Page 181, Fig. 3-4 (given here in postscript)
is misleading. The caption for (a) says that this is the character
sequence sent by the network layer. If this was the case, a DLE should
be stuffed before the DLE's at the beginning and end because they could
not be framing characters. Framing characters would be appended by
the data link layer, not the network layer.
Note that problem 2 on page 239 uses the exact same sequence as in Fig. 3-4.
As an aside,
why is a DLE stuffed before the existing DLE and not after? :-)