Can tidal disruption event models reliably measure black hole masses?
Tidal disruption event (TDE) light curves are increasingly used to infer the masses of quiescent supermassive black holes ($M_{\rm{BH}}$), offering a powerful probe of low-mass black hole demographics independent of host-galaxy scaling relations. However, the reliability of most semi-analytic TDE models assume full stellar disruption, despite theoretical expectations that partial disruptions dominate the TDE population. In this work we test the robustness of current TDE models using three repeating partial TDEs (rpTDEs), in which the multiple flares produced by the same surviving stellar core must yield consistent black hole masses. We present spectroscopic observations establishing AT 2023adr as a rpTDE, making it the third such spectroscopically confirmed event. We independently model the flares of the three rpTDEs; 2020vdq, 2022dbl, and 2023adr, applying fallback-accretion fits, stream-stream collision scaling relations, luminosity-based empirical relations, and cooling-envelope fits. After accounting for statistical and model-specific systematics, we find that all TDE models generally return self-consistent $M_{\rm{BH}}$ values between flares, and are broadly consistent with host-galaxy $M_{\rm{BH}}$ proxies, recovering $M_{\rm{BH}}$ to within 0.3-0.5 dex. However, the convergence of fallback models towards unphysical stellar masses and impact parameters reveals limitations in the existing fallback model grids. We also show that light curve coverage, particularly in the near-UV, is critical for constraining model parameters. This has direct implications for interpreting the thousands of TDE light curves expected from upcoming surveys such as the Rubin Observatory's Legacy Survey of Space and Time, where from simulations, we find that $M_{\rm{BH}}$ may be underestimated on average by 0.5 dex without additional follow-up.