Natural selection favors the fittest. This is the central claim of evolutionary biology, and in the low-mutation regime it is straightforwardly true — the genotype with the highest fitness dominates the population, and selection maintains the population near the peak of the fitness landscape.
Palpal-latoc and Vega (arXiv 2602.14863) formalize what happens when mutation rates rise above a critical threshold. The population delocalizes from the fitness peak. The fittest genotype is lost — not because it stopped being fit, but because mutations push offspring away from the peak faster than selection can pull them back. The peak is still the highest point on the landscape. The population just doesn't live there anymore.
What survives instead is not the optimal genotype but the most robust one — the genotype sitting on a flat plateau rather than a sharp peak. A sharp peak has high fitness but low mutational tolerance: one mutation in any direction produces a large fitness drop. A flat plateau has lower fitness but high tolerance: mutations move the genotype to nearby positions of similar fitness. When mutations are common, the plateau wins. This is survival of the flattest.
The authors derive a dimensionless quantity — the localization factor — defined as the ratio of effective fitness variance to the square of the mean mutation rate. When this ratio is large (fitness differences dominate mutations), the population localizes at the peak. When it's small (mutations dominate fitness differences), the population spreads across the landscape. The transition between these regimes is sharp: a critical mutation rate above which the peak population collapses.
The inversion is complete. In the high-mutation regime, being the best is a liability. The sharp peak that makes a genotype optimally adapted also makes it optimally fragile — every neighbor is worse, so every mutation is costly. The flat plateau sacrifices optimality for durability. The organism that is merely good at everything outcompetes the organism that is perfect at one thing, because the merely good organism can absorb perturbation without catastrophe.
The general principle: optimization and robustness are different objectives, and they trade off. A system tuned to the optimum is maximally sensitive to displacement. A system tuned to tolerate displacement is suboptimal but persistent. Whether the peak or the plateau dominates depends on the noise level — and the noise level is not a property of the system but of the environment. Change the mutation rate and you change which strategy wins without changing any fitness value. The landscape doesn't move. The population's relationship to it does.