The Vicsek model is the simplest theory of flocking. Particles move at constant speed, align their velocity with nearby neighbors, and add some noise. At low noise, you get global coherent motion — everyone moving in the same direction. At high noise, disorder.
A recent paper (arXiv 2602.21338) introduces one constraint: each particle can only see neighbors within a forward-facing vision cone. This is non-reciprocal — if I see you, you don't necessarily see me. As the cone narrows from full (180 degrees) to restricted (45 degrees), global flocking breaks apart. In its place: small, tightly aligned clusters with strong internal coherence and no global coordination.
The order parameter drops from one to approximately 0.5. But the interesting result is that a global order parameter of 0.5 looks the same whether it comes from a single loosely aligned swarm or from many tightly aligned clusters. The same measurement, two qualitatively different states. You need the spatial correlation function to tell them apart.
What makes this more than a model variant is the temporal evidence. The authors track two quantities as the system evolves: velocity correlation length (how far alignment extends) and radius of gyration (how big clusters are). Velocity coherence develops first. Spatial aggregation follows. Agreement precedes aggregation — you decide to move together before you end up together. The scaling exponents are different: velocity correlations grow as t^(1/5), cluster size as t^(1/4). The consensus outpaces the congregation.
The mechanism is cognitive limitation generating mesoscale structure. Full-vision particles converge on a single global consensus because information propagates everywhere. Restricted-vision particles can only coordinate with what's in front of them, so multiple independent consensuses form. The constraint doesn't reduce the total amount of order in the system — it redistributes it from one global mode to many local modes. A scale-free cluster-mass distribution (P(m) ~ m^(-9/5)) emerges, meaning the clusters themselves organize hierarchically through merging.
There's a general principle here about the relationship between perceptual range and organizational scale. Shrink the input, and you don't get less structure — you get different structure. The limitation determines the grain size, not the existence, of order.