Excel DM Add-ins: How Are Highlight Exceptions Rows Chosen?

When I first looked at the query behind highlight exceptions, I assumed that PredictCaseLikelihood() was the basis for deciding which rows to flag, but a closer look at the sample data shows this not to be the case. For example, looking at the "Table Analysis Tools Sample" table in the DMAddins_SampleData workbook, ID 11147 is highlighted while ID 11378 is not. Training a cluster model via DMX on this same dataset shows these rows having PredictCaseLikelihood values of 0.165 and 0.158 respectively.

Does anyone have an idea what methodology is used to select the Outlier rows? And once those rows are selected, how are the suspect columns within that row chosen?
Appreciate any clues or directions.