Extended-source stress test at T = 300 K. We perturb the per-object ﹩{L}_{{\rm{wh}}}^{{\rm{\max }}}﹩ for “extended-like” sources (large w1rchi2, w2rchi2, w3rchi2, or w4rchi2, or ext_flg>0) by +10% on the limiting band and recompute f ₉₅. The curve shifts by ≤3% at midthresholds and leaves the high-threshold plateau unchanged, confirming that any residual extended-source bias is subdominant to the RSR bandpass effect and mask choices.