Written answers are scored against an answer you write yourself, and every attempt is now kept.
Marking used to compare a written answer with whatever the model thought the answer was. It now compares it with the answer you wrote while approving the set, so the standard is yours.
Retakes are kept separately instead of overwriting each other, so you can see whether the second attempt actually moved.
Smaller things: a private page shows whether its learner has opened it, and a set can be unpublished without being deleted.
