Because you are using 1<>file to redirect standard output, the shell
does not truncate the file before executing awk.  When awk writes to
the file it is overwriting existing data.  Your transformation
shortens the data, and therefore there is some of the old data
left at the end when awk has finished writing the new data.

The 1<>file trick is neat, but it should only really be used when
you are doing a one-to-one transformation.  It's also only safe if
the transformation is idempotent (i.e. repeating the change does
not alter the result).  This is so that if awk is killed part way
through, you can just restart it from the beginning.

An example of a safe usage is:

$ cat for_test
abcabc
$ awk '{ gsub(/b/, "B"); print }' for_test 1<>for_test
$ cat for_test
aBcaBc

--
Geoff Clare <netn...@gclare.org.uk>