Shared, dependency-light primitives for landscape-genetic network methods: a lightweight deme/landscape graph (vertex coordinates and an undirected edge list) with constructors from coordinates; genetic covariance and distance from biallelic or multiallelic data (the Yang-style normalized-dosage covariance and the Dyer-style multivariate covariance); and antisymmetric per-edge directional covariate builders (the gradient of a scalar potential, and the projection of a vector flow field). Used by 'terradish' (symmetric resistance). No compiled code.