Constraining local primordial non-Gaussianity through scale-dependent bias requires an accurate determination of the PNG bias parameter $b_φ$, which is completely degenerate with $f_\mathrm{NL}$ in the galaxy power spectrum. The standard universality relation predicts $b_φ$ from the linear bias, but breaks down for realistic galaxy populations due to PNG assembly bias. The time evolution estimator offers a promising alternative, inferring $b_φ$ from the redshift evolution of tracer number counts. We test it against the separate universe estimator using the IllustrisTNG and SIMBA suites of CAMELS. Our main contribution is to demonstrate that the estimator can be extended to realistic observer-frame photometric selections, provided correction factors derivable from the galaxy SEDs are applied to account for luminosity-distance dimming and $K$-corrections. We further show that physical evolution not directly tracing the growth of matter fluctuations can also be corrected under assumptions about the star formation history: magnitude cuts require correcting for passive stellar fading, after which agreement with the separate universe estimator is restored. This establishes the robustness of the time evolution estimator of PNG bias with realistic survey selections, paving the way to future applications with real data.