This preprint (arXiv:2605.28928, v1, May 2026) presents a theoretical time-dependent model coupling viscous disk spreading with repeated EMRI-star impacts to explain slow, massive outflows in TDEs that produce radio emission years later. The mechanism differs from jet-delay or viewing-angle scenarios by tying the delay directly to the viscous timescale for the compact TDE disk to reach the EMRI orbit, ejecting material at ~0.02-0.1c with masses up to ~1 solar mass and energies ~10^51 erg. As a preprint it remains unpeer-reviewed, relies on parameterized viscosity and collision efficiency without direct observational fitting, and lacks sample statistics since it is purely analytic/numerical modeling rather than a population study. Related work on QPEs (Miniutti et al. 2019, Nature) shows star-disk collisions can power X-ray eruptions in the same systems, while late-time radio surveys (e.g., Anderson et al. 2020, ApJ) document rebrightenings whose energetics now align with this outflow channel. The model predicts a subset of radio-loud TDEs should overlap with QPE hosts, offering a multi-messenger test once LSST and ngVLA transient streams mature, yet it underplays potential contamination from circumnuclear medium density variations that could mimic the signal.