A group of people (usually supervisors—either the same people who generated the critical incidents initially or another group) clusters the incidents into a smaller set of performance dimensions (usually five to ten) that they typically represent.
The result is a given number of performance dimensions, each containing several illustrative critical incidents.