Empirical variance calibration prior to adaptive smoothing, shown for the example target J0910b. Adapted from Z. Lin et al. (2025). The measured standard deviation in blank-sky regions is compared with the expectation assuming uncorrelated noise as a function of kernel size. The variance increase is well described by a logarithmic relation, ﹩{\sigma }_{{\rm{meas}}}={\sigma }_{{\rm{nocov}}}[1+\alpha \mathrm{log}({N}_{{\rm{kernel}}})]﹩, where N_kernel is the number of spatial pixels in the box smoothing kernel. The relation approaches an asymptotic value at large kernel sizes, reflecting the finite correlation length introduced during coaddition. This calibrated variance model is used during adaptive smoothing to obtain statistically consistent uncertainty estimates.