We analyze the Susceptible-Infected-Recovered-Susceptible (SIRS) process, a continuous-time Markov chain frequently employed in epidemiology to model the spread of infections on networks. In this framework, infections spread as infected vertices recover at rate 1, infect susceptible neighbors independently at rate $λ$, and recovered vertices become susceptible again at rate $α$. This model presents a significantly greater analytical challenge compared to the SIS model, which has consequently inspired a much more extensive and rich body of mathematical literature for the latter. Understanding the survival time, the duration before the infection dies out completely, is a fundamental question in this context. On general graphs, survival time heavily depends on the infection's persistence around high-degree vertices (known as hubs or stars), as long persistence enables transmission between hubs and prolongs the process. In contrast, short persistence leads to rapid extinction, making the dynamics on star graphs, which serve as key representatives of hubs, particularly important to study. In the 2016 paper by Ferreira, Sander, and Pastor-Satorras, it was conjectured, based on intuitive arguments, that the survival time for SIRS on a star graph with $n$ leaves is bounded above by $(λ^2 n)^α$ for large $n$. Later, in one of a few mathematically rigorous results for SIRS, Friedrich, G{ö}bel, Klodt, Krejca, and Pappik provided an upper bound of $n^α\log n$, with contains an additional $\log n$ and no dependence on $λ$. We resolve this conjecture by proving that the survival time is indeed of order $(λ^2 n)^α$, with matching upper and lower bounds. Additionally, we show that this holds even in the case where only the root undergoes immunization, while the leaves revert to susceptibility immediately after recovery.