As far as I understand all test run on a given text file. And compare this
to a expected output. Then they run again but this time using the result
from the first run.
I guess the purpose is to soo that it does not get changed again
because it is already the expected output.
After running all tests there are 3 or so in the repo that fail in the
second run. How important is it to fix these? Because they were
there for some time.