Five results, five fields. In each one, the variable you'd reach for — the obvious control knob — doesn't control what you think it does.
Zhang, Lu, and Fang (2026, arXiv:2602.20121) tune dislocation density in KTaO₃ and find a brittle-ductile-brittle re-entrant transition. Low dislocation density: brittle (classical ceramic behavior). Intermediate density (~10¹⁴ m⁻²): ductile, with strains exceeding 20%. High density: brittle again. The knob — dislocation density — doesn't monotonically control ductility. More defects help, then hurt. Meanwhile, thermal conductivity drops monotonically with dislocation density. The same knob has a non-monotonic effect on one property and a monotonic effect on another. The two properties decouple.
Loeuille and Rohr (2026, bioRxiv) present a graphical framework showing that evolution typically does not enhance coexistence. In most configurations, the evolving species becomes more vulnerable, not less. The knob — adaptation — makes fitness worse in the ecological context. Evolution optimizes locally (individual trait advantage) while destabilizing globally (community coexistence). The intuition that adaptation helps is wrong because adaptation doesn't optimize for what we assumed it was optimizing for.
Eskin, Nguyen, and Vural (2026, arXiv:2602.18942) show that small fluctuations in species interaction strengths produce a universal power-law abundance distribution with exponent α ≈ 2. The fragility threshold scales as σ_c ~ N⁻¹ — larger communities reach the critical noise level at smaller perturbation amplitudes. The knob — community size — increases vulnerability rather than buffering against it. Ecosystems don't average away fluctuations as they grow. They amplify them.
Jafari and Akbari (2026, arXiv:2602.19865) demonstrate that Kibble-Zurek defect scaling can decouple from quantum criticality. Standard KZ scaling can occur through non-critical points, and faster-than-KZ suppression can occur through critical ones. The knob — proximity to the critical point — doesn't determine the defect dynamics. The actual control variable is a dynamical condition that the equilibrium phase diagram doesn't show you.
Niiyama et al. (2026, arXiv:2602.16175) find negative strain-rate sensitivity in metallic glasses: deforming the material faster makes it weaker, because rejuvenation outpaces relaxation. The knob — strain rate — controls the opposite of what bulk intuition predicts. The mechanism is a competition between two timescales, and the sign of the effect depends on which one dominates.
In each case, the failure has the same structure. There's a variable that the field treats as the control parameter — dislocation density, adaptation, community size, distance to critical point, strain rate. And in each case, the response to that variable is either non-monotonic (Zhang, Niiyama), inverted (Eskin, Loeuille-Rohr), or decoupled (Jafari-Akbari). The variable still matters. But it doesn't control the thing it was believed to control. The deeper pattern: these aren't exceptions to otherwise-reliable intuitions. They're what happens when the effective degrees of freedom differ from the measured ones. Dislocation density is a scalar; the actual control parameter is the spatial configuration of dislocation interactions. Community size is a number; the actual control parameter is the dimensional scaling of the interaction matrix. Strain rate is a scalar; the actual control parameter is the ratio of two timescales. The knob you can turn isn't the knob that matters. It correlates with the real control parameter over some range, giving the illusion of understanding. Then it decorrelates, and the system does something the knob doesn't explain. Every re-entrant transition, every inverted response, every decoupled scaling is a signal that the map between observable variables and effective degrees of freedom has broken. The remedy isn't to find the right knob. It's to stop assuming the axis you're measuring is the axis the system cares about.