A mechanism has a typical effect. Tilting octahedra in perovskites stiffens phonons. Evolution improves fitness. Dissipation degrades information. Increasing community size buffers against fluctuations. These are the textbook directions. Six recent results show the same mechanisms producing the opposite effect when the system changes.
In SrTiO₃, octahedral tilting enhances thermal conductivity by suppressing certain phonon scattering channels. The same structural distortion in SrSnO₃ (2026, arXiv:2602.20142) produces “a completely opposite effect” — acoustic softening that suppresses thermal transport by 19% per degree of tilt. Same mechanism, same crystal family, opposite sign. The difference is the mass ratio of the ions: strontium-titanium vs. strontium-tin. The geometry of the distortion is identical. What it does to heat flow is not.
NiTi shape-memory alloys undergo martensitic transformation — the mechanism that gives them their characteristic dimensional instability. Heated, they change shape. But by training the crystallographic texture through combined mechanical and thermal cycling (2026, arXiv:2602.18315), the same martensitic mechanism is repurposed for near-zero thermal expansion — smaller than Invar, the alloy specifically engineered for this property. The mechanism that normally makes the material dimensionally unreliable makes it dimensionally stable.
In MoTe₂ (2026, arXiv:2602.19630), coherent phonons couple to electronic bands — but the coupling strength varies by band. The same lattice vibration shifts one electronic state by several meV and barely touches another. The phonon is one event. The electronic response is many, and they don't even have the same sign. The mechanism is phonon-electron coupling. Whether it matters depends on which electron you're asking about.
Evolution normally improves individual fitness. Loeuille and Rohr (2026, bioRxiv) show that in most multi-species configurations, the evolving species becomes more vulnerable to extinction, not less. The mechanism (adaptation) is intact — the individual is optimizing locally. But the ecological context transforms the sign. What helps the individual hurts the community member.
Increasing community size normally buffers against perturbation through statistical averaging. Eskin, Nguyen, and Vural (2026, arXiv:2602.18942) show that the critical noise threshold scales as σ_c ∼ N⁻¹ — larger communities are more fragile, not less. The averaging mechanism is real. But the interaction matrix grows faster than the averaging can compensate, and the equilibrium wanders into infeasible regions. The buffer becomes a liability.
Dissipation degrades quantum information. Except in the dissipative toric code (2026, arXiv:2602.19288), where continuous energy loss to the environment drives the system toward the code space — self-correcting errors in a regime where classical error correction has no finite threshold. The mechanism (coupling to environment) is the same mechanism that normally decoherences. Here it corrects.
The textbook writes down a mechanism and assigns it a sign: positive, stabilizing, enhancing, degrading. These six results show that the sign isn't a property of the mechanism. It's a property of the mechanism-in-context. Octahedral tilting has no inherent thermal sign. Evolution has no inherent fitness sign. Dissipation has no inherent information sign. The mechanism contributes; the system determines the direction. This is stronger than saying "context matters." Every physicist knows context matters. The claim is that the sign — not the magnitude, not the rate, but the direction — flips between systems that share the mechanism. You cannot know, from knowing the mechanism alone, whether it helps or hurts. The perovskite lattice doesn't know whether its octahedra are helping or hindering heat flow. The ecosystem doesn't know whether its evolution is stabilizing or destabilizing. The quantum memory doesn't know whether its dissipation is error or correction. The sign depends on the system, and nothing about the mechanism will tell you which system you're in.