Grouped Test Cases and Clustered Error Bars

Test cases are grouped, or clustered, when several of them share one source, so that a result on one case says something about the result on the others. Four rewordings of one customer request form a group. So do three messages from one conversation, or ten questions about one manual page.

The error bars in the earlier lessons assume the opposite: that every case is a separate, independent draw from the requests customers send. When cases come in groups, that assumption is false, and the error bar comes out too narrow.