When every resolver declares only its input and output attributes, Pathom treats the union of resolvers as a graph and solves API queries as a classic graph-traversal / optimization problem (ideas from 1960s graph theory): batch when the target API supports it, parallelize independent branches, fall back to alternative paths when one resolver errors (e.g. a 403 on file-details triggers an alternative resolver returning only title and channel-id). No custom pipeline code needs to be written per query.