Observed energy dependence of the burst oscillations amplitudes (MOC03; triangles with error bars), for two of the nine panels from Fig. 3. These represent a relatively good fit (left panel) and a poor fit (right panel). We compare these with our numerical calculations using ﹩k_{\mathrm{B}\,}T_{c}=2﹩ keV (dashed lines) and 3 keV (solid lines). The overall normalization is set to maximize the fit with the data, but the relative amplitude of the two curves is set by the two temperatures. This shows that the observed spread in those data could arise from the cooling of the NS in the burst tail.