It is an interesting idea.
I aready record Disproven items… so that it isn’t necessary to re-run the disproving exercise the next time the same bad info pops up. (As an example, I often see the Christening information that was mistaken for Birth data. Such a widely held misimpression can often be traced back to single Source being transcribed incorrectly.)
So marking other data as low-probability ‘theory’ would also be good if Gramps is to be a research tool rather than simple filing system. But it also it raises the spectre of data transfer problems. If the receiving system CANNOT annotate something as theoretical or Disproven, all of the transferred data becomes ‘fact’ instead ‘fiction’.