analysis

Data splitting criteria

•Static analysis ¨

•Distinguished values: zero, source literals, mode, outliers, extrema

•Exceptions to detected invariants

•User-selected

•Exhaustive over random sample


	[Don’t be defensive about how simple my static analysis is.]
	Say that this is a policy decision.

	I chose to use a trivial static analysis: boolean conditions and functions.
	No sense doing something complicated until you have demonstrated that simple approaches are insufficient; and on simple examples, this was sufficient. I will revisit this decision if necessary.