A music teacher is in charge of evaluating student performances at a
musical competition.  Numerous judges are available to judge the 
students, but each judge can only evaluate two or three performances.
The response variable is based on a performance-rating scale.

In the original analysis, 50 students were evaluated by 50 judges.
Judges were split into teams of two or three, and each member of the
team evaluated the two or three student performances assigned to that
team. Results were compared across judges to determine a winner. The
instructor was dissatisfied with this approach and would like to
analyze additional competition data using a better model.

Identify the original design.  Identify any weaknesses in the design. 
Faced with the same sort of constraints, how might you design this
experiment?

Students recognize that having groups of only two or three judges 
reviewing two or three performances creates problems with judge
effects.

I had answered this by considering a BIBD with k=r=3 and a=b=7; there
is still a problem with judge effects, but it should not be as bad.
Student solutions include using a smaller panel of judges to review
a larger number of student performances (presumably you could find
a smaller group of judges with a higher level of commitment).  Other
students tried to find a BIBD (with b and a approximately 50) for the
entire experiment.  Such solutions are possible but k tends to be a
little too high.  One solution was to use a slightly large number of
judges and winsorize the scores--quite a good idea.