remarkable growth of the field increasingly requires precise, widely-applicable, and modular methods that can model the quantum electrodynamics of the physical circuits, and even of their more-subtle renormalization effects. Here, we present a computationally-efficient method satisfying these criteria. The method partitions a quantum device into compact lumped or quasi-distributed cells. Each is first simulated individually. The composite system is then reduced and mapped to a set of simple subsystem building blocks and their pairwise interactions. The method operates within the quasi-lumped approximation and, with no further approximation, systematically accounts for constraints, couplings, parameter renormalizations, and non-perturbative loading effects. We experimentally validate the method on large-scale, state-of-the-art superconducting quantum processors. We find that the full method improves the experimental agreement by a factor of two over taking standard coupling approximations when tested on the most sensitive and dressed Hamiltonian parameters of the measured devices.