Some of the handled errors need to know the context in order to be detected or classified. We have a problem if an error occurs on the last note: mechanisms for error detection do not have enough visibility to determine what has been done.
That happens for exemple on wrong note problems: if a g-state is got for the last note, we don't know if this note was wrong or not played. We have also a problem is this last note is too short or too long, since nothing happens after it...
Sometime the problem may appear if an error occurs on the first note. Alignment is different in that case since first state to be reached is a g-state.