Abstract
Social networks are built through local decisions, yet the value of a connection depends on paths that individuals cannot fully observe. How such bounded information shapes collective efficiency remains unclear. Here we develop a computational model in which agents revise links using a limited view of the network, an expectation about connections beyond that view, and stochastic choice. Starting from an inefficient empty network, the model reaches near-efficient structures within an intermediate regime: agents must value unseen opportunities enough to initiate connectivity, but not so strongly that continued linking prevents the removal of redundant ties. This favorable regime narrows as groups grow, although broader visibility partly restores it and agent heterogeneity preserves its qualitative form. Earlier action or link receipt is associated with eventual hub status, while stable agent-level decision parameters are associated with persistent payoff ranks across repeated networks. Introducing one agent that forms links without considering payoffs reduces welfare in small groups but increases it in larger groups, while concentrating costs on the builder. These results identify a scale-dependent balance between exploration, local representation and value-sensitive pruning through which bounded individual decisions can generate collective efficiency and structural inequality.