This seems to be over-desing-patternized. Methods needed by standard
eval plugins (ie part of standard distribution) are part of strategy
API. Contributed eval methods will have to patch strategy plugins if
they need some other data from them.
The only weak point here is that we store a raw grade into
workshop_grades and not a direct id of the selected level. Therefore,
when re-assessing, we need to actually guess what level the assessor
chose previously. This is not problem if there are not two levels with
the same grade. Such case is not common when using Rubric. In the
future, this may get refactored so Rubric would use its own storage of
filled assessment forms.
David Mudrak