipative channel preserves the symmetry individually) and weak (only the total dynamics preserves it). The symmetry group doesn't change. What changes is how it's distributed across the system's interaction with its environment. The critical behavior depends on that distribution.
Zaccone and Samwer (2026, arXiv:2602.16390) derive a universal ratio between the energy of a minimal dislocation loop and the thermal energy at the melting point: approximately 25.1 for face-centered cubic crystals, independent of elastic moduli and chemical composition. The same topological excitation — a dislocation loop — exists in every crystal, but the universal ratio at which it proliferates is a purely geometric constant determined by the loop's shape and the lattice structure. Different materials, same geometry, same melting criterion. The universality is hidden because the individual parameters (shear modulus, lattice spacing, Burgers vector) all vary; only their ratio at the transition is fixed.
Hernández-García and Velázquez-Castro (2026, arXiv:2602.16129) study gene regulatory networks with intrinsic fluctuations and find that in very small systems — few molecules — oscillation is suppressed. The same negative-feedback architecture that produces robust oscillation in large networks fails to oscillate in small ones. This is the inverse of Caupin et al. (2026, arXiv:2602.18785), where small systems oscillate because they can't sustain the steady state that large systems achieve.
Same mechanism — finite-size effects. Opposite outcome. Caupin: smallness creates oscillation (the free energy barrier is crossable). Hernández-García: smallness suppresses oscillation (the fluctuations wash out the limit cycle). The difference is what the size constrains. In confined fluids, size constrains the energy barrier between phases. In gene networks, size constrains the signal-to-noise ratio of the feedback loop. The finite-size effect is real in both, but what it does depends on what degree of freedom the size couples to.
Mehling and Dijkstra (2026, arXiv:2602.17235) find that the Atlantic Meridional Overturning Circulation response to Greenland meltwater is similar at high and low ocean model resolution. The community expected that explicitly resolving mesoscale eddies (1/10° versus ~1°) would change the answer — eddies transport heat and salt, and their inclusion should affect how freshwater disrupts the overturning. Instead, the background ocean state matters more than whether eddies are resolved. The same forcing, applied to the same equations at different resolutions, produces the same result. Resolution is the wrong variable.
Four contexts, one pattern. The thing you'd expect to determine the outcome — the symmetry group, the material chemistry, the system size, the model resolution — doesn't. What determines the outcome is how that thing is embedded in its context. Lee: same symmetry, different realization → different critical behavior. Zaccone-Samwer: different materials, same geometric embedding → same melting point. Hernández-García vs. Caupin: same finite-size effect, different coupling → opposite sign. Mehling-Dijkstra: same equations at different resolution → same answer (background state is the real variable). The standard move in physics is to classify systems by their components: the symmetry, the size, the resolution, the material. These results show that classification by components is insufficient. The same component — literally the same mathematical object — produces different physics depending on how it's realized, embedded, or coupled. The symmetry group is necessary but not sufficient. You also need to know how it sits inside the dynamics. This is not relativism. The outcomes are precise and measurable. Lee's dynamical exponents are specific numbers. Zaccone's ratio is 25.1, not 24 or 26. The claim is structural: the mapping from symmetry to behavior has more degrees of freedom than the symmetry alone specifies. The extra degrees of freedom live in the embedding.