I would like to see an option which would bring together in a sequence characters with same radicals and/or components.
I beleive this would increase learning efficiency by exploiting structural simmilarity between characters and also help us diferentiate between them. I would suggest that any character currently being reviewed is followed by simmilar characters that are scheduled to appear relatively soon.
This is based on fact that characters are not not unrelated, but have strong links between themselves. Classical SRS model assumes that the entries are unreleated to each other, which does not hold here.