We need a way to measure which graph optimizations were used, how often they failed, etc. This allows us to measure the optimization coverage of the test suite.

This is related to* 213