We present various works aimed at maximizing polarization in the Electron Storage Ring (ESR) of the soon-to-be-built Electron-Ion Collider (EIC), and describe in detail the polarization properties of the ESR throughout its design evolution. Most significantly, we present a novel method called "Best Adjustment Groups for ELectron Spin" (BAGELS) that achieves simultaneous control of the polarization, orbit, and optics in electron storage rings by use of special vertical orbit bumps constructed via dimensionality reduction. Using BAGELS, we nearly double the asymptotic polarization in a 1-interaction point (IP) ESR lattice, and more than triple it in a 2-IP lattice. We also use BAGELS to construct knobs that can be used for global coupling correction, and knobs that generate vertical emittance for beam size matching, all while having minimal impacts on the polarization and orbit/optics. Furthermore, we present SciBmad, a new, modular, differentiable, and high performance accelerator physics software that can be used easily in Python or Julia. SciBmad's symplectic integrators, which include spin, are universally polymorphic, forwards-/backwards-/Taylor-differentiable, and CPU/GPU parallelizable. Also included are a high-order automatic differentiation library, and routines for doing perturbation theory with nonlinear (possibly damped) Hamiltonian maps using Lie algebraic methods. SciBmad's machine learning-enabled ecosystem aims to be a powerful tool for modern particle accelerator design and simulation.