This is a good idea.

Will incorporate false-positive rates into the rubric from the next run onwards.

At winfunc, we spent a lot of research time taming these models to eradicate false-positive rates (it's high!) so this does feel important enough to be documented. Thanks!