The crash_pipeline.py file has a lot of tricky issues about needing the ``__init__`` and ``run`` methods to all take the same arguments. stgao suggests that this may be because the pipeline class doesn't expect/allow subclasses to implement their own ``__init__``. The only reason we define ``__init__`` is because we need to cache some things for ``finalized`` to use (both for logging, and for setting the status in the ndb.Model).

We should see if there's some way to clean all this up, simplifying the code to avoid triggering so many design constraints.