If task shells are successful at identifying and holding constant
the elements in a task that affect its difficulty and discrimination, it
may be possible to reduce pretesting requirements because
pretesting a sample of tasks generated by the shell would provide
data that could be applied to all of the tasks generated by the shell.
It is an empirical question whether or not the pretested sample of
tasks generated by the shell will be similar enough in their operating
characteristics to reduce the need to pretest all of the tasks generated
by the shell. If the variable elements can be sufficiently specified,
task shells can facilitate the automated generation of tasks.