The data before 5301 s were taken during the XRT anomaly. The PC data may be unreliable.
From 2016 January, the light curve plots support interactive zooming and exploring. See the jsPlot documentation for details.
Also, from the same date, the first plot on this page is the observed flux
in the native band of each instrument (including UVOT where available).
All data uses absorption NH=2.957 × 1022 cm-2 at z=0. (Source: the late-time spectrum from the XRT spectrum repository).
The XRT data in WT mode, and in PC mode prior to the switch to WT mode were taken during the anomaly. During this time, XRT was not collecting data below ~0.5 keV, which means that the XRT hardness ratio at this time is artificially increased. Once XRT was restored (at T+5300 s) to normal operations, data between 0.3—0.5 keV were again included. In the Burst Analyser, this causes an artificial strong hard-to-soft spectral evolution in the early XRT data. This evolution is not physical but is an artifact of the transition of XRT from an anomaly state to normal operations.