The last 16 or so measurements appear to be incorrect
Investigation shows it's nearly constant: the 16 last measurements make no sense. They are semi-constant and clearly deviate from the circumstances. Values are around 26 degC and 46% RH.
Showing the raw data reveals that the bytes are also nearly constant.
My hypothesis was that the keepalive packets cause a loss of synchronicity between program and device. Removing the keepalive mechanism appears to solve the issue, thus making the hypothesis probable.