I am struggling a bit with different classifications of things detected.
1 - the chance of it actually being an error (sure, probably, maybe, possibly)
2 - the severeness of it (unacceptable, considered wrong by most, considered wrong by some, a matter of taste)
3 - by type of error (current categories more or less)
class 1 and 2 could help the user setting how strongly he wants to be advised... Was ever thought about introducing these kind of classes?